Updated NeuronRain FAQ - http://neuronrain-documentation.readthedocs.…

…io/en/latest/

index.rst

@@ -290,7 +290,7 @@ assigning time slices dynamically according to priority classifier. It is unknow
 	process1 = <pid1, executabletype1, executablename1, size1, cpu_usage1, memory_usage1, disk_usage1>
 	process2 = <pid2, executabletype2, executablename2, size2, cpu_usage2, memory_usage2, disk_usage2>
 	distance(process1, process2) = euclidean_distance(process1, process2)
-Psutils Dictionary Encoding of a process and Diff edit distance between two processes has been implemented in https://github.com/shrinivaasanka/asfer-github-code/blob/master/python-src/software_analytics/DeepLearning_SchedulerAnalytics.py. Socket Streaming Analytics Server of Process Statistics has been implemented in https://github.com/shrinivaasanka/asfer-github-code/blob/master/python-src/software_analytics/ which analyzes stream of process JSON dictionary data and can write out analytics variables read/exported by VIRGO Linux kernel_analytics driver which in turn are readable by OS Scheduler (requires Scheduler rewrite). This is an ideal solution for self-healing OS kernels which learn from process performance in userspace and change scheduler behaviour dynamically. Analytics variables can be directly written to /etc/sysctl.conf or by sysctl  if alternative to /etc/kernel_analytics.conf is preferred. Sysctl has config variables for VM Paging, Scheduler, Networking among others which are read by kernel live (kernel.sched.*)- if kernel provides comprehensive sysctl variables for Scheduler policy, it removes necessity for Scheduler rewrite. Presently sysctl apparently exports Round Robin timeslicing only. Similarly, USBmd 32 and 64 bit drivers for Wireless LAN traffic analytics can directly write learnt analytic variables to /etc/sysctl.conf (https://www.kernel.org/doc/Documentation/networking/ip-sysctl.txt lists various TCP tuning configs - net.* - e.g Corking consecutive frequent read/writes into one read/write, SYNACK retries, fastopen, receive buffer size)
+Psutils Dictionary Encoding of a process and Diff edit distance between two processes has been implemented in https://github.com/shrinivaasanka/asfer-github-code/blob/master/python-src/software_analytics/DeepLearning_SchedulerAnalytics.py. Socket Streaming Analytics Server of Process Statistics has been implemented in https://github.com/shrinivaasanka/asfer-github-code/blob/master/python-src/software_analytics/ which analyzes stream of process JSON dictionary data and can write out analytics variables read/exported by VIRGO Linux kernel_analytics driver which in turn are readable by OS Scheduler (requires Scheduler rewrite). This is an ideal solution for self-healing OS kernels which learn from process performance in userspace and change scheduler behaviour dynamically. Analytics variables can be directly written to /etc/sysctl.conf or by sysctl  if alternative to /etc/kernel_analytics.conf is preferred. Sysctl has config variables for VM Paging, Scheduler, Networking among others which are read by kernel live (kernel.sched.*)- if kernel provides comprehensive sysctl variables for Scheduler policy, it removes necessity for Scheduler rewrite. Presently sysctl apparently exports Round Robin timeslicing only. Similarly, USBmd 32 and 64 bit drivers for Wireless LAN traffic analytics can directly write learnt analytic variables to /etc/sysctl.conf (https://www.kernel.org/doc/Documentation/networking/ip-sysctl.txt lists various TCP tuning configs - net.* - e.g Corking consecutive frequent read/writes into one read/write, SYNACK retries, fastopen, receive buffer size). GRAFIT Course Material in https://github.com/shrinivaasanka/Grafit/blob/c8290348b916e5b35044c3834f56a825b4db23e4/course_material/NeuronRain/AdvancedComputerScienceAndMachineLearning/AdvancedComputerScienceAndMachineLearning.txt describe an example performance analytics of OS Scheduler (clockticks-to-processes) hypothetically implemented as LSH. Simulating this in a linux kernel may not be straightforward. But there are performance tools like perf (http://www.brendangregg.com/perf.html#SchedulerAnalysis) and SAR which can create a streaming text dataset of kernel scheduler runqueue after some script processing and write kernel.sched.* variables based on analytics.
 
 **Who can deploy NeuronRain?**