1man(1) General Commands Manual man(1)
2
6 nfsometer - NFS performance measurement tool
7
9 nfsometer [options] [mode] [[<server:path>] [workloads...]]
10
12 nfsometer is a performance measurement framework for running workloads
13 and reporting results across NFS protocol versions, NFS options and
14 Linux NFS client implementations
15
17 Basic usage (no mode specified):
18 nfsometer <server:path> [workloads...]
20 This will fetch needed files, run traces, and generate reports,
22 same as running the the 'fetch', 'trace' and 'report' stages.
23 Advanced usage (specify modes):
25 nfsometer list
27 List the contents of the results directory.
29 nfsometer workloads
31 List available and unavailable workloads.
33 nfsometer notes
35 Edit the notes file of the results directory. These notes will
37 be displayed in report headers.
38 nfsometer loadgen <server:path> <workload>
40 Run in loadgen mode: don't record any stats, just loop over
42 <workload> against <server:path>. Only one -o option is allowed.
43 Use the -n option to run multuple instances of the loadgen work‐
44 load.
45 When running more than one instance, the intial start times are
46 staggered.
47 nfsometer fetch [workloads...]
49 Fetch all needed files for the specified workload(s). If no
51 workloads are specified, all workloads are fetched.
52 Fetched files are only downloaded once and are cached for
53 future runs.
54 nfsometer trace <server:path> [workloads...]
56 Run traces against <server:path>. The traces run will be:
58 (options + always options + tags) X (workloads) X (num runs)
59 This will only run traces that don't already exist in the results
60 directory.
61 nfsometer report
63 Generate all reports available from the results directory.
65 nfsometer example
67 Show examples from man page
69
71 -r <dir>, --resultdir=<dir>
72 The directory used to save results. default: '/home/dros/nfsome‐
73 ter_results'
74 -o <mount.nfs options>, --options=<mount.nfs options>
78 Mount options to iterate through. This option may be used multi‐
79 ple times. Each mount option must have a version specified.
80 -a <mount.nfs options>, --always-options=<mount.nfs options>
84 Options added to every trace. This option may be used multiple
85 times.
86 -t <tags>, --tag=<tags>
90 Tag all new traces with 'tags'. This option may be used multiple
91 times.
92 -n <num runs>, --num-runs=<num runs>
96 Number of runs for each trace of <options> X <tags> X <workloads>
97 default: 1
98 --serial-graphs
102 Generate graphs inline while generating reports. Useful for
103 debugging graphing issues.
104 --rand
108 Randomize the order of traces
109 -h, --help
113 Show the help message
114
116 Example 1: See what workloads are available
117 $ nfsometer workloads
119 This command lists available workloads and will tell you why
121 workloads are unavailable (if any exist).
122 Example 2: Compare cthon, averaged over 3 runs,
125 across nfs protocol versions
126 nfsometer -n 3 server:/export cthon
128 This example uses the default for -o: "-o v3 -o v4 -o v4.1".
130 To see the results, open results/index.html in a web browser.
131 Example 3: Compare cthon, averaged over 3 runs,
134 between v3 and v4.0 only
135 nfsometer -n 3 -o v3 -o v4 server:/export cthon
137 This example specifies v3 and v4 only.
139 To see the results, open results/index.html in a web browser.
140 Example 4: Compare two kernels running iozone workload, averaged
143 over 2 runs, across all nfs protocol versions
144 nfsometer can compare two (or more) kernel versions, but
146 has no way of building, installing or booting new kernels.
147 It's up to the user to install new kernels.
148 In order for these kernels to be differentiated, 'uname -a'
149 must be different.
150 1) boot into kernel #1
152 2) nfsometer -n 2 server:/export iozone
154 3) boot into kernel #2
156 4) nfsometer -n 2 server:/export iozone
158 5) open results/index.html in a web browser
160 To see the results, open results/index.html in a web browser.
162 Example 5: Using tags
165 Tags (the -t option) can be used to mark nfsometer runs as
167 occurring with some configuration not captured by mount options
168 or detectable tags, such as different sysctl settings (client side),
169 different server side options, or different network conditions.
170 1) set server value foo to 2.3
172 2) nfsometer -o v4 -o v4.1 -t foo=2.3
174 3) set server value foo to 10
176 4) nfsometer -o v4 -o v4.1 -t foo=10
178 What is passed to -t is entirely up to the user - it will not be
180 interpreted or checked by nfsometer at all, so be careful!
181 To see the results, open results/index.html in a web browser.
183 Example 6: Always options
186 The -o flag specifies distinct option sets to run, but sometimes
188 there are options that should be present in each. Instead of
189 writing each one out, you can use the -a option:
190 nfsometer -o v3 -o v4 -a sec=krb5 server:/export iozone
192 this is equivalent to:
194 nfsometer -o v3,sec=krb5 -o v4,sec=krb5 server:/export iozone
196 Example 7: Using the "custom" workload
199 A main use case of nfsometer is the "custom" workload - it allows
201 the user to specify the command that nfsometer is to run.
202 NOTE: the command's cwd (current working directory) is the runroot
204 created on the server.
205 export NFSOMETER_CMD="echo foo > bar"
207 export NFSOMETER_NAME="echo"
208 export NFSOMETER_DESC="Writes 4 bytes to a file"
209 nfsometer server:/export custom
210 This will run 3 traces (v3, v4, v4.1) against server:/export of
212 the command: echo foo > bar.
213 Example 8: Using the loadgen mode
216 Loadgen runs several instances of a workload without capturing
218 traces. The idea is that you use several clients to generate
219 load, then another client to measure performance of a loaded
220 server. The "real" run of nfsometer (not loadgen) should mark
221 the traces using the -t option.
222 1) On client A, run the cthon workload to get a baseline of
224 a server without any load.
225 nfsometer trace server:/export cthon
227 2) When that's done, start loadgen on client B:
229 nfsometer -n 10 loadgen server:/export dd_100m_1k
231 This runs 10 instances of dd_100m_1k workload on server:/export.
233 It can take several minutes to start in an attempt to stagger
234 all the workload instances.
235 3) once all instances are started, run the "real" nfsometer
237 trace on client A. Use the -t option to mark the traces
238 as having run under load conditions:
239 nfsometer -t "10_dd" trace server:/export cthon
241 4) Explain how the tests were set up in the result notes.
243 This should be run on client A (which has the traces:
244 nfsometer notes
246 5) Now generate the reports:
248 nfsometer report
250 Example 8: Long running nfsometer trace
252 The nfsometer.py script currently runs in the foreground. As
254 such, it will be killed if the tty gets a hangup or the connection
255 to the client is closed.
256 For the time being, nfsometer should be run in a screen
258 session, or run with nohup and the output redirected to a file.
259 1) screen -RD
261 2) nfsometer -n 2 server:/export iozone
262 3) close terminal window (or ^A^D)
263 ...
264 4) reattach later with screen -RD
265 5) once nfsometer.py is done, results will be in results/index.html
266
268 mountstats, nfsstats
269
271 No known bugs.
272
274 Weston Andros Adamson (dros@netapp.com)
275nfsometer 1.7 man(1)