Skip to content

snalli/whisper

 
 

Repository files navigation

WHISPER 1.0

WHISPER, or Wisconsin-HPL Suite for Persistence is a comprehensive benchmark suite for emerging persistent memory technologies. For more details :

An Analysis of Persistent Memory Use with WHISPER. ASPLOS'17.

Sanketh Nalli, Swapnil Haria, Mike Swift, Mark Hill, Haris Volos, Kim Keeton.

research.cs.wisc.edu/multifacet/whisper/

IMPORTANT

We are currently testing WHISPER in various environments. PLEASE USE AT YOUR OWN RISK. Report issues and suggestions to Swapnil (swapnilh at cs dot wisc edu) or Sanketh (sankey at cs dot wisc dot edu).

Eliza is a tool to process traces generated by WHISPER applications. You may use it or develop your own for your analysis.

To download:

   	$ git clone --recursive https://github.com/swapnilh/whisper.git
	$ cd whisper

OR

   	$ git clone https://github.com/swapnilh/whisper.git
	$ cd whisper
	$ ./script.py -d

To build:

	$ cd whisper
	$ ./script.py -b \
		-w	{ycsb,tpcc,echo,redis,ctree,hashmap,memcached,vacation,nfs,exim,sql,all}

To run (tracing is optional, but need to be root for it):

	$ cd whisper
	$ ./script.py -r [-t] -z {small|med|large} \
		-w  {ycsb,tpcc,echo,redis,ctree,hashmap,memcached,vacation,nfs,exim,sql,all}

To clean:

	$ cd whisper
	$ ./script.py -c \
		-w {ycsb,tpcc,echo,redis,ctree,hashmap,memcached,vacation,nfs,exim,sql,all}

To update:

	$ cd whisper
	$ ./script.py -u
	git submodule update --remote
	git submodule update status

For help:

	$ cd whisper
	$ ./script.py -h

usage: script.py [-h] [-d] [-b] [-r]
              [-w {ycsb,tpcc,echo,redis,ctree,hashmap,memcached,vacation,nfs,exim,sql,all}]
              [-z {small,med,large}] [-t] [-c] [-s] [-u] [-p]

Buildi, Clean, Run, Update WHISPER

optional arguments:
  -h, --help            show this help message and exit
  -d                    Download workload
  -b                    Build workload
  -r                    Run workload
  -w {ycsb,tpcc,echo,redis,ctree,hashmap,memcached,vacation,nfs,exim,sql,all}
                        Workload
  -z {small,med,large}  Set workload size
  -t                    Enable tracing. Need to be root.
  -c                    Clean workload
  -s                    Clean workload and dependencies
  -u                    Update benchmark
  -p                    More help for a workload

For more help:

Example,
	$ ./script.py -w vacation -p 

Usage: ./build/bench/stamp-kozy/vacation/vacation [options]

Options:                                             (defaults)
    c <UINT>   Number of [c]lients                   (1)
    n <UINT>   [n]umber of user queries/transaction  (10)
    q <UINT>   Percentage of relations [q]ueried     (90)
    t <UINT>   Number of [t]ransactions              (131072)
    r <UINT>   Number of [r]ows		       (5767168)
    u <UINT>   Percentage of [u]ser transactions     (80)
    e <UINT>   Enable trac[e] collection             (0)

FAQs :

1. For Nstore-{YCSB, TPCC} and Echo I get a runtime error that says "no free memory of size 128 available".

What do I do ?

Nstore and Echo create a file in /dev/shm (zfile for nstore and efile for echo} which act as persistent memory pools. So does every application in WHISPER. The default size is 1GB. When space is low, the above error is thrown. We will make this a run time parameter in the future. For now, simply increase it by altering the PSEGMENT_RESERVED_REGION_SIZE in Nstore and Echo. You may find this variable by using grep, cscope or any indexing tool for browsing source code. Its location varies with application. Delete the old pool, recompile and re-run. Suggested sizes are as shown. Please pay attention to the format of defining the size i.e. XXX * 1024 * 1024 * 1024. This is in bytes.

#define PSEGMENT_RESERVED_REGION_SIZE (2UL * 1024 * 1024 * 1024) // 2GB
#define PSEGMENT_RESERVED_REGION_SIZE (4UL * 1024 * 1024 * 1024) // 4GB
#define PSEGMENT_RESERVED_REGION_SIZE (8UL * 1024 * 1024 * 1024) // 8GB

You may also reduce the size as follows.

#define PSEGMENT_RESERVED_REGION_SIZE (512 * 1024 * 1024) // 512MB
#define PSEGMENT_RESERVED_REGION_SIZE (256 * 1024 * 1024) // 256MB
#define PSEGMENT_RESERVED_REGION_SIZE (128 * 1024 * 1024) // 128MB

About

WHISPER is a comprehensive benchmark suite for emerging persistent memory technologies.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%