Robin, I have attached the log file for the test fail below, from the netperf log dumps of the 2 systems, it seems that 20 sessions from both sides were running fine ,so i am not sure why the MAXQ concluded that only 37 out of 40 sessions passed.Could you pls tell me what exactly is the error here?
Thanks Somnath --- On Thu, 12/24/09, Somnath kotur <so...@yahoo.com> wrote: From: Somnath kotur <so...@yahoo.com> Subject: Re: [networking-discuss] NICDRV test 15 - max sessions test To: "tian robin luo" <robin....@sun.com> Cc: networking-discuss@opensolaris.org, driver-disc...@opensolaris.org Date: Thursday, December 24, 2009, 5:07 AM Robin , I just was rerunning the entire test suite again on the same system that passed the netstress ,only this time i am seeing similiar errors in test01 itself, as in for TCP traffic with 65000/1460 , only 36/38 sessions manage to connect while it works just fine for the rest of the suite ....? Can this be attributed to the same cause as below or is this something i need to worry about? Thanks Somnath --- On Tue, 12/22/09, tian robin luo <robin....@sun.com> wrote: From: tian robin luo <robin....@sun.com> Subject: Re: NICDRV test 15 - max sessions test To: "Somnath kotur" <so...@yahoo.com> Cc: dngire...@gmail.com, networking-discuss@opensolaris.org, driver-disc...@opensolaris.org Date: Tuesday, December 22, 2009, 7:51 AM Somnath kotur: Hi Robin, I am trying out the 'netstress' test while testing my 10G NIC Driver , while the 12 hr stress test passes and then moves on to the 'max sessions test' it invariably ends up in a situtation where an odd session fails to connect initially ... like the MAX results are always 2997/29998 out of 3000 sessions:Failed ... There have been instances where i have seen the same pass as well... . I tried recreating the same testcase by hand, i.e start netservers on both SUT and client , run start_netperf.sh with appropriate paramenters (1500 sessions) from client ,waited for a little while and then did the other way around, ran start_netperf.sh from SUT ... I got the following error msg 4 times : netperf: data send error: Transport endpoint is not connected i typed 'ps -A | grep netperf/netserve | wc -l ' on both the systems and confirmed that the SUT was 4 netperf sessions short, I immediately created 4 netperf sessions by hand with the same parameters as expected in the test case and sure enough there were a total of 3K sessions running and they all ran fine... Usually that's what seems to be happening , there seems to some glitch while setting up the sessions initially , but on trying it again ,they all seem to work just fine.. Is this really a bug in the driver that i need to look at or is this something not too serious? Thanks Somnath Hi, Somnath Yes, we've also encountered this small problem when testing 3000 sessions on 10G NIC, though everything is OK in 1G NIC driver. It seems be a problem in netperf when creating multiple sessions. I suggest you to use uperf to test it, it works well on a huge number of sessions, it can be got from http://www.uperf.org/ We are also working on change our test suite nicdrv from based on netperf to uperf. An workaround is to configure the DRIVER_SPEED variable to 10G in the config.vars file. (The default value is 1G) Thanks, Robin -----Inline Attachment Follows----- _______________________________________________ networking-discuss mailing list networking-discuss@opensolaris.org
journal.S150.20091223125900.execute.amd64.tar.gz
Description: application/gzip
_______________________________________________ networking-discuss mailing list networking-discuss@opensolaris.org
