Skip to content
Permalink
Branch: master
Commits on Mar 3, 2019
  1. analyze response times by operation type

    bengland2 committed Mar 3, 2019
    allows you to spot cause of trends in response time during a run
  2. modify smallfile/smallfile_rsptimes_stats.py to parse fs-drift *rspti…

    bengland2 committed Mar 3, 2019
    …mes.csv
    
    filename format is different
Commits on Mar 1, 2019
  1. Merge branch 'master' of https://github.com/parallel-fs-utils/fs-drift

    bengland2 committed Mar 1, 2019
    pulling README.md from github
  2. gaussian distribution almost works

    bengland2 committed Mar 1, 2019
    no more exceptions being thrown
    saving context in per-thread FSOPCtx member variables
    experimenting with spreading out gaussian distribution
    velocities and centers
  3. typo

    bengland2 committed Mar 1, 2019
  4. pass user-specified random distribution to FSOPCtx correctly

    bengland2 committed Mar 1, 2019
    until now it was always doing uniform dist.
  5. let FSOPCtx class know host and thread id

    bengland2 committed Mar 1, 2019
    This will enable it to save per-thread state later
    so that different threads can be operating on different
    gaussian distributions with different velocities
Commits on Feb 26, 2019
  1. clarify meaning of --host-set

    bengland2 committed Feb 26, 2019
    add that hosts can be placed in a file and the pathname can be supplied in place of the list of hosts
    add that password-less ssh access is a requirement
Commits on Feb 20, 2019
  1. make it easier for user to see input parameters

    bengland2 committed Feb 20, 2019
    print them after parsing them
    indent first row, column-align values also
Commits on Jan 29, 2019
  1. add option for starting fs-drift-remote.py by daemon not ssh

    bengland2 committed Jan 29, 2019
    fix timeout message
    don't use 1/10 sec delay between threads if by daemon
    let user alter python program and location of fs-drift-remote.py
  2. add option to launch fs-drift-remote.py by daemon not ssh

    bengland2 committed Jan 29, 2019
    for security reasons, hardcode the path to fs-drift-remote.py
    but let user override that with environment variables
Commits on Jan 28, 2019
Commits on Jan 22, 2019
  1. correct IOPS computation

    bengland2 committed Jan 22, 2019
    had been using files instead of ios
  2. make results have same fields for cluster, host or thread

    bengland2 committed Jan 22, 2019
    makes it easier to parse results
Commits on Jan 21, 2019
  1. check for negative file size returned by os.fstat()

    bengland2 committed Jan 21, 2019
    encountered this problem on a Gluster fs, trying to catch it early
  2. output file descriptor number along with exception if there is one

    bengland2 committed Jan 21, 2019
    Just to ensure that fd is sane
Commits on Jan 14, 2019
  1. no data parser needed

    bengland2 committed Jan 14, 2019
    from now on data will be output in JSON format
    parsing becomes json.loads()
Commits on Jan 11, 2019
  1. add filesystem fullness check and inhibit writes when it's full

    bengland2 committed Jan 11, 2019
    why: this benchmark depends on the filesystem to return results
    so it won't work if the filesystem is so full that it can't write files
  2. handle case where new files/links cannot be created

    bengland2 committed Jan 11, 2019
    when filesystem is full, this can happen even on a rename
  3. random number must be between 1 and max_files

    bengland2 committed Jan 11, 2019
    we divide randomness into directory path component
    and filename component
Commits on Jan 10, 2019
  1. after you read in verbosity, wait poll_rate sec before unlinking

    bengland2 committed Jan 10, 2019
    Otherwise most threads will not see it
    also catch exception if you can't unlink it
    because another thread got there first
  2. dont use same part of random int for filename and for dirpath

    bengland2 committed Jan 10, 2019
    Otherwise you don't generate full set of max_files pathnames
  3. get rid of randstate member and use of it

    bengland2 committed Jan 10, 2019
    may be synchronizing threads' random number generators
  4. do not use same random seed every time

    bengland2 committed Jan 10, 2019
    this was something smallfile needed to do, but
    using same random seed means we never create new files,
    get EEXIST, we really want sequence to be different each time
Older
You can’t perform that action at this time.