- Wednesday, December 28, 2016

release-team@conference.openafs.org

Wednesday, December 28, 2016< ^ >

Room Configuration

Room Occupants

GMT+0
[01:47:55] kadukoafs@gmail.com/barnowlE45450E9 leaves the room
[01:48:04] kadukoafs@gmail.com/barnowlE45450E9 joins the room
[01:52:02] kadukoafs@gmail.com/barnowlE45450E9 leaves the room
[01:52:11] kadukoafs@gmail.com/barnowlE45450E9 joins the room
[03:48:30] mvita joins the room
[05:33:41] mvita leaves the room
[14:42:11] meffie joins the room
[14:59:58] wiesand joins the room
[15:00:49] <wiesand> Hello
[15:00:57] jgorse joins the room
[15:01:00] <jgorse> hello
[15:01:11] <wiesand> Hi Joe
[15:01:34] <wiesand> I wasn't actually expecting anyone today…
[15:01:47] <wiesand> All the same: do you have any Linux news?
[15:01:59] <jgorse> heh… this is probably why I could make it today. =)
[15:02:16] <jgorse> 4.9 is running well in all the tests I have performed thus far.
[15:02:29] <wiesand> Great, thanks!
[15:02:45] <wiesand> Have you happened to looked at 4.10-rc1 yet?
[15:03:01] <jgorse> I have been learning about eBPF and Brendan Gregg’s performance tools on 4.9.
[15:03:13] <jgorse> I have not yet tested 4.10-rc1 yet.
[15:03:19] <wiesand> er, what's eBPF?
[15:03:28] <jgorse> extended Berkley Packet Filter
[15:03:42] <wiesand> Ah thanks.
[15:03:54] <meffie> jgorse: some profiling on gerrit 11794 would be helpful
[15:04:02] <jgorse> this is basically an in-kernel processor that can do data flow with static safety.
[15:05:27] <jgorse> mmeffie: shall I do the bench_fs_syscall test or something different?
[15:05:48] <meffie> not sure yet.
[15:06:10] <jgorse> at the least, I will have an A-B setup.
[15:06:17] <wiesand> Ben +1'ed the "shake harder" stack ( openafs-stable-1_6_x (shake-harder) <https://gerrit.openafs.org/#/q/status:open+project:openafs+branch:openafs-stable-1_6_x+topic:shake-harder> ), so that's headed for 1.6.21. Additional testing would be most welcome.
[15:06:18] <meffie> some sort of parallel "build" load would do it think
[15:06:36] <meffie> a-b parallel loads
[15:06:50] <jgorse> hmm… I will try the bench_fs_syscall in parallel
[15:07:51] <jgorse> is it possible to have two OpenAFS clients running at the same time on a system, different domains (e.g. /afs1 /afs2)?
[15:08:01] <meffie> 11794 is fast path for dentry revalidate
[15:08:24] <wiesand> Joe: unlikely ;-)
[15:08:49] <wiesand> Mike, was this about those "stat like crazy" workloads?
[15:09:19] <meffie> yes
[15:09:40] <meffie> for example "make -j 32"
[15:10:10] <meffie> (with headers in /afs)
[15:10:30] <wiesand> If so: Have a huge tree in AFS, und run "find | xargs stat" over it, n times in parallel where n >= 2
[15:10:56] <meffie> yes
[15:11:23] <jgorse> weisand: I have a “find …” load which did not use xargs stat. this would be an easy thing for me to try.
[15:11:42] <wiesand> Worth a try I think.
[15:11:42] <meffie> network file systems have to revalidate dentries, with out these patches, we hit the GLOCK
[15:12:45] <meffie> this fixes that for at least the non-fakestat case, if i understand Andrew.
[15:14:14] <meffie> (there are other bottlenecks, but this is the first one.)
[15:15:37] <wiesand> I'm not sure whether the actual use case still exists. But performance was so abysmal that it was scary…
[15:16:39] jgorse kicks off the openafs-a build
[15:17:06] <meffie> you'll want to write the files to local disk.
[15:17:18] <meffie> so, it should be an out of tree build.
[15:18:13] <jgorse> so run configure via relative path?
[15:19:37] <jgorse> our configure is so slow
[15:19:56] <jgorse> I wish we did not unit test kernel details in the configure
[15:20:38] <meffie> i dont recall how to do the "out of tree" builds. i thought the bsd builds worked that way tho.
[15:21:25] <jgorse> autotools can invariably be hacked to comply. first I will try weisand’s find xargs stat test.
[15:25:39] <meffie> ben can give us a pointer later, i know he does the out of tree builds.
[15:28:01] <meffie> thanks jgorse
[15:28:50] <wiesand> mkdir build; cd build; ../openafs-x.y.z/configure; make
[15:34:20] <wiesand> More to discuss today?
[15:35:46] <meffie> oh, thanks.
[15:37:13] <meffie> no thank you wiesand, have a happy new year!
[15:37:29] <wiesand> Happy new year everyone!
[15:37:29] <jgorse> I think I am good for now.
[15:37:36] <jgorse> Happy new year!
[15:37:38] <wiesand> Thanks alot!
[15:37:42] wiesand leaves the room
[15:39:19] meffie leaves the room
[16:00:42] meffie joins the room
[16:00:55] meffie leaves the room
[16:00:56] meffie joins the room
[16:01:40] meffie leaves the room
[19:54:58] mvita joins the room
[20:13:29] <jgorse> Updated with some profiling data. Hopefully we can merge this guy:
[20:13:29] <jgorse> https://gerrit.openafs.org/#/c/11794/4
[20:26:17] mvita leaves the room