Bob, I'm a little behind in getting to this, but I will. Thanks for all the data.
Please also watch for a post on TCP cases in general. Cheers, Phil Bob Palowoda wrote:
Phil Harman wrote:Bob Palowoda wrote:There is probably nothing wrong with connection it just toke123sec on a 3.2ghz machine. I just didn't wait long enough.Phew! 123sec is a long time. Maybe your machine was busy doing other stuff in the background? Before 0.3.0 we used to run each test case for 5 seconds, irrespective of the number of samples obtained. For 0.3.0 Bart reworked the iteration control to gather a statistically significant and stable set of samples. Some test cases are much more sensitive to background activity than others. Or it may be that we need to tune the new code a little more.Well something funny is going on with connection. And it dosn't seem to affect the results but the runtime. I ran the test on a Linux Suse 9.3 box and it completed in ~4sec. So I decided to run it on another Solaris nv_13 box (orig ran it on a nv_19 box) to see if made a difference. Nope still about 125secs of wall clock time but the results where normal. So you would think if I did a differential timing with truss I'd see some big delay. Now here is where it gets strange. I get nothing that shows the big delays from truss but I do get these outputs with no time associatedwith a pollsys call. Look at the bottom of the file: http://www.fiver.net/misc/connection.out script. Notice the1114: 0.0001 connect(457, 0x08068518, 16, SOV_DEFAULT) Err#150 EINPROGRESS1114: pollsys(0x080473C8, 1, 0x00000000, 0x00000000) (sleeping...) ^^^^^^^^^^^^^^^ [EMAIL PROTECTED]:/usr/local/src/bench/libMicro/libMicro-0.3.0> exitThat is where I get these huge 1 to 3 second delays. They are not recorded in the output of the script file. Screen shoot of the test:http://www.fiver.net/misc/snapshot13.pngSo something is causing a runtime delay maybe my X envionment but I don't really have anything running on the machine that can cause such huge delays. It's only me on the system. I have to try this with no X and at the console to see if it is related. By the way how can truss not record a system call timing?This is the results of connection on the machine. Nothing abnormal other than a few calls way out therebut the runtime is 128sec. ---Bob[EMAIL PROTECTED]:/usr/local/src/bench/libMicro/libMicro-0.3.0> bin/connection -E -C 200 -L -S -W -N connection -B 256 Running: connection# bin/../bin-i86pc/connection -E -C 200 -L -S -W -N connection -B 256 prc thr usecs/call samples errors cnt/samp connection 1 1 57.24556 198 0 256 # # STATISTICS usecs/call (raw) usecs/call (outliers removed)# min 41.55809 41.55809 # max 26573.72019 13450.21279 # mean 2437.95974 1952.25946 # median 57.69142 57.24556 # stddev 5699.50421 4601.10870 # standard error 401.01572 326.98656 # 99% confidence level 932.76256 760.57073 # skew 2.37616 2.04170 # kurtosis 5.09603 2.18166 # time correlation 19.84323 16.45937 # # elasped time 128.25268 # number of samples 198 # number of outliers 4 # getnsecs overhead 375 # # DISTRIBUTION# counts usecs/call means # 170 0.00000 |******************************** 89.90810 # 0 500.00000 | - # 0 1000.00000 | - # 0 1500.00000 | - # 0 2000.00000 | - # 0 2500.00000 | - # 0 3000.00000 | - # 0 3500.00000 | - # 0 4000.00000 | - # 0 4500.00000 | - # 0 5000.00000 | - # 0 5500.00000 | - # 0 6000.00000 | - # 0 6500.00000 | - # 0 7000.00000 | - # 0 7500.00000 | - # 0 8000.00000 | - # 0 8500.00000 | - # 0 9000.00000 | - # 0 9500.00000 | - # 0 10000.00000 | - # 0 10500.00000 | - # 0 11000.00000 | - # 0 11500.00000 | - # 0 12000.00000 | - # 0 12500.00000 | - # 18 13000.00000 |*** 13235.64168## 10 > 95% |* 13302.14447# # mean of 95% 1348.54217 # 95th %ile 13250.48379 for 128.25649 seconds ^^^^^^^^^^^^^ [EMAIL PROTECTED]:/usr/local/src/bench/libMicro/libMicro-0.3.0>
smime.p7s
Description: S/MIME Cryptographic Signature
_______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
