[netperf-talk] Additional "histogram" statistics courtesy of Google

[Rick Jones]

Folks -

The folks at Google were kind enough to extend the histogram code to enable 
optionally keeping some additional statistics about the times kept in the 
historgram.  I have taken that code and merged it into the omni tests.  The new 
output specifiers are MIN_LATENCY, MAX_LATENCY, P50_LATENCY, P90_LATENCY, 
P99_LATENCY, MEAN_LATENCY, and STDDEV_LATENCY.  You can either wade through the 
kitchen-sink omni output, or you can create an output specifier file thusly:

raj at tardy:~/netperf2_trunk$ cat foo
ELAPSED_TIME,THROUGHPUT,THROUGHPUT_UNITS,RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY

and use it like this for an RR test for example:

raj at tardy:~/netperf2_trunk$ src/netperf -j -t omni  -l 1 -- -r 1 -o foo
OMNI TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to localhost (127.0.0.1) port 0 
AF_INET
Elapsed Time (sec),Throughput,Throughput Units,Round Trip Latency 
usec/tran,Minimum Latency Microseconds,Maximum Latency Microseconds,50th 
Percentile Latency Microseconds,90th Percentile Latency Microseconds,99th 
Percentile Latency Microseconds,Mean Latency Microseconds,Stddev Latency 
Microseconds
1.00,27404.75,Trans/s,36.490,26,110,33,40,41,35.00,3.68

If you have previously done a ./configure --enable_omni ... when builting 
netperf.   The global -j option tells netperf to track the statistics.

A related change is that now much of the histogram code is compiled-in 
unconditionally. This is in support of the -j option. However, to get a classic 
histogram, one must still add --enable-histogram to the ./configure and set the 
verbosity to 2 or higher.  I'm not completely adverse to always doing histograms 
at verbosity 2 or higher if the user base wishes it.  Until them I'm simply 
trying to minimize surprise :)

The bits are in the top of trunk in the repository:

http://www.netperf.org/svn/netperf2/trunk

happy benchmarking,

rick jones