nfsometer (1) - Linux Manuals

nfsometer: NFS performance measurement tool


nfsometer - NFS performance measurement tool


nfsometer [options] [mode] [[<server:path>] [workloads...]]


nfsometer is a performance measurement framework for running workloads and reporting results across NFS protocol versions, NFS options and Linux NFS client implementations


Basic usage (no mode specified):

 nfsometer <server:path> [workloads...]

  This will fetch needed files, run traces, and generate reports,
  same as running the the 'fetch', 'trace' and 'report' stages.

Advanced usage (specify modes):

 nfsometer list

 List the contents of the results directory.

 nfsometer workloads

 List available and unavailable workloads.

 nfsometer notes

 Edit the notes file of the results directory. These notes will
 be displayed in report headers.

 nfsometer loadgen <server:path> <workload>

 Run in loadgen mode: don't record any stats, just loop over
 <workload> against <server:path>.  Only one -o option is allowed.
 Use the -n option to run multuple instances of the loadgen workload.
 When running more than one instance, the intial start times are

 nfsometer fetch [workloads...]

 Fetch all needed files for the specified workload(s).  If no
 workloads are specified, all workloads are fetched.
 Fetched files are only downloaded once and are cached for
 future runs.

 nfsometer trace <server:path> [workloads...]

 Run traces against <server:path>.  The traces run will be:
 (options always options tags) (workloads) (num runs)
 This will only run traces that don't already exist in the results

 nfsometer report

 Generate all reports available from the results directory.

 nfsometer example

 Show examples from man page


-r <dir>, --resultdir=<dir>
The directory used to save results. default: '/home/dros/nfsometer_results'

-o <mount.nfs options>, --options=<mount.nfs options>
Mount options to iterate through. This option may be used multiple times. Each mount option must have a version specified.

-a <mount.nfs options>, --always-options=<mount.nfs options>
Options added to every trace. This option may be used multiple times.

-t <tags>, --tag=<tags>
Tag all new traces with 'tags'. This option may be used multiple times.

-n <num runs>, --num-runs=<num runs>
Number of runs for each trace of <options> X <tags> X <workloads> default: 1

Generate graphs inline while generating reports. Useful for debugging graphing issues.

Randomize the order of traces

-h, --help
Show the help message


Example 1: See what workloads are available

  nfsometer workloads

  This command lists available workloads and will tell you why
  workloads are unavailable (if any exist).

Example 2: Compare cthon, averaged over 3 runs,
     across nfs protocol versions

nfsometer -n 3 server:/export cthon

  This example uses the default for -o: "-o v3 -o v4 -o v4.1".
  To see the results, open results/index.html in a web browser.

Example 3: Compare cthon, averaged over 3 runs,
     between v3 and v4.0 only

  nfsometer -n 3 -o v3 -o v4 server:/export cthon

  This example specifies v3 and v4 only.
  To see the results, open results/index.html in a web browser.

Example 4: Compare two kernels running iozone workload, averaged
     over 2 runs, across all nfs protocol versions

  nfsometer can compare two (or more) kernel versions, but
  has no way of building, installing or booting new kernels.
  It's up to the user to install new kernels.
  In order for these kernels to be differentiated, 'uname -a'
  must be different.

1) boot into kernel #1

2) nfsometer -n 2 server:/export iozone

3) boot into kernel #2

4) nfsometer -n 2 server:/export iozone

5) open results/index.html in a web browser

  To see the results, open results/index.html in a web browser.

Example 5: Using tags

  Tags (the -t option) can be used to mark nfsometer runs as
  occurring with some configuration not captured by mount options
  or detectable tags, such as different sysctl settings (client side),
  different server side options, or different network conditions.

  1) set server value foo to 2.3

  2) nfsometer -o v4 -o v4.1 -t foo=2.3

  3) set server value foo to 10

  4) nfsometer -o v4 -o v4.1 -t foo=10

  What is passed to -t is entirely up to the user - it will not be
  interpreted or checked by nfsometer at all, so be careful!

  To see the results, open results/index.html in a web browser.

Example 6: Always options

  The -o flag specifies distinct option sets to run, but sometimes
  there are options that should be present in each.  Instead of
  writing each one out, you can use the -a option:

  nfsometer -o v3 -o v4 -a sec=krb5 server:/export iozone

  this is equivalent to:

  nfsometer -o v3,sec=krb5 -o v4,sec=krb5 server:/export iozone

Example 7: Using the "custom" workload

  A main use case of nfsometer is the "custom" workload - it allows
  the user to specify the command that nfsometer is to run.

  NOTE: the command's cwd (current working directory) is the runroot
  created on the server.

  export NFSOMETER_CMD="echo foo bar"
  export NFSOMETER_NAME="echo"
  export NFSOMETER_DESC="Writes 4 bytes to a file"
  nfsometer server:/export custom

  This will run 3 traces (v3, v4, v4.1) against server:/export of
  the command: echo foo bar.

Example 8: Using the loadgen mode

 Loadgen runs several instances of a workload without capturing
 traces. The idea is that you use several clients to generate
 load, then another client to measure performance of a loaded
 server. The "real" run of nfsometer (not loadgen) should mark
 the traces using the -t option.

 1) On client A, run the cthon workload to get a baseline of
 a server without any load.

nfsometer trace server:/export cthon

 2) When that's done, start loadgen on client B:

nfsometer -n 10 loadgen server:/export dd_100m_1k

 This runs 10 instances of dd_100m_1k workload on server:/export.
 It can take several minutes to start in an attempt to stagger
 all the workload instances.

 3) once all instances are started, run the "real" nfsometer
 trace on client A.  Use the -t option to mark the traces
 as having run under load conditions:

nfsometer -t "10_dd" trace server:/export cthon

 4) Explain how the tests were set up in the result notes.
 This should be run on client A (which has the traces:

nfsometer notes

 5) Now generate the reports:

nfsometer report

Example 8: Long running nfsometer trace

  The script currently runs in the foreground.  As
  such, it will be killed if the tty gets a hangup or the connection
  to the client is closed.

  For the time being, nfsometer should be run in a screen
  session, or run with nohup and the output redirected to a file.

1) screen -RD
2) nfsometer -n 2 server:/export iozone
3) close terminal window (or ^A^D)
4) reattach later with screen -RD
5) once is done, results will be in results/index.html


No known bugs.


Weston Andros Adamson (dros [at]


mountstats, nfsstats