/
NEWS.txt
2195 lines (1293 loc) · 133 KB
/
NEWS.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# GPU Preview 0.5.0 for developers
GPU is currently undergoing its 4th rewrite, codenamed freedom (3th was named solaris which reached version 0.963, 2th polaris, 1th unnamed). The preview package has version 0.5.0. This package is solely intended for developers as there is still no client working, just a bunch of unit tests.
The package is intended to sketch the whole idea of the next generation GPU. Some people might be disappointed that GPU moves away from peer-to-peer to become a client-server architecture, but we believe this is the only drawback. The highlights of the coming release (probably 6 months from now) are:
- client/server architecture should allow scaling of nodes up to several thousands of clients
- client is in Freepascal and can be compiled on Windows, Linux and Mac OSX
- Server is in PHP/mySQL and runs everywhere where LAMP stack runs
- All software used to develop GPU II is Open Source and available freely
- The plugin architecture of GPU I is retained, although the plugin layer has been rewritten (available in preview) and plugins need to be ported as well
- Communication is entirely over HTTP, workunits can be downloaded, and resulting workunits can be uploaded via HTTP back to the server (available on GPU preview package)
- HTTP makes GPU transparent to most proxies, so that clients can run also in companies and not only at home
- GPU II retains features like channels (to run whiteboard, chat and sensor networks), job management, clustering services, all rewritten in light of the experiences
done with GPU I (available on GPU preview package)
- GPU II will stay a multipurpose distributed computing architecture, so that any client and any server will support any computation (unlike the BOINC architecture)
- Each server should be able to support 40-100 clients according to some initial computations, the network should scale with number of servers
- One superserver will manage the whole network of servers
The package contains most of the server, communication layer client and server side, plugin layer clientside.
we hope the release will fuel some interest into the GPU project.
Have fun!
2011-02-14
#
0.960: climate simulation simulates vulcano Eyjafjallajoekull
GPU 0.960 features a simple climate simulation. With the latest update (0.9607) which can be retrieved with the menu "Tools->Autoupdate to latest version" on the main GPU application, the climate simulation
also shows the path of ashes spread by the vulcano Eyjafjallajökull and the CO2 production of the world population.
However, it is difficult to calibrate the parameters of the model. Anyone can try simple tasks like: having a world with stable temperature at 15.5 Celsius, start a nuclear war, move Earth down to Mercury orbit or up to Iupiter, stop earth rotation or revolution, etc.
A screenshot of the climate simulation can be seen at https://sourceforge.net/project/screenshots.php?group_id=58134&ssid=132912
Source code of this frontend can be found under /lib/libclimate and /src/frontend/thedayaftertomorrow.
2010-07-20 10:42:05 UTC by dangermaus
Super C.R.A.P.
This is a preview of what is being built right now. Super C.R.A.P is an acronym for Super Cluster Ready At Processing, core of the system is a Rollin KVM switch with 8 outputs (VGA, mouse, keyboard) bought for 218 CHF (=140 €). Additionally, 8 cables with VGA, mouse and keyboard are needed (each one costs about 15 €). An old 10 MBit Ethernet Switch will provide connectivity to the nodes of the cluster.
The cluster which now features 4 computers (2 XP, 1 Gentoo and 1 Sun) will be extended to 8 computers, all computers are coming from people who substituted their one and therefore are a little bit old, but very cheap (got them for free). The Rollin KVM Switch does not support the Sun Ultrasparc, so it will be put up separately.
The cluster will be both a showcase and a sounding board for GPU. Particular care will be needed when operating the 8 computers simultaneously, the necessary power will be taken from different phases. Estimated consumption is about 8x500W + 200 W (LCD screen) = 4200 W which is quite a big oven :-)
The cluster should be fully operational for beginning of 2010.
2009-10-08 06:24:48 UTC by dangermaus
#
Two breathtaking ideas / 0.952 released
The first one is from nanobit who fortunately rejoined the project: to use a script engine (e.g the one of Thexa4 released with 0.9256) as basis for distributed stored procedures. Stored procedures are a well-known thing for people working with databases, they are most of the time SQL procedures stored for convenience in what a database system can administer at best: a table with a column for the row number and another for the lines of the script. Using the distributed WAN database firstly introduced by nanobit in 0.943 joined with the stored procedure concept would allow another way of doing distributed computing with one major advance: it is possible for the user to distribute new code in almost real time.
The second one is to have sensors which broadcast collected data via GPU to the nodes. A special frontend collects data and filters it by sensor types (e.g earthquake sensors, temperature, insulation) and by streamers (the nodes which have a sensor attached and which broadcast its data). There is a prototype on CVS and we have two earthquake sensors, one in Holland and one in Switzerland. It will take a fairly amount of time and development effort until GPU people will be able to look at both signals and to compare them and average them. The averaging effect should in principle remove noise from the signal (like bus, or neighbour walking around).
0.9253 introduces an important bugfix in the milestoning mechanism of Eternity II plugin. Duncan White kindly reported it, and helped to test it. And we still are working on the duplicate packets issue which plagues the GPU network since 2004. Maybe this time we will find a solution, at least we identified once for all in source code where the problem is. Last but not least, this news is written with the 100$ laptop, a dream who is becoming true,see http://www.olpcnews.com
0.953 will be release at mid March. Keep an eye on the ChangeLog on CVS for the latest details.
0.952 released
--------------
0.952 is released on sourceforge.net. It contains important fixes to the connection layer done by Alex Fihman. This should be the most stable GPU ever.
Additionally, the command:
'add', whichplugin
will tell that basic.dll is the plugin that implements the function add. This will help users in deciding which plugins to unload, if some kind of jobs creates problems.
2008-03-05 23:06:53 UTC by dangermaus
#
Torrentes, Wiki and deltasql
During last month GPU experienced a short popularity blast with at least 20 nodes online all the time and with peaks of 25-26 nodes. Though this blast finished (now we have 10-15 nodes as usual in the last two years), the blast brought new improvements in the way the community can work together to achieve the goal of a flexible Internet Supercomputer.
Thanks to everclear64, GPU is availabe as torrent and can be downloaded with a Torrent client like ABC (available at http://pingpong-abc.sourceforge.net ). The GPU torrent itself can be downloaded at http://www.tek-zone.com/gpu/GPU.torrent .
The GPU project also got a wiki at http://www.gpu-grid.net/wiki as home to collaborative notes about the project, powered by MediaWiki. MediaWiki powers not only Wikipedia, but it is used also in other important projects like One Laptop Per Child. Thexa4 and everclear64 started filling it with some content available on the official website. If you want to help, you can open an account on the wiki and add your own content.
A deltasql was added to the project ( http://deltasql.sourceforge.net ). Deltasql is a tool that helps managing databases, a Database Version Control. As more server side components are added like the File Distributor, deltasql will help keep track of the database schemas. A JobPipe and ResultCollector component are planned and will be released during 2008.
Finally, Thierry noticed a mistake in how the Eternity plugin places the central piece. Its rotation is wrong in 0.950. The mistake was introduced in 0.9493, during a small refactoring introduced for the hints support. The problem is fixed with 0.9502 which is available on Tools->Autoupdate to latest version in the main GPU window.
Merry Christmas and Happy New Year!
2007-12-20 16:33:14 UTC by dangermaus
#
Monotone Boolean Functions and Script Engine
GPU in its latest version 0.949 has a new plugin written by Alex Fihman called Monotone Boolean Functions (mbf.dll). The plugin attempts to approximate Dedekind's problem by means of a Monte Carlo approximation. This is the first "real-world" GPU plugin that tries to solve a mathematical problem.
More information on this plugin can be found at http://www.fihman.com/mbf/mbf.html
0.949 also contains Thexa4's Script Engine, which is a novel frontend written in Visual Basic. This frontend adds scripting capabilities to the GPU stack language. It handles GPU answers from different nodes as a set with unique elements, stored into particular variables. The frontend has also loop and echo support. For more about the GPU stack language go to http://gpu.sourceforge.net/virtual.php
0.949 is also released with tons of bugfixes: in particular long strings were producing Access Violations (fixed by Alex Fihman), a memory leak was discovered. The job modifiers as described in chapter 5.4 of gpu_ext.pdf were not working. The upgrade from the lite to the full version did not work for the Orbit Reconstruction Extension. New entry nodes are added (Thexa4 and hisoka), and guevara is moved to a more accessible port (8080).
Double clicking in chat was not opening a search browser anymore. The mailing list menu item is replaced by a Wiki one. In the current version on autoupdate (0.9491), the total uptime bugfix is solved.
If the period July, September was very bad, the period October, November revealed itself as one of the most proficous for the project :-)
Stay tuned
2007-11-16 22:32:42 UTC by dangermaus
#
Eternity II plugin launched!
Eternity II plugin
------------------
On autoupdate there is a new GPU plugin which can help people solving the Eternity II puzzle. Eternity II is a puzzle with 256 pieces, which have to be placed on a 16x16 board. Each squared piece has 4 faces with a particular sign on it. Adiacent faces of neighboured pieces need the same sign.
A 2 million $ price will be paid to the first valid solution submitted to the puzzle creators.
A much simpler online demo of Eternity II is available at
http://uk.eternityii.com/try-eternity2-online/
The presentation of Eternity II with Lucy Pinder as youtube video is here: http://www.youtube.com/watch?v=RzjIctATZs8
Instructions to install the Eternity II plugin inside GPU are here: http://gpu.sourceforge.net/eternity.php
Network
-------
In June, a lightning hit one of the entry nodes (spartacus). Damages included the windows partition configuration files and the water cooling pump. After that until mid September, only Laurent's lwm-media node was available as entry node, the GPU network never was so close to shutdown. In September, Ninja tried to revive spartacus, but also the processor turned out to be burned. Ninja then upgraded the guevara node (an AMD Athlon 700 MHz) with low-noise fans and recovered DVD and 1Gig-Ethernet card from spartacus. An attempt to install Windows Vista failed due to the much more restrictive Genuine program. Therefore guevara is now running Debian 4.0 with wine and will act as entry node for GPU in the future.
Next scheduled release
----------------------
0.948 will be released only in mid November.
Have fun!
2007-09-20 12:47:35 UTC by dangermaus
#
0.946: Single Work Unit Set projects
The File Distributor component ( hosted at http://www.gpu-grid.net/file_distributor ) grew into a
complete solution to distribute work units and applications, suitable for small-sized projects.
The GPU plugin integrates with it, and a standalone client is available as well. The PHP set is
now used to collect computer information on the cluster. The hope is that this data will help
to understand the current scalability issues. With 0.946, projects with Single Work Unit set (typical
for rendering projects) are supported.
Below the release notes for the last three releases:
0.946: - Important File Distributor Plugin bugfixes -
FD server with computer/user/team statistics - FD server and plugin with Single Work Unit Set support -
GPU core reports periodically status to FD server (will be used later to create better clusters).
0.945 File Distributor plugin and standalone client ready for tests in the GPU community.
Possibility to distribute applications as .zip files. Uploading results to a FTP server.
0.944 Stability improved (no out of memory errors) /
PHP File Distributors enhanced: XML jobs, possibility to upload work units as zip files,
possibility to upload applications as .zip files, ftp upload settings /
File Distributor GPU plugin still in work... This is mainly a release for production purposes.
2007-06-01 18:52:58 UTC by dangermaus
#
0.943 Distributed P2P WAN DB by nanobit
First, the GPU organization thanks Laurent Wattieaux ( http://www.lwm-media.org )
for mantaining an own FTP server where occasional users can upload the Terragen images
computed by the GPU network.
After the file distributor, a new paradigm ideated and implemented by nanobit (René Tegel)
enters into the GPU arena. Technically implemented as two new plugins called pwansrw.dll and gpu2pwandb.dll,
the P2P WAN database uses environment variables, which can be synchronized between 2 nodes or the entire network by using hashes.
The paradigm is powerful and could be used to exchange results of computations between nodes without having any loss of data.
The distributed P2P WAN DB is described in the document pwandb_paper.odt.doc, to be found in the docs folder.
Several other improvements followed the release of 0.938.
- 0.939 introduced user management into the File Distributor (user statistics are planned in a future release).
- 0.940 introduced XML messages for netmapper. Due to the switch, version prior to 0.940 cannot see each other and vice versa.
This lead also to a simplification of the code.
- 0.941 was released only on autoupdate, its purpose being to stabilize the distributed P2P WAN DB plugin, which
was crashing on some platforms
- 0.942 introduced the Distributed P2P WAN DB by nanobit
- 0.943 has a new screensaver with zooming and autoupdate information, fixes for the Vista and Linux Wine platform,
and meanings to update the lite version to a full version.
The project finally sketched a roadmap, which can be found at http://gpu.sourceforge.net/roadmap.php
As usual, we thank all people who sustain the network by running the GPU client and who feed us with new ideas and source code.
2007-03-23 14:01:07 UTC by dangermaus
#
0.938: File Distributor Release
GPU is a client for distributed computing over P2P.
0.938 is released with a set of PHP scripts (based on preliminary work done by Justin Adams), which allow clients to download work units in a simple manner. The GPU client has an improved 'downloader' plugin, which can deal with the new file distributor. The plugin uses exponential waiting to avoid webserver overload. To get an idea, you can visit the project FD at http://spartacus.is-a-geek.net/file_distributor
More information on FD is also here: http://gpu.sourceforge.net/file_distributor.php
0.937 introduced labels on the screensaver, an internal refactoring which regarded all 3D frontends. GPU was recompiled with the latest
Synapse release 37 (instead of 36). Attempts to unbound UDP Gate on GPU Core failed. Attempts to add a screenshot to the computed html statistics failed, due to the old glut library GPU uses.
0.936 reports back data to the search engine in XML format. Some other maintenance was done in the automatic VM tests.
2007-01-02 15:47:43 UTC by dangermaus
#
0.935: Project reverts to plain GPL
After an internal discussion between team members, we decided to release 0.935 with the unmodified (GPL version 2), and to remove the public released versions beginning from 0.910 up to 0.934.
This for two main reasons: one is that Sourceforge.net hosts only projects that are licensed under the Open Source Definition. The project is not big enough to provide its own CVS and web space.
The second one is that GPL cannot be modified without changing its name. So we should have chosen a name like "No military Use Public License".
There was discussion going on for the GPL version 3 that regards a restriction for military use. Read for example David Turner's blog: http://www.fsf.org/blogs/licensing/20050211.html
Release 0.935 includes a new search plugin and frontend by nanobit, an updated DelphiPackagerTool by DelphiFreak and an attempt to include JPL planets inside Orsa.
2006-08-19 15:10:16 UTC by dangermaus
#
Discussion about modified GPL
What started like a little taunt suddenly got another dimension. The GPU project has modified the GPL license a little by adding Asimov's first law of robotics.
Meanwhile, we have been written be members of the Free Software Foundation, asking us to reconsider the change or at least not violate their copyright by removing the preamble and altering the name. We are aware modifying the GPL is not allowed by the GPL license itself, but did it without bad intentions. We go consider what is appropiate. After all, we're not after a legal conflict with the FSF. Give us some time for internal debate, we'll keep you informed.
2006-08-15 03:32:11 UTC by artee
'Seen once statistics', Search Engine Improvements, Bugfixes
The statistics engine was slightly improved (in the latest fix on autoupdate 0.9331) to keep record of all nodes which ever connected to the network.
We are now able to compare our community with the bigger ones in the distributed computing arena. Well, after running the statistics engine for some
months, we collected one year of uptime (which is only uptime, because the network is still idle most of the time). We indexed about 80 GB
of internet data. According to nanobit, the amount of crawled data should be 3/4 times the size of the indexed data. And fortuntately, filters are included by default, so we avoid to crawl quite a lot of crap :-#
Still some nodes can appear twice in the statistics, but the bug will automatically disappear as soon as everyone runs a version equal or higher 0.933.
The search engine improvements and a call for help is included in the "message from nanobit" which can be read in the
mailing list archive at: http://sourceforge.net/mailarchive/forum.php?thread_id=10295891&forum_id=40848
Remember to give a try at the search engine, the address is http://search.centraldatabase.org
Cluster statistics are here: http://www.gpu-grid.net/gpu/statistics/gpu_statistics.html
0.933 comes out with several bugfixes. If we compare with 0.930, we discover that we fixed a bug that involved duplicate packets on the initiator node.
Another bug, which caused some Intel dual cores to be wrongly detected disappeared. The behaviour of speed buttons in the core client is now almost correct.
The screensaver startup was fixed as well.
Other things happened outside in the real world: luciolis generated a terragen video on the cluster, the energia team
fought hard to get the c++ plugin working. nonobotae and lwm kept competing for the first rank in the hall of fame.
and some developers still have to learn the magics of jboss clusters.
#
Spanish Translation, Total Uptime, TCP/IP patch
The monthly release schema went broken due to time constraints and stress at work. Anyway with 0.930 we have a stable release which can run for entire months. The latest changes are on CVS and on autoupdate as usual. To get the changes described below, install 0.930 and then choose from the menu 'Tools' the item 'Autoupdate to
latest version'.
Spanish translation
-------------------
Alberto Pérez Garcìa-Plaza filed a clean spanish localization file. We thank him a lot :-)
To get Spanish, go to Edit->Settings->Language Options and choose 'Change Language'.
Total uptime computation
------------------------
As a long awaited feature, the total uptime of a GPU node is stored in a temper proof manner.
It still has some issues, if the node crashes, though. Total uptime got also the most
weighted factor inside the statistics formula.
Patch to unblock TCP/IP limit of 10 connections
-----------------------------------------------
We run on spartacus node the patch described here:
http://www.speedguide.net/read_articles.php?id=1497
which moves the limit of 10 concurrent connections on XP service pack 2 up to 50 connections.
This should help people who cannot connect to the GPU network. Note that the patch is not needed for occasional users, but it is important for nodes which have very long uptimes. We are running investigations to understand why so many people cannot connect. We might move GPU to some other port than the 6346 Gnutella port, which might be often firewalled. If you find out something regarding this matter, we would be glad to hear from you!
The TCP/IP patch by hacker LvLlord can be found here:
http://www.lvllord.de/
Miscellaneous
-------------
The Screensaver was slightly modified: colors of the graph were changed, and UTC time was added.
After autoupdating, you should choose Tools->Screensaver->Install to copy screensaver.exe inside the windows folder and then make 'Gpu 3d Network' as your default screensaver manually.
Thanks to the donation of blueapple2 autoudpate service was moved last week to http://www.gpu-grid.net/gpu. The domain http://www.gpu-grid.net is active as well and might host some of our webpages later in time.
What comes next:
----------------
We are planning the move to the SVN repository, which should solve some long outstanding CVS issues. Keep watching the mailing list for updates on this regard.
2006-04-12 08:03:30 UTC by dangermaus
#
0.930: Screensaver, XML, Uptime...
As often in life the release 0.930 did not follow exactly the roadmap with the Orsa plugin. It ended instead with a new screensaver, inspired by the BOINC screensavers. People often online on our network will not be too surprised, to see a 3D network mapper blown to full screen.
At the beginning of January, we released a .zip file containing statistics between November 04 and December 05. Beginning from January 06, we also generate and collect XML statistics, which might be parsed to a later point in time.
We additionally monitor uptime of nodes (which can reach 15 days and more) and approximate the scalability of the cluster by simply summing up the number of free connection slots of all nodes.
<lwm> backboned the network with his huge cluster of nodes. <nanobit> improved Search Engine data collection, although now we run the old webinterface due to some (ehm) technical issues. <paulatreides> is studying molecular dynamics [1,2] and the GAMESS program which might be bound in GPU. <simbasabi> is already at work to build a follower of spartacus, which is quite noisy, even if watercooled.
As long term roadmap for 0.940, it would be good if we would allow external developers to easily bound their .exe applications inside our network (applaunch improvements, file distributors are the keywords here).
Stay tuned.
[1] http://folding.mesdiscussions.net/forum2.php?config=foldinghome.inc&post=2&cat=12&cache=cache&sondage=0&owntopic=%20%20
[2] http://perso.wanadoo.fr/paulatreides/fichiertest/eau_calcul_frequences_IR.pdf
2006-02-27 20:31:00 UTC by dangermaus
#
More C/C++ examples, new orsa released soon
While Tiziano works sweat and tears to finish the exiting 'liborsa for pascal and gpu' project, others have put some effort to create the much requested C examples.
A simple C plugin we already had, but we were requested for frontend examples. User edsatori has put a small GUI example using BCB (Borland C builder).
I (Rene) made two examples using minGW/Dev-c++ demonstrating UDP communications with the gpu core. One shows how to register to netmapper or chat messages and how to execute a gpu command and fetch the results. The other is the gpu chat, interactive, but on the console as frontend. Both demonstrate how easy udp works, also in single-threaded applications. As suprise is a port from the famous megahal chat AI as gpu chatbot. Stuff for study for the C programmers who are interested in gpu.
2005-12-27 20:31:16 UTC by artee
#
C/C++ compatibility, Water Cooling, Nova-TS Team
In "Tapping the matrix" [1], a talk given by Carlos Justiniano (founder of Chessbrain), GPU is described as an active project with one big drawback. It is implemented, we quote, in an "obscure Pascal dialect". Ok, this was like a trigger for nanobit:
C/C++ compatibility achieved for plugins
----------------------------------------
Since the beginning, GPU was intended to load standard dynamic link libraries. We used pointer of chars instead of the very practical Delphi Strings for the communication between core and plugins. Using Bloodshed Dev-C++ [2], an amazing C++ IDE written in Delphi (ehm) and under GPL, nanobit was able to compile a C-DLL. The source code for the C plugin is under /src/dllbuilding/c_example. We had to change a little the TStack definition, so that the parameters were understood by the plugin. This prompted us to recompile all plugins, and recompiling everything was easy step thanks to Samuel Herzogs "DelphiPackageTool". So, we can now easily plug-in any sort of C/C++-code into the GPU system.
As final remark, frontends written in any language can communicate with the core, if they use the UDP/IP channel mechanism. We will try to provide some of them in some crazy languages, if we manage to get them working.
Water Cooling
-------------
Running a GPU node for long time is a challenge, one has to fight with dust, noise and neighbourhood.
A big thank to Stefano Godenzi (Ninja), who managed to install the Thermaltake [3] Watercooling system inside spartacus, one of the titled computers of the network. The pump as delivered by factory was wrong cabled, and some adjustement was needed. We plan to cover his work into a webpage, which we will made available on this site.
Nova-TS Team
------------
Laurent Wattieaux put up to 6 computers on the network, so that his Nova-TS [4] team often was top team in the GPU Cluster statistics [5]. By the way, we are collecting statistics since about a year, and the network itself is operating since one and a half years. We probably had an average of 6 nodes always on the network, with tops of 16-18 nodes. Ah, nodes with the LNX flag, are GPUs running under wine emulator.
Version 0.926 is the current stable release. It is compiled for C/C++ compatibility, has a whiteboard able to display nanobit's seismograph (more on this in a later news ;-), a netmapper which resizes better and let's the cluster spin on the center of mass. Bugfixes to autopdate, to the core as usual. The source release now includes a Changelog file, such that developers can better follow changes. Do not hesitate and ask our mailing list [6] for developer CVS access!
[1] Tapping the Matrix, http://bof.linuxforum.dk/2005/slides/Carlos/tappingthematrix.html
[2] Bloodshed Dev-C++, http://www.bloodshed.net/devcpp.html
[3] Thermaltake, http://www.thermaltake.com
[4] Nova-TS team, http://www.nova-ts.com/
[5] GPU Cluster Statistics, http://www.tea.ch/gpu/statistics/gpu_statistics.html
[6] GPU mailing list, gpu-world@lists.sourceforge.net
2005-11-08 11:18:08 UTC by dangermaus
#
Universal Time Plugin, Jedi Formatted Code
GPU is an experimental grid computing project which targets the rise of a big and cheap supercomputer over Internet, run by volunteers.
This is the project status for October 12, 2005.
First, we thank gratefully lwm (Laurent Wattieaux) for adding a new entry host for the GPU network, which solves the problem of the momentaneously broken spartacus node. A big thanks to paulatreides, for fixing some problems in earthsim plugin and reformatting most of the GPU source code with the Jedi formatter. Now the source code looks most of the time nicer and more readable. With time, we will reformat some parts where Jedi did not perform so well. The formatted source code is already on CVS and will be released with 0.925.
Lot of credits to sabybuld (Stefano Godenzi) for buying and planning the water cooling system which will be built in spartacus very soon. Delphifreak (Samuel Herzog) keeps his support to the DelphiPackagerTool, too. pisces computer is now running version 0.9194 for more than 850 hours (35 days) without application crashes.
Red's movie ShipRockFinal was published on the Movies section as well.
Universal Time Plugin
---------------------
nanobit (Rene Tegel) realized a new plugin which keeps synchronized the clock of any computer running GPU with the atomic clock of the U.S. National Institute of Standards and Technology. The plugin queries time.nist.gov at 5 minute intervals and consequently adjusts the system clock of the operating system in little steps of 100 milliseconds, so that other applications are not affected.
With the GPU command "timeerrorms, nodename", it is possible to see the difference in milliseconds in respect to universal time. This plugin is very important, if we would like to turn the GPU network into a sensor network. It would be then easier to compare data from sensors like weather stations or even seismographs. More on http://nist.time.gov
Search Engine News and Manual
-----------------------------
nanobit wrote an interesting news describing the progress done by the search engine here:
http://www.tea.ch/gpu/updates/gpuse_news_sept25-2005.pdf
Try to query the GPU network at http://search.dubaron.com with some keyword like "grid computing". The query interface is really impressive, especially if viewed with Firefox or Mozilla browsers.
A new manual on how to use GPU together with the Search Engine is available at
http://www.tea.ch/gpu/updates/gpuse_manual.pdf
Other improvements
------------------
GPU Core improved only slowly, providing new functions to load and discard dynamic link libraries one by one. This lead to DLL Manager fixes and to an autoupdater which does not always need to close GPU to perform its work. There will be more work on Core in the upcoming month of November (in particular fixes to the frontend manager, allowing more frontends to register to the same service).
Terragen shipped with GPU was updated from 0.9.21 to 0.9.43. This step should fix some rendering bugs.
Finally GPU is shipped with a full (16MB) and lite (6MB) install. Releases are now marked as "test" when they are released. Depending on how they perfom, they will be renamed in "stable" or "unstable". In fact, 0.923 was quite weird, we apologize for any inconvenient it might have created to our community!
Finally, be sure to check out the Seismograph section on http://gpu.sourceforge.net!
2005-10-12 15:18:25 UTC by dangermaus
#
Search engine boosts, new icons, GPL for non military use
GPU is a software for clustering over P2P. This is the project status for 31 August, 2005. Current release is 0.921. Here the areas were there was improvement.
Search Engine
nanobit (Rene Tegel) did a lot to improve the search engine. The new library sqlite3.dll (that replaces sqlite.dll) is able to handle UTF8 characters, so that languages like Chinese are supported. The updated search plugin now does not commit each time a new webpage is crawled but waits for several changes and then commits. As consequence, the journaling of NTFS is much less stressed, and disk usage is reduced a lot. Finally, the webinterface on http://search.dubaron.com that runs mysql is able to work in synergy with the sqlite plugins on the GPU network.
Look & Feel
All images inside the client are externalized to the subdirectory icons. An ini file (boot.ini) allows to customize some of the frontends buttons. Sabybuld (Stefano Godenzi) created a new set of icons for the frontends, which definitely look better now. Additionally, some details on GPU core were corrected: many bug chats were corrected (including the annoying bug of occasionally missing end of line). Closing GPU now closes all frontends as well. The hooks to show Pastella connections were missing until now in GPU, they are now included, but we need more work on this.
Improvements to whiteboard
Again nanobit included webcam support to the whiteboard, if users of the cluster are equipped with a webcam, they can create a sort of virtual view of the cluster on the whiteboard. Rumors say, that even dangermaus black & white TV was broadcasted on the whiteboard :-)
GPL for no military use
Following inquires of Nick Johnson (npj), we decided to create our own version of the GPL. The text can be read here http://gpu.sourceforge.net/GPL_license_modified.txt . Goal of the modification is to prohibit military use, without giving away the advantages provided by GPL
Miscellaneous
Website FAQ was updated, a new site http://gpu.sourceforge.net/titan_movie.php tries to present the work done on January to recreate Titan's conditions using Huygens space probe data. Red produced a new movie "ShipRockFinal" which is available as a preview on our ftp servers.
However, you might encounter problems in connecting to the GPU network. spartacus entry point (also built by Stefano Godenzi) got too noisy and waits for a new water cooling. If you would like to provide a computer running GPU as entry point, please read http://gpu.sourceforge.net/advanced.php#publish and contact us on gpu-world@lists.sourceforge.net
Thx for your help!
[comments from rene]In fact the change from sqlite2 to sqlite3 has nothing to do with utf-8 support. Webpages in foreign (detectable) code pages gets converted to utf-8 before indexing. This works as well for sqlite2 as sqlite3. For developers, sqlite 3 may be more convenient due to widestring/utf16 support.
The change to sqlite3 was merely to be forward compatible and due to some small changes in design it was a convenient moment to drop existing databases and switch to sqlite3 simultaniously.
2005-08-31 14:00:42 UTC by dangermaus
#
File Distributor, Delphi 2005, Whiteboard, Balloons
GPU is a cluster software with a P2P based architecture. This is the project status for July, 17, 2005.
Despite summer hole and high temperatures, both the cluster and the development kept running at sustained rhythm.
Network Status
On July 12, <Budtske> connected his Internet Coffe to the GPU system. During that day, 24 nodes formed one unique entity and joined their forces to compute a movie by <Red>. Smoking like an old steam motive, the cluster broke the 30 gigaflops barrier.
http://www.tea.ch/gpu/statistics/july05/gpu_stats_120705_80GHz.html
The Network Mapper showed overheated connections, so work to reduce bandwidth will be soon unavoidable.
A movie generated by <Camsinny> some time ago, "Around the peak" is now available for download, too.
File Distributor
Justin Adams (<shakezilla>) wrote a set of PHP scripts that can be used by any centralized or P2P based project to download files and mark them as processed. The downloader plugin was adapted to implement exponential waiting. An uploader plugin is on works. However, scripts still do not work through Apache server. It might take some more months until we get something running, but the path is set.
http://gpu.sourceforge.net/file_distributor.php
Delphi 2005
<paulatreides> fought a battle with the new Delphi IDE and got everything compiled into Delphi 2005. As important result, we got some bugs resolved caused by the Delphi 7 compiler into GPU core. Additionally, Delphi 2005 Personal Edition was distributed for free in computer magazines last month. Modifications compatible with Delphi 2005 and 7 are on CVS and will be in the next 0.919 release.
Whiteboard
<nanobit> wrote another frontend, a whiteboard to share draws among people on the cluster. Lines, ellipses, images travel through the chat service channel and form funny draws, formulae and network charts.
Balloons
The trayicon of GPU informs the user with balloons if new user enter or leave the cluster. If GPU is minimized, chat sentences pop up, but with a frequency of one in two minutes. This feature can be deactivated in the configuration settings, if it should bother too much.
Additional improvements included in release 0.918 are:
- GPU detects SMP computers and computes their MHz in a more accurate manner.
- Terragen wrappers work with multiple FTPs
- Some more bugfixes
Might the force be with you, rebels of cyberspace ;-)
2005-07-17 17:34:12 UTC by dangermaus
#
GPU 0.916: DelphiPackageTool and additional 2D statistics
This is the project status for GPU, a "Gnutella Processing Unit" on May, 22, 2005. GPU is a software able to cluster computers around the internet in a random graph.
Network status
----------------------
Some problems for the entry points <iap_mohan> and <spartacus> seem solved now. Great thanks to <swiftstream> and <leo> nodes that kept the network running despite all troubles. In order to survive, the GPU network needs more entry points. Read here how to create one: http://gpu.sourceforge.net/advanced.php#publish
Unfortunately, there were no video computations going on these days, too. Additional anonymous ftp servers that could host computed images would be cool...
2D graph on network mapper
------------------------------------------
The network mapper features a new 2D graph (realized with Delphi TChart), that keeps track of GFlops, Terrastones, Crawlostones, Number of nodes, Traffic and Cluster load in real time. To increase the special effect, statistics frequency increased from once in 2 hours to once in 2 minutes without stability loss. E.g., it is now possible to follow how the amount of crawled data increases. Or to look at the amount of available GFlops computed by short FFT calls issued each 2 minutes, while the node is idle. By keeping netmapper running for several hours, the fluctuation of all parameters becomes evident.
Package changes
--------------------------
DelphiPackagerTool is included in 0.916. It will help developers installing GPU faster into Delphi IDE. DelphiPackageTool is located in the directory \install\dev and can compile in one shot several plugins, frontends and the core client.
Menus for the core client
------------------------------------
The core client has now menus following request of many users. It will take a while until translations make their way through, though.
Other
--------
Some initial experiments with lazarus and Freepascal were done. One far day, we hope to move all code there :)
Ok, that's all. Let's hope Bzzz. the project will not fall on the dark side of the power. Bzzzz. Pop.
2005-05-22 19:44:21 UTC by dangermaus
#
New open source connection layer for GPU: Pastella
Pastella is an Open Source protocol implemented as Delphi library and based on Synapse, that allows applications of any type to build their own P2P network. Pastella is in early beta stage and was written and designed by Rene Tegel (nanobit). Pastella is an alternative p2p protocol, just like gnutella is. The main goal is to perform network optimalization, at the cost of some memory power and cpu cost for some intelligent caching and hashing. It's design goals are: less network overhead, avoidance of redundant data, good broadcast support, minimalization of packet loss, host-to-host routing by other hosts and walkers and agent support.
Why is it called Pastella? Well. there are several fantasy-rich theories about that. For example, that we replaced the GNU from GNUtella to PAStella to honour the language Pascal. Another one, cause some of us like Pasta :)
New GPUs will first offer an hybrid between the commercial TGnutella component written by Kamil Pogorzelski and TPastella.
While TGnutella was very stable and suitable for commercial purposes, (and Kamil Pogorzelski was very kind to allow us to use it on the GPU project with only one license), it is now time to try to reach independence with Pastella. Kamil's TGnutella is doing an excellent work, but we would like to have an open source license on the connection layer and solve some issues that raise up with porting it to Linux.
Right now, we have problems solving the network entry point problem, which is not related to TGnutella nor Pastella, but more to trouble in running our network infrastructure. We are sorry for the inconvenients.
Pastella is available at http://sourceforge.net/projects/visualsynapse
TGnutella is available at http://delphi.pogorzelski.de
2005-04-23 16:50:55 UTC by dangermaus
#
GPU 0.912: new package structure, chatbots
This is the project status for GPU, a "Gnutella Processing Unit" on March, 24, 2005. A list of recent changes, known issues and information on network status follow here.
Before we go on, we thank a lot blueapple2 for his donation, which boosted our motivation quite a bit. We still have to decide what to do with it.
Package structure
-----------------
Samuel Herzog (delphifreak) sent us a new proposal for the directory tree of GPU. He also adapted his DelphiPackagerTool, available from http://www.novabit.ch, to help developers to install the myriad of components needed by GPU in one shot. All frontends, plugins and mainapp contain relative paths to libraries and components, now.
The new file tree is described in the Readme.txt at the top level of the source zip file. From now on, the zip package will not contain binaries anymore, so that we have a clean separation between source and binary distribution. gpu_solar module was succesfully imported into CVS yesterday.
The structural change is important and therefore GPU gets a brand new codename: solar substitutes matrix. 0.912 is still transitional: the last glitches should be fixed in the next two-three releases.
Chatbots
--------
gina.dll is a new chatbot by nanobit that will entartain users in the sad hours, when the chat is emptier than deep space. It is a port of his markov-chain chatbot. To create a new genious of the caliber of an Einstein, Kant or Shakespeare, one should first collect a clean ASCII file with their work. This ASCII file should be named Einstein.brn and put into the main gpu directory. Issuing the command "gina_enable" followed by "1,'Einstein.brn',gina_brain", Einstein will pop up on channel 1 and enlighten us on general relativity issues. Similarly, Kant could teach us about the Kritik der reinen Vernunft. Any language works, thanks to the markov chain principle.
GPU implements a new chat service, similar to the netmap service, allowing frontends and plugins to listen for chat activity. The simple frontend allows registering for both services, so that developers can figure out how the system works. Documentation on this can be found under gpu_ext.pdf and is still far from completion.
Huygens movie of touchdown on Titan
-----------------------------------
Some Huygens movies can be seen here: ftp://spartacus.is-a-geek.net/movie/new_ones
In particular we thank MM, nanobit and paulatreides for their effort. MM used a picture of Cassini and got better results. Huygens image quality is very poor and presents difficulties when imported into a Terragen terrain file. More work on Huygens movie is planned.
Other changes
-------------
GPU system underwent more changes, including direct messages support as requested by Dr_Moebius. The ORSA extension is able to simulate doppler oscillations of Pioneer transmissions, mysql search engine improved speed and is available as webservice on http://search.dubaron.com , chatbot service and simple frontend improvements were discussed before.
Network status
--------------
We thank first Manic Guru, swiftstream, lwm1 and Seeschloss for their support in running the GPU application.
Network load and activity was lower in the period february, march and was partially dued by structural changes. iap_mohan upgraded OS from Windows 2000 to XP. iap_mohan covered night activity (GMT+1) and is currently not ready for operations.
spartacus moved house. spartacus is a 486 box with a 586 heart, built by simbasabi (Stefano Godenzi). simba told dangermaus, be careful with the feets of spartacus, cause the holes for screws do not match exactly. 8 months later the warning was lost in time, and dangermaus managed to throw downstairs spartacus for one stair as the foot broke, while moving house. However, nothing serious broke internally thanks to the robust design of simba and sparta is now operational 24h a day.
Known issues for 0.912
----------------------
- file sharing is disabled
- terragen stats keep reporting -2 for work done
- using big fonts creates layout problems on GPU
- search engine does not work on more than one machine, because reporting of partial results seems broken
The report for 0.912 finishes here, 73 over.
2005-03-24 09:50:06 UTC by dangermaus
#
Computing the movie of Huygens touchdown on Titan...
After a journey of seven years with companion Cassini, Huygens space probe landed on Titan's surface on 14 January.
The probe sent back around 350 images to mother ship Cassini, which relayed data back to Earth.
As soon as the raw images were published, computer enthusiasts pounced on the images immediately, and improved them using a range of free or commercially available software before swapping their pictures in Internet chatrooms. One of these tools was Terragen. ( http://www.nature.com/news/2005/050117/full/050117-7.html ).
Nanobit managed to import an ESA image as Terragen terrain file as well. GPU is now computing a movie of Huygens touchdown in distributed fashion, although we still do not know many parameters varying from sun disc size, atmosphere thicked (is it all cloudy?), redness of light on ground and so on...
Everyone who wants to help us to compute the movie can download the GPU program and autoupdate it to a version which should be at least 0.9089. Then, it is necessary to make sure that the Terragen plugin is activated (GPU window - Advanced plugins and frontends). Release of 0.909 is planned in the next days and will make autoupdate step not mandatory.
We plan to attach the radar sound of the touchdown ( http://www.esa.int ). After showing the touchdown, the movie will take a ride on Titan's rivers.
We also invite Terragen artists and everyone who can provide us with information on how to make the movie more realistic to contact us at gpu-world@lists.sourceforge.net
Thank you!
2005-01-23 23:08:17 UTC by dangermaus
#
Project status (January 2005)
This is the status report for the project GPU, a Gnutella Processing Unit for January 2005.
Cluster statistics
----------------
Since 8 November 2004, the cluster is monitored. The first manually generated graphs reporting gigahertz of speed, FFT benchmarks, crawled data and generated images are here:
http://www.tea.ch/gpu/statistics/stats_graph.html
Current statistics are now in HTML format:
http://www.tea.ch/gpu/statistics/gpu_statistics.html
Thanks to johnatemps and Skystation for their presence on the cluster.
Generated data
---------------
For the first time, the framework was used in a sort of productive manner.
Around fifteen movies were generated with Terragen. Worth to mention are paulatreides project_lavelanet which explains how the camera moves on the landscape, nico's Lord of the Rings, Red's incredible basejump and Red's Visit of the Monument Valley in Utah with real data of the U.S. Geographical Survey.
Client improvements
-------------------
Between 0.900 and 0.908 the main improvement was in stability. nanobit fixed such an amount of memory leaks that dangermaus position changed from developer to memory leaker. The best way to recognize it is to run GPU under wine emulator and this for weeks (disabling applauncher.dll is recommended :-) Other improvements
like plugin manager for advanced extensions and reorganization of layout with panels were done. nanobit rewrital of the core loop of TComputationThread increased speed. The netmapper is now able to display connection traffic with a heat color map (black: no gnutella packets, white: more than 10kbit/s bandwidth).
Documentation
--------------
Documentation was splitted into Users and Developers section. New user's docs include an introductory tutorial in Quicktime format (camera: Martin Waldburger) and a powerpoint presentation presented at the coffee break. New developer's doc include a preview of nanobit's improvements called "GPU Extensions for Terragen and the Distributed Search Engine". While forging one line of code, developers think hundreds more they would like to implement. This effect created a semester/diploma thesis proposals page:
http://gpu.sourceforge.net/thesis-ideas.php
Long term goals
---------------
In the powerpoint presentation ( http://www.tea.ch/gpu/GPU_with_extensions.ppt ) long term goals for the project are stated. Licensing issues are shortly addressed: keep core under GPL, move away from commercial TGnutella, allow commercial extensions. There is hope to get nanobit's TPastella with Distributed Hash Tables as new connection layer. Integrating the Orsa framework by Pasquale Tricarico into GPU might be another target. Orsa is a framework to compute and analyze orbits inside the solar system, including Near Earth objects that might hit the Earth.
2005-01-07 12:57:22 UTC by dangermaus
#
GPU 0.900: the sleeping giant is awaking
GPU, a Gnutella Processing Unit is a Gnutella Client that can perform computations on a peer-to-peer based grid.
Version 0.900 includes a plugin able to render virtual landscapes. The plugin is a wrapper for the famous Terragen program [1], free for non-commercial use. Videos already generated by the framework can be downloaded from here:
http://www.dubaron.com/terragen/movie/
We thank Terragen artists nikoala, paulatreides and nico for their engagement.
The GPU core experienced improvements: the core loop
inside ComputationThread was rewritten by nanobit and results follow: GPU now works without bugs on symmetric multiprocessor machines (SMPs like <gemini>, one of our official FTP servers), the core loop is at least ten time faster and the nasty bug with non-terminating threads is solved.
Many bugs in chat were fixed as well, including channel bug and 'is typing a message' bug. Chat was improved with clickable URLs. Felipe Mendon?a sent the Portogues e translation. New documentation is available in the /docs/gpu_ext subdirectory of GPU partially explaining Terragen and the Distributed Search Engine.
Aggregated statistics for the cluster can be seen here:
http://gpu.sourceforge.net/clusterstats/gpu_statistics.txt
They are published on a daily basis on sourceforge, but can be computed at any time but any GPU using the network mapper.
At time of writing, the framework delivers about 10 gigaflops of speed while 2 gigabytes of virtual landscapes were generated. 10 users are online mostly all the time. The cluster should be able to grow up to about 50 users and then hit some scalability limit.
Also, we have to deal with slow ftp servers and full hard disks. Is the sleeping giant awaking?
[1] PlanetSide Software http://planetside.co.uk/terragen
2004-10-27 13:19:27 UTC by dangermaus
#
GPU 0.898 - Render a virtual world movie
With the newest GPU (use autoupdate) you can help rendering movies of a virtual world, using the new plugin 'earthsim' that launches terragen to create a frame of this movie. Normally creating these animations can take weeks. Using distributed computing, terragen artists have a way of rendering their work in strongly reduced time.
As of version 0.898, the GPU plugin for terragen http://www.planetside.co.uk/terragen/ is declared 'stable - release candidate'.
If no more serieus issues are reported (they will be fixed with version 0.899), gpu is ready for a very interesting 0.9 release!
2004-10-09 10:39:28 UTC by artee
#
Help us to crawl it all :-)
GPU 0.883 is released with the latest robust search plugin written by nanobit (Rene Tegel). We hope some new users will join our network to collect impressions, bugs and new ideas. At time of writing, it is possible to search on about 20000 crawlostones of Internet data (20 GB). This number experiences fluctuations dued to nodes living and joining.
The search frontend features now a Visual Synapse webserver, so that any GPU can offer a query interface on the web. Additionally, for experienced users, a mySQL version of the plugin is available (more details on http://gpu.sourceforge.net/search_engine.php ).
GPU itself implements now a fast Visual Synapse DNS client, instead of the broken and slow Delphi 5 Indy DNS component. The underlying TGnutella component is now more robust, duplicates are less frequent, although more work should still be done here.
The question on how much GPU really scales is still open, probably with 0.883 we could reach maybe around 100 nodes.
The Visual Synapse webserver component opens new horizons, cause we will be able to offer other services, even to people that do not run the GPU client.
The network was well shaped these days, and we thank <Cheese> and <lwm> for their continous support.
2004-08-12 17:38:57 UTC by dangermaus
#
GPU 0.878: color maps - chat improvements - CVS
GPU 0.878 (available on autoupdate) features several improvements, most of them done by nanobit (Rene):
Changes in current version
-------------------------
- chat was improved with user's list, notification of typing and bell. The way the duplicate's problem is solved for chat will soon be implemented for GPU packets as well.
- a much faster search plugin is implemented. Database is split into four files, so that the database can stay on FAT32 up to 16GB of data (each file on FAT32 can be 4GB at maximum). FAT32 is much faster than NTFS for crawling purposes.
- the GPU engine features now callback for plugins to send back multiple results. Callbacks as implemented by nanobit are easier to use.
- GPU has a list of hosts tried once a minute and stored in gpu_fix.txt. The reason we use an additional list to the well known gpu_hosts.txt is simple: it will allow to setup a sort of backbone for the network.
- Feynman-Kac frontend implements now several interpolated color maps, to display function results graphically.
- Chess plugin displays score of actual move.
CVS setup
----------
CVS module gpu_matrix is now actively used in team and in a productive way, although sometimes we implement features twice :-). Tortoise CVS and Cygwin are very useful.
Network status
--------------
Best network supporters were <Cheese> from New Zealand, <lwm> from France and <kosia> from Poland. Connections from new users dropped to one in two days.
Backbone of the network held by <nanobit>, <oleander> and <spartacus>-<iap_mohan> tandem.
New node added: <iron-linux> running GPU with wine under SuSE Linux 9.1.
Crawlers
--------
<nanobit> and <spartacus> collected about 12'000 crawlostones (12 GB) of Internet data. Database is now dropped because of new database definition.
As final remark: July seems a cold month with lot of programming weather :-)
2004-07-12 21:03:57 UTC by dangermaus
#
GPU 0.868: Distributed Search Engine by nanobit
0.868 and 0.874 autoupdate feature a new plugin/frontend couple written by nanobit. This extension allows GPU to crawl the web and index it using a SQL database. Users can then search each database created indipendently by GPUs for strings contained in the website or in the URL.
Using the search frontend, the user first submits to the crawler initial websites (Submit URL tab). Then, the user can enable a different number of crawlers. After selecting a place for the database, local statistics will show the visited URLs and display graphically statistics about CPU and network bandwith usage. The new GPU command "crawlostones" will return the number of megabytes crawled by the local GPU (Compute locally button) or crawled by the entire GPU network (Compute globally button).
At time of writing and in a few days, around 6000 crawlostones were collected and about 300'000 pages indexed. This is about 1/10000 in respect to the Internet size. Right now, GPU Search Engine (abbreviated with SE) does not implement robots.txt, so that some websites might decide to block access from the crawler's
IP.
In the recommended autoupdate to 0.874, the new command "speedostones" is introduced: this will perform
a benchmark of the computer running GPU using the FFT routines in fftplug.dll.
0.874 core client enhances frontend management so that multiple jobs can be sent at once and implements an additional UDP/IP channel with VisualSynapse for communication with frontends. 0.874 has the ability to run GPU with a normal set of plugins, or with the normal set plus a set of experimental plugins.
Finally, some network status information will be found in the forums. Shortly, <spartacus> had some problems, <nanobit> entered as new permanent host and substituted <X_spider>, <andromeda> laptop broke its 30 GB harddisk and <iap_mohan> schedule was extended a little bit.
lwm reported GPU running under Suse Linux 9.0 with Wine emulator. Months before, seeschloss reported GPU running under Linux with VMWare framework.
2004-06-08 12:54:25 UTC by dangermaus
#
GPU 0.860 string as parameters
0.860 and further versions on autoupdate contain an extension of the virtual machine that allows strings to be passed as parameters to plugins.
New plugins were introduced: strbasic.dll shows how to use strings as parameter. downloader.dll is
able to get URLs as parameter and start a file download through HTTP. downloader.dll is the first step toward the "The centralized model as subset of GPU" ( http://gpu.sourceforge.net/gpu_p2p/node27.html ).
0.862 introduced FFT routines by Joost Ruysch in fftplug.dll. The core routine is optimized in assembler. Speed of this routine is said to be around 20 times faster than FFT routines used in Seti@home.
0.856 introduced thread priority settings and user/local time for chat entries. 0.862 (on autoupdate) introduced the possibility to load and unload experimental plugins with extension .exp. Plugins are marked as experimental if they are not stable and cause GPU crashes. 0.864 introduced ping intervals of 10 seconds to keep connections open if they go through firewalls.
nanobit (Rene) is working on a Distributed Search Engine driven by GPU. The results are already quite impressive and can be seen by autoupdating the application. Configuration is still a little bit difficult, ask developers lurking in chat for further help... Source code for this new plugin-frontend couple will be released as soon as the database structure is almost definitive.
Future directions for the projects include: UDP/IP visual synapse as further channel for communication between plugins and frontends, webscript to get data from a webserver. See the Help Wanted section for more details.
2004-05-23 00:11:12 UTC by dangermaus
#
GPU 0.851: bug fixes by Bjorn Madsen
Changes in current version
-------------------------
Bjorn Madsen fixed two dangerous bugs in GPU version 0.851: he fixed a memory leak in the procedure Gnutella1SearchResult (in gpu_mainform.pas) that was reserving memory each time a request for files was issued. This might have been the cause for slow GPU responsivity after days of running.
The second bug he fixed was in TGnutella: uploaded files from GPUs clients were one byte shorter. This is seldom a problem if the file is music, but it is very serious is the file is compressed. Another thing he noticed and that was fixed is that the downloaded list with GWebCaches was useless because all caches were closed in the meanwhile. Other minor changes in Network Mapper were done.
0.853 is available on autoupdate: GPU collects but does not display statistics data if the statistics are not shown. This significantly reduces CPU-time consumption (from 2% to 0%) if GPU is idle. On <X_spider>, the fan was running faster because of this problem.
0.853 contains now a primitive priority queue for incoming packets: for each computer, only 5 jobs can be put in another computer's queue. This to avoid flooding attacks, mainly caused by improper use of the Feynman Kac frontend.
Autoupdate service officially moved from ETH to tea.ch server. This because dangermaus account will be closed very soon.
Network status
--------------
<X_spider> from Canada, owned by Simon Blondeau was new permanent host and reliable meeting point. <lwm>, <lwm-zoom>, more recently <nanobit>, <glc> and a node from Hong Kong were very often connected.
The tandem <iap_mohan>,-<spartacus> was only backing them. In particular, <spartacus> had some trouble in standig up in the morning and got a virulent Sasser influence while the student's network was experiencing problems as usual.
Interesting discussions were held in Chat, and some of the proposed ideas might be part of GPU very soon.
2004-05-09 15:42:10 UTC by dangermaus
#
Thesis out, Trigger, Donations
The thesis with title "GPU, a framework for Distributed Computing over Gnutella" is finally out
and published as html and pdf here: http://gpu.sourceforge.net/doc.php
Please inform us of mistakes and in particular if we forgot to mention some contributors at the end.
<iap_mohan> resumed normal operations on April 9, although it will run only at night between 00:30-6:15 (GMT+1). During the day, <iap_mohan> will serve customers in Poschiavo as normal Internet Access Point.
We use the little 'Trigger' utility, available in the package 'GPU components', to start and stop GPU at given times.
<spartacus> will run approximately between 09:30-23:30 (GMT+1). There are time windows, where no hosts are available as meeting point:
between 06:15-9:30 in the morning and 23:30-00:30 in the evening. Unfortunately, we do not provide plugins on <iap_mohan> anymore, because we had stability problems and abuses.
On the contrary, <spartacus> runs with all plugins.
We added a note in the download section to inform people that port forwarding is needed in order to have more interesting network topologies than the current centralized one, where all clients connect to the same meeting point. Any network topology is allowed in a LAN by using the 'Connect' button. We also tried to move the focus on the webiste from file-sharing to grid computing.
The project is open to donations and to eventual sponsorships. On our side, we will try to mantain the current infrastructure working (tandem <iap_mohan>-<spartacus>) and add some more nodes. We will slowly move away autoupdate services from ETH infrastructure to the http://www.tea.ch servers. For this, a GPU update is not required.
No real 'breakthrough' in development is planned for the next months. And in fact, work is still needed to solve the duplicate's problem. As always, we are open to your suggestions, in how
to improve and continue the project.
Finally as personal note, let me thank from this column David Jensen, a researcher at IBM Watson Labs in NJ, for installing the Globus Toolkit on my laptop.
2004-04-10 11:00:41 UTC by dangermaus
#
GPU scales up to 64 machines, <spartacus>, Polls
Last wednesday, we tested GPU in a particular configuration: each computer was connected to 4 neighbours only and we had a sort of 2D-grid. Doing so, we did not have an exponential growth of results. After adding 23 computers (21 on a student's room, one in a student's house and one in Russia <tm>) we had almost linear speedup. Because of TTL constraints, the grid can get only 8x8 wide, so GPU scales up to 64 machines for now.
I think, we are not far away to get normal Gnutella scaling (up to 500'000) nodes. A photo of the cluster with the Network Mapper view is here: http://gpu.sourceforge.net/download.php
<spartacus> is finally in place thanks to Nigia. It will be the meeting point for the next months. However, it is difficult to sleep with that noise, so it is online only during the day in Europe. We might write a little trigger program that starts and stops GPU during the night on <iap_mohan>.
We have now polls on the website thanks to Ayhan (on the left of the window). You can answer more pools here: http://gpu.sourceforge.net/polls . Thanks for your feedback!
The thesis is finished and out the pipe, but it will take until next wednesday, until we get the final ok to publish. There will be a PDF, ps and HTML version (created with the amazing html2latex tool).
Autoupdating to 0.849 will delete the crabs_simulator, that creates problem because of the popup window. We will work on a version that can run only locally.
2004-03-26 16:51:03 UTC by dangermaus
#
Call for help on Thursday 18 for a benchmark
It would be very nice if some of you could connect GPU to the international network on Thursday 18 at 21:00 (Time of Zuerich, GMT+1) by pressing the Start button. To get times right, the dynamic clock of Ayhan ( http://gpu.sf.net/map.php ) should be helpful. From then on, for about one hour, we will send out jobs and perform measuraments on how many packets we receive back. Results will be published in the thesis at the end of March. Remember to set your nickname and country in the chat window. Thank you! For more info mail to dangermaus@users.sourceforge.net.
Current network status is bad: <iap_mohan> will be down for the entire week: a new webcam and an usb smartcard added to the internet access point seem to create problems with GPU: lot of access violations are reported. However, we hope GPU is not the cause of these problems in this era of viruses and spam.
Beginning from next sunday, I hope to publish a new node on the GPU network with the help of Nigia, at least during the day (8:00-22:00 GMT+1). Its name will be <spartacus>.
Thanks to Ayhan, we have a new poll section on the website; scroll down on the left to see it. The homepage received a book section, where we advertise a little for Wes book.
The address for GPU statistics was wrong in the last news. Correct address is http://gpu.sf.net/stats
2004-03-14 11:15:00 UTC by dangermaus
#
Project status (March 2004)
Sorry for being late with the weekly news, but stress around here breaks us apart.
We were working on two theoretical problems for a while: Wes right found now the solution for the coupon collector problem. We have an argument for the Milgram's experiment, too: according to our little estimation, Gnutella should be a fractal with dimension about d=7.8. But this is not new as well, there are books on social physics and fractals on Amazon. Anyway the problems are interesting. How many people are between you and Musharraf? How many coupons do you have to collect until your figure's book is completed? In which way do they relate to GPU? Answer should come around the 25 of March.
dangermaus Delphi is broken, so we are performing some benchmarks with version 0.847. Nicely, if we connect nine computers in a line, the last computer will not compute because packets contain a TTL of 7, cool eh? If we connect this 9 computers in a star, we have packet loss because the central computer is overloaded. We still have to perform tests in the tree and random graph topology.
The website received a partner and book section. Sorry for being so late in acknowledgeing people who help us.
Network status was fine: We saw often <nanobit> and <nono> in the beginning of the week. <rv73-USA> and <SeeSchloss> often were online in the second week. <HE9ZMM> is another machine at ETH Zuerich.
GPU statistiscs (public, you can see them at http://gpu.sf.net ) had jumps due to the fact that http://framasoft.net and http://ratiatum.com provided a short description of GPU on their homepage. We thank them, France is top nation for the month of March because of this.
Ayhan experimented with a script to add time zones to the network maps. It works fine under Mozilla, but not under IE. You can see it under http://gpu.sf.ent/map1.php . If you have ideas on how to solve the compatibility problem, we would be glad to hear from you.
Phyrum donated a Pentium II 100 MHz to the project. We will use it to test GPU on Windows 95-98. Thanks!
Finally, I do not have time to perform tracebacks on IAP, if someone is interested I can mail a filtered log file.
ciao!
2004-03-07 14:56:08 UTC by dangermaus
#
Welcome to Iran, Brasil, Malaysia!
A warm welcome to Malaysia, Iran and Brasil. These
are three countries we did not previously see in the
logfile of <iap_mohan>! Hundreds of tracebacks can be seen in the Network Status forum.
A thank to <cloudtee@am> and <ChandraX> for their useful tips, too.
The project is now in a documentation phase: new in the docs sections is the "Advanced Tips and Tricks" document.
The thesis will be published on March 24, although now it is clear that GPU scales only to around 20 computers. This is a bug and not a feature, of course, because normally Gnutella can scale to 500000 nodes. We will fix it, but it is difficult to say when.
On Friday, we published the Floppyx distribution of GPU. The name is similar to Knoppix, a Linux distribution that can be started from a CD. GPU can
run from a floppy disk, although there was not enough place to save the Network Mapper and many plugins. A batch file to create the disk is provided as well.
This week was signed by little failures:
- <iap_mohan> was down for unkown reasons since
Friday 13 (!) for three days. <SeeSchloss> and <SeeSchloss[2]> jumped in heroically to stop the hole
- Delphi 7 of dangermaus broke for unknown reasons
- ETH AFS where the autoupdate files are hosted was unreliable
- Little sections of the thesis went lost
2004-02-22 13:02:06 UTC by dangermaus
#
GPU 0.845: Feynman-Kac initial release
GPU 0.845, a Gnutella client for sharing CPU-time, is released.
Following improvements were done from 0.840:
- Feynman-Kac plugin and frontend solve a partial differential equation in a 3D domain (ellipse). Documentation will follow to explain it a little.
- Netmapper displays multiple users behind the same IP (useful for example in student houses where all connections are behind a router with Network Address Translation). Late answer management is improved: a late GPU is now added correctly to the list, the sphere changes from gray to a color.
- advanced seed management: the first time GPU is installed, a seed file is generated. This allows reproducibility of results (one of the fundaments of science) at least on one machine.
- Bug fixed: connecting to hosts with different ports (like seeschloss.no-ip.org:1789) does not generate an error
- possibility to disable writing in logfiles in the preferences tab
- gpu_hosts.txt included; if ETH account is down, people can connect to the network anyway.
- Frequently Asked Question added.
Autoupdate 0.846:
following a request from Stackem, 0.846 does not clean the panel at the end of the computation
Network Status:
Backbone formed by <Seeschloss>,<iap_mohan> and <albatros>. Support from <bernina>. Less users connecting from around the world.
2004-02-14 12:36:07 UTC by dangermaus
#
Network crash not caused by GPU
I retold at least 5 times to my friends how GPU could bring down the network in the computer room, the network of the entire computer science building and wireless access. I then went at home, and while searching for the address of support, I realized that the website of the department was down, too.
On Saturday morning I sent a letter of excuses. A man came complaining that mailserver was down, too. I prefered to stay silent.
However, support people answered very kindly: they think that is very unlikely that a simultaneous write access (seven times) to the same file can bring down the Andrew File System. Outage was caused by a router/switch crash, most likely.
Anyway, 0.842 is available on autoupdate with an option to disable writing in the logfile. Network Mapper supports now a little bit better late answers and people behind the same IP (typical for student houses, e.g.)
F.A.Q were updated with two new questions and two mirrors for the Windows setup are available on the download site now.
2004-02-09 16:08:24 UTC by dangermaus
#
Statistics, FAQ, First big network crash
Many thanks to Seeschloss that gave more detailed statistics to the GPU project, they are available if you add /stats to the project's web address.
Frequently Asked Questions are available as well. "so far, we had no big flaw" should be updated with what follows.
During this week, <permanode>, <iap_mohan> and <albatros> were the network's backbone with <SeeSchloss> and <dangermaus> as supporters.
However, this friday evening, based on the fact that very few people were in the university computer room, I was tweaking GPU to get more accurate 3D graphs. Unfortunately, I was logged in 7 computers on the same account and all GPUs were writing the same logfile at the same time... this brought down the network for the entire computer science department, wireless access and webpage as well. I will write a letter of excuse as soon as they are back online and hope they will not beat me too much. Please do not autoupdate (to version 0.841, in particular) until we fix the problem!
<albatros> and the <mammut-x> cluster did not survive the crash, <iap_mohan> is still available and people should connect to "iap.mine.nu" until we reestablish normal operations.
What's new in 0.840? 0.840 gathers IPs from outside each six hours and reestablish connections to permanent hosts if they were lost. The second release of the Feynman-Kac plugin is provided.
The setup file for 0.835 was missing the Dllbuilding subdirectory, this is fixed in 0.840. Other minor install problems (by running several times Setup.exe and by uninstall) were fixed.
2004-02-06 22:08:26 UTC by dangermaus
#
GPU enters 3D era, <yaymiss> node, globetrotter map
Plenty of news for this week:
First, many thanks to yaymiss (James Dixon) for providing a host running GPU and this in Goodyear, Arizona. Last week, three permanent nodes <yaymiss>, <albatros> and <iap_mohan> were forming the network's backbone. <Seeschloss>, <stackem> and <dangermaus> heavily supported it with long uptimes. Around 30 users were registered by the logfile of <iap_mohan>.
Note that news on network infrastructure and tracebacks moved to the new "Network Status" forum.
Then in 0.835 (and 0.836 with some fixes, available through autoupdate), GPU enters the 3D era for frontends with the Network Mapper and for backends with the OpenGL Teapot plugin, an example to show how OpenGL can be "hacked" to run inside the little Graphics window of GPU.
A 3D Feynman-Kac plugin is on works, but should not be ready till the end of next week. Results for this plugin should be more scientific and less show-biz...
Finally, notice the new "guestbook" of GPU: a worldwide map provided by the Globetrotter project (powered by Xenolth) where everyone running a GPU can "eternize" his/her effort. You are welcome to add yourself there. Address is http://gpu.sf.net/map.php .
Thank you for supporting the project!
2004-02-01 16:20:31 UTC by dangermaus
#
GPU 0.830: See through the eyes of <albatros>!
GPU is a Gnutella client for sharing files and CPU-time.
0.830 is a milestone release: it includes a frontend, called GPU Network Mapper to visualize in three dimensions how the GPU network looks like. It is enough to be connected to one permanent host (for example <albatros> or <iap_mohan>) to get information about load of all other nodes (color green: no load and red: busy) and how they are interconnected. The Network Mapper works like the other frontends: it sends a "gpustatus" request, all connected GPUs answer with information about load and their connections. This information is a graph, displayed in 3D by the Frontend thanks to the OpenGL routines made by Tom Nuydens (tom@delphi3d.net) and Chris Rorden (chris.rorden@nottingham.ac.uk). The flare effect is also by Tom.
This weekend, you can expect to view on the GPU network the following nodes: <Seeschloss[2]>, <albatros>, <dangermaus>, and <iap_mohan>.
GPU with lower version numbers than 0.827 are plotted as wire spheres, the Network Mapper knows their existence because newer clients reported them into their connection list.
IP information is corrected in 0.830 (sometimes wrong in 0.829 and 0.828, if GPU was behind a router): the page of siwi with the IP script is visited to retrieve it.
If you check 'Periodic update', the Network mapper will rebuild the graph about every 40 minutes. Please consider that higher update rates might hurt the Gnutella network.