VBENCH is a fork of MOSBENCH but allows running benchmarks inside VM as well.

• Currently, we have included the following benchmarks for evaluation:
  • Gmake: Fork intensive, but spends most of the time in the userspace.
  • Exim: Crazily stresses mm, fs, and process sub-systems.
  • Psearchy: Stresses the interrupt service, used to stress the writer side of the mmap_rwsem.
  • Metis: Stresses the reader side of the mmap_sem.
  • Hist: Stresses the reader side of the mmap_sem.

• First compile the relevant benchmarks as follows:

  • Exim: make all-exim
  • Psearchy: make all-psearchy
  • Metis: make all-metis
  • Hist: make all-hist

• Configure the following variables in the config.py:

    ROOT = "~/vbench" # Provide the relative path (this is also used in the VM)
    CPU_CNTS = 160 # Total number of cores (cores * sockets)
    CORES_PER_SOCKET = 10
    CPU_STEP = 10 # Numbers of cores to increase until CPU_CNTS in each run

• Modify the hostname in the hosts.py. You can set it to 127.0.0.1 and disable sanity check in mparts/host.py

• Create multiple tmpfs partitions with the following command:

  $ sudo ./mkmounts tmpfs-separate # sudo is mandatory

• After this, you can use config.py to execute benchmarks.

 $ ./config.py -h

Options:
  -h, --help            show this help message and exit
  -d, --dryrun          dry run (default: sanityRun = False)
  -t TRIALS, --trials=TRIALS
                        trials (default: TRY = 1)
  -c CORES, --cores=CORES
                        number of cores for dryrun, default = 80
  -n, --profile-nos     get numbers using perf stat
  -r, --perf-record     record all functions
  --kvm-record          perf record for guest, only works with -r option
  --guest-only-record   only records the guest, no host side recording
  -p, --pin             pin all the tasks in qemu process
  -l, --lockstats       get lock stats
  -m, --multivm         run two VMs simultaneously
  --kvm-stat            get the kvm stats if possible (only possible for PVM)
  --trace-kvm           get the results using trace-cmd

• To run a benchmark, here exim, across all cores, execute the following:

 $ ./config.py exim # runs the exim benchmark from core 1 to core 160.

• After running the benchmark, results are displayed on the terminal and are stored in the results folder, which can be used for plotting by using the ./graph script.