On Thu, 22 Mar 2001, Rajesh Agrawal wrote:
> I ran polygraph on freebsd for cache engine and generated the
> report. I was unable to find number of persistent connections
> mentioned anywhere. Is there any calculation i need to get such
> parameter ?
See the "use" row in the "Connection Length" table (below the "Stream
Totals" table). Connection use tells you the number of HTTP
transactions per TCP connection (minimum, mean, and maximum). If no
pconns were in use, the maximum will be at most 1. PolyMix-3 [offered]
maximum is 64.
For example, here are the readings of "mean transactions per connection"
taken during the last cache-off: http://www.measurement-factory.com/results/
Aratech-2000 1.00
CinTel-iCache 14.86
Compaq-C2500 14.88
Compaq-b17 14.86
Dell-100 14.87
Dell-200x4 1.41
Swell-1450 14.84
F5-EDGE-FX 14.85
IBM-220-1 14.87
IBM-330 14.87
IBM-220-2 14.88
IBM-230 14.87
iMimic-Alpha 14.86
iMimic-2600 14.87
iMimic-1300 1.41
iMimic-2400 14.86
Lucent-100z 2.32
Lucent-100 2.32
Lucent-50 2.30
Microbits-C 14.89
Microbits-P 14.88
Microsoft-1 14.86
Microsoft-2 14.86
NetApp-C6100 14.80
NetApp-C1105 14.73
Stratacache-F 14.86
Stratacache-E 14.87
Stratacache-D 14.87
Squid-2.4.D4 14.84
Note that the mean of a zipf(64) distribution is ~14.85 (median is
about 7.5). PolyMix-3 robots are configured to send no more than
zipf(64) requests per connections. The number of transactions per
connection will be much lower for transparent configurations, of
course.
If you do need to know the number of persistent connections (i.e., the
number of connections that were closed after more than one HTTP
transaction), you probably should analyze the conn_close.*.use.hist
histograms in the /tmp/polyrep/*/clt.All.lx file. The clt.All.lx file
is created when you run label_results. It is an ASCII file that is
relatively easy to parse for a human, especially after reading
http://polygraph.ircache.net/doc/lxoutput.html
Note that such a counter would depend on request rate and duration of
the test. You may be better off with computing a percentage of pconns
instead. The computation is relatively easy to do with a Perl or awk
script; I bet you have a script kiddie in your team :).
HTH,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:18 MDT