Switch to DuckDuckGo Search
   March 5, 2013  
< | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | >

bottom
[00:02:16] *** gkyildirim has joined #smartos
[00:04:26] *** gkyildirim has left #smartos
[00:15:15] *** ipalreadytaken has quit IRC
[00:15:46] *** neophenix has quit IRC
[00:16:36] *** ipalreadytaken has joined #smartos
[00:16:37] <pringlescan> Is there anyway to fix the high resolution timer when running Windows 2008 as a guest? I can't run benchmarks due to it being wildly inaccurate (sometimes it shifts 1 minute in a second)
[00:17:27] *** ipalreadytaken has quit IRC
[00:18:30] <rmustacc> pringlescan: What release of SmartOS is this?
[00:19:30] <pringlescan> rmustacc, latest USB image I downloaded today
[00:19:43] <pringlescan> it's running great by the way… it was easy sailing once I got it booted
[00:20:16] <ryancnelson> what's uname -a say?
[00:20:43] <pringlescan> SunOS 00-24-e8-7a-83-c4 5.11 joyent_20130222T000747Z i86pc i386 i86pc
[00:20:59] <rmustacc> Well, I haven't heard any reports of the slow hpet stuff for a while.
[00:21:09] <rmustacc> So I don't really know what's going on there without doing a full investigation, sorry.
[00:21:24] <pringlescan> It happens when I run Geekbench, I wasn't sure if it was a known issue.
[00:21:38] <pringlescan> it could also be my screwy bios
[00:21:51] <ryancnelson> did you convert that windows image from something else? or just create it from scratch, or what?
[00:21:55] <rmustacc> Haven't heard of it for a while.
[00:22:00] <pringlescan> I created it from scratch
[00:22:03] <rmustacc> People don't generally run the multimedia timers.
[00:22:27] <pringlescan> http://browser.primatelabs.com/geekbench2/compare/1719223/1703884 SmartOS beat baremetal in a couple of tests
[00:22:44] <rmustacc> That sounds a bit suspicious.
[00:22:46] <pringlescan> I upgraded the RAM to a different configuration using less channels, so I think if I put the original RAM back in and test it, the scores will be almost even.
[00:22:48] <ryancnelson> hurray, ARC cache!
[00:23:15] *** brendang has joined #smartos
[00:23:50] <pringlescan> I think those few tasks benefitted from hyperthreading to the point where treating the threads as individual cores may have helped? I'm not sure how HT is handled in SmartOs
[00:24:02] *** ira_ has joined #smartos
[00:24:26] <wesolows> they appear as virtual CPUs and can be scheduled onto
[00:24:38] *** ira has quit IRC
[00:24:38] *** ira_ is now known as ira
[00:25:04] *** szaydel has quit IRC
[00:25:25] <richlowe> tries to be smart about it, but can fail to if we get bad (or unexpected) info from acpi.
[00:25:37] <pringlescan> so if I want my guest o/s to have full access to a dual CPU system with 16 threads and 8 cores, should I specify 16 vcpus in the JSON file?
[00:25:57] <wesolows> there's some awareness of how the CPUs are laid out with respect to sockets, packages, etc.  psrinfo is a starting place.
[00:26:33] <rmustacc> pringlescan: Is the only thing this machine is going to is be running this one guest?
[00:26:36] <pringlescan> wesolows, it's weird, the first entry (0) was live 3 seconds before the rest of the entries (up to 15)
[00:26:54] <pringlescan> rmustacc, no it'll have between 1-4 VMs, light duty 95% of the time, and some batch jobs
[00:27:09] <rmustacc> Then I would not give the guests 16 CPUs.
[00:27:12] <rmustacc> Just as many as you think they need.
[00:27:43] <pringlescan> I figured schedulers were smarter than I was and would give out a fair share? but I guess that's putting undue burden on the hypervisor?
[00:28:19] <rmustacc> Well the host schedular is smart.
[00:28:23] <rmustacc> But it doesn't know anything about the guests.
[00:28:42] <rmustacc> So it could easily deprioritize one of your threads that is holding a spin lock because it has 15 other virtual cpus to run.
[00:28:46] <rmustacc> For your one process/zone.
[00:29:21] <pringlescan> That makes sense.
[00:29:39] <pringlescan> is there a book out I can read on smartos or illumos, I didn't see anything on amazon
[00:30:38] <rmustacc> What are you interested in?
[00:31:13] *** bluezenix has quit IRC
[00:31:35] <pringlescan> I don't know anything about solaris, I've only run linux. I have to play developer/sysadmin and the stack is nginx/postgresql/node.js
[00:31:36] <ryancnelson> the Opensolaris Bible is pretty good reading, and is mostly still relevant... lets you know about zfs and zones, and crossbow, at least.
[00:31:50] <ryancnelson> nothing in there on our kvm hypervisor, though.
[00:32:11] <pringlescan> that's more fun to know, probably somewhat less relevant for the immediate future
[00:32:13] <ryancnelson> there was a blog post recently about what chapters are still valid
[00:33:54] <pringlescan> okay, that's on its way.
[00:34:21] <rmustacc> If you're interested in internals, Solaris Internals 2nd edition is still useful.
[00:34:32] <rmustacc> If you're doing to be doing dev/sysadmin stuff, you're going to want to learn DTrace.
[00:34:34] <pringlescan> I'm going to try to blog anything I have trouble with to contribute to what comes up in google
[00:34:39] <rmustacc> For that I really suggest the DTrace book in Brendan.
[00:34:49] <rmustacc> DTrace book by Brendan Gregg.
[00:34:54] <pringlescan> Dtrace was a deciding factor
[00:35:14] <rmustacc> It has a lot of one liners and good advice on how to get started looking at things in subsystems where you've never been before.
[00:35:33] <rmustacc> I personally use it a lot, when I'm not bothering him directly.
[00:36:12] <Triskelios> it's the Brendan you can carry
[00:36:21] *** jamesd2 has joined #smartos
[00:36:23] <pringlescan> solaris books are expensive, it's fascinating
[00:36:27] *** ipalreadytaken has joined #smartos
[00:36:36] <pringlescan> they're also offered for rental as if they're academic
[00:37:03] <ryancnelson> this is the opensolaris bible to illumos rosetta stone i was talking about:  http://wiki.smartos.org/display/DOC/The+OpenSolaris+Bible
[00:37:09] <rmustacc> This is probably the best bang for buck: http://www.amazon.com/DTrace-Dynamic-Tracing-Solaris-FreeBSD/dp/0132091518/ref=sr_1_1?ie=UTF8&qid=1362440212&sr=8-1&keywords=DTrace
[00:37:16] <Triskelios> Solaris Internals *is* regularly found in academia
[00:39:09] *** ipalreadytaken has quit IRC
[00:39:28] <pringlescan> I'll probably do a series of Ubuntu -> SmartOS blogs detailing my adventure, I try to feed any of the weird things I run into back into the interwebs
[00:39:42] *** ipalreadytaken has joined #smartos
[00:41:23] <pringlescan> I ordered the opensolaris book and dtrace one, thanks for all your help guys.
[00:42:28] *** pringlescan has left #smartos
[00:43:05] <ryancnelson> looking forward to the blog posts
[00:44:07] *** ipalreadytaken has quit IRC
[00:45:04] <ryancnelson> i got one of these last weekend:  http://en.code-bude.net/2013/01/29/how-to-build-a-20dollar-low-cost-openwrt-router-flash-openwrt-to-tp-link-wr703n/ ... considering turning it into a teeny PXE boot server to carry with me to the datacenter
[00:46:20] *** ipalreadytaken has joined #smartos
[00:48:26] *** trentster1 has joined #smartos
[00:49:23] *** trentster has quit IRC
[00:56:07] *** wolfeidau has quit IRC
[00:56:18] *** wolfeidau has joined #smartos
[01:01:53] <nahamu> ryancnelson: nice
[01:04:32] *** Cpt-Oblivious has quit IRC
[01:06:25] *** trentster1 has quit IRC
[01:07:57] *** ipalreadytaken has joined #smartos
[01:11:59] *** darjeeling has quit IRC
[01:13:43] *** ipalreadytaken has quit IRC
[01:16:43] *** ipalreadytaken has joined #smartos
[01:17:41] *** szaydel has joined #smartos
[01:25:26] *** tonyarkles has quit IRC
[01:25:53] *** trentster has joined #smartos
[01:26:54] <trentster> hey all, my connection dropped, if there were any responses, please could someone repaste them
[01:27:35] <trentster> I am pxe booting now, and everything is working, with smartos=true in the pxeboot menu, except no ssh to the box
[01:27:41] <trentster> networking is working
[01:30:00] *** axonpoet has quit IRC
[01:33:30] <nahamu> trentster: responses to what?
[01:34:25] <trentster> I asked a question and then my internet went down..so I did not see any responses if they came in.
[01:34:30] <nahamu> Alternately, http://echelog.com/logs/browse/smartos/ :)
[01:34:50] <trentster> nahamu: aaah thanks
[01:34:51] <nahamu> I'm not sure your question made it through...
[01:36:58] <trentster> I am pxebooting via linuxpxe
[01:37:35] <trentster> I am having a problem, where the node boots up properly, all the vms start, networking is working, but it seems ssh does not start
[01:38:28] <trentster> So I can not ssh to the GZ of the node
[01:39:31] <trentster> https://gist.github.com/trentster/76f44ed0deabc7372f39
[01:39:38] <trentster> thats what I am using to pxe boot
[01:39:43] <trentster> any ideas?
[01:40:31] *** ipalreadytaken has quit IRC
[01:41:06] *** ipalreadytaken has joined #smartos
[01:41:27] <ryancnelson> can't even connect, or just can't log in?
[01:41:31] <rmustacc> I'd get on the console.
[01:41:37] <rmustacc> And then see what's happened on the machine.
[01:44:59] *** ipalread_ has joined #smartos
[01:46:15] *** leecallen has quit IRC
[01:46:38] <ryancnelson> if you have an IP conflict, it'd probably boot ok, but then turn off the IP in the GZ, with "DUPLICATE" in the ifconfig flags
[01:46:39] *** leecallen has joined #smartos
[01:47:31] <ryancnelson> ... and all your vm's, having different addresses, would be ok, too
[01:48:33] *** ipalreadytaken has quit IRC
[01:49:19] <ryancnelson> if you look at the arp listing from your client machine (assuming you're on that lan), what's the mac-address that *you* see look like?  does it match what's on the headnode's nic?
[01:53:40] *** potatosalad has quit IRC
[02:00:19] *** Forced has joined #smartos
[02:01:22] *** willmurnane has joined #smartos
[02:01:48] <willmurnane> is there a way to establish a boot dependency or delay with smartos?
[02:02:12] <willmurnane> i.e., before starting vm X, wait until vm Y has been started for 30 seconds
[02:02:22] <ryancnelson> oh
[02:02:28] <ryancnelson> huh... that's an interesting question.
[02:02:37] <ryancnelson> the answer is "sure, why not" :)
[02:02:56] <willmurnane> basically, i want to run my LDAP server as a VM, and have all the other VMs wait until it's up to try to start
[02:02:57] <ryancnelson> ... mark your zone/vm (a KVM vm lives in a zone)
[02:03:04] <ryancnelson> ... as "autoboot=false"
[02:03:38] <ryancnelson> ... then create a SMF service, and, as it's dependencies, have something that checks for your pre-conditions
[02:03:47] <ryancnelson> (that's the high-level fix)
[02:04:09] <jesse_> ryancnelson, except that autoboot=true is set whenever you start a zone
[02:04:46] <ryancnelson> oh, yeah... the act of turning it on (with vmadm) sets it to true
[02:05:03] <willmurnane> i guess i could make my new service come before the vm-starting service, and have it turn off autoboot on everything but the LDAP server...
[02:05:15] *** ipalread_ has quit IRC
[02:07:18] <ryancnelson> you might look at /lib/svc/method/svc-zones ... if you can modify that, you can decide which zones boot, and when
[02:08:35] <ryancnelson> would a hack that says "find all zones that *aren't* your ldap server, and shut them down, then boot them again a little later" be ok?
[02:08:46] <ryancnelson> ... run that at boot time?
[02:09:43] <willmurnane> yeah, i guess that would work. i'll play around with the concepts and see what i can come up with
[02:10:38] <ryancnelson> you're familiar with the "/opt/custom/smf/mymanifest.xml" trick?
[02:10:44] <willmurnane> yes
[02:10:49] <ryancnelson> coolio
[02:12:04] *** cheeseplus has joined #smartos
[02:19:48] *** trentster has quit IRC
[02:23:13] *** arekinath has joined #smartos
[02:23:14] *** arekinath has joined #smartos
[02:25:06] *** trentster has joined #smartos
[02:41:34] *** denizr has joined #smartos
[02:42:08] *** ryancnelson has quit IRC
[02:44:08] *** notmatt has quit IRC
[03:08:01] *** denizr has quit IRC
[03:11:17] *** darjeeling has joined #smartos
[03:11:19] *** denizr has joined #smartos
[03:26:22] *** miine___ has joined #smartos
[03:26:54] *** miine_ has quit IRC
[03:27:54] *** miine___ has quit IRC
[03:28:03] *** miine has joined #smartos
[03:32:08] *** denizr has quit IRC
[03:38:39] *** dap has quit IRC
[03:39:29] *** dap has joined #smartos
[03:39:52] *** denizr has joined #smartos
[03:40:35] *** dap has quit IRC
[03:42:04] *** arai1 has left #smartos
[03:52:51] *** potatosalad has joined #smartos
[03:56:37] <nahamu> looks like the new imgadm doesn't like my image server :-/
[03:57:43] *** darjeeling has quit IRC
[03:57:51] <nahamu> time to write a new one, I guess...
[03:58:00] *** darjeeling has joined #smartos
[04:01:05] <trentster> nahamu: we got the exact same problem with project-fifo
[04:01:23] *** darjeeling has quit IRC
[04:01:27] <trentster> and MerlinDMC has to rewrite datasets.at as well
[04:04:21] <nahamu> alas
[04:04:39] <nahamu> I thought MerlinDMC had said that the new imgadm was working fine with datasets.at
[04:06:49] *** ira has quit IRC
[04:09:00] <richlowe> I thought the heads-up said that it should, too.
[04:14:44] *** szaydel has quit IRC
[04:20:41] *** denizr has quit IRC
[04:24:39] *** axonpoet has joined #smartos
[04:27:27] *** krisa has joined #smartos
[04:38:51] *** denizr has joined #smartos
[04:39:49] *** szaydel has joined #smartos
[05:04:32] *** prasm has quit IRC
[05:05:51] *** gregburd has quit IRC
[05:06:19] *** deedubs has quit IRC
[05:07:29] *** Azbruh has quit IRC
[05:07:56] *** Raynos has quit IRC
[05:08:05] *** hungryblank has quit IRC
[05:09:08] *** Azbruh has joined #smartos
[05:24:16] *** abhishekvasisht has joined #smartos
[05:25:02] *** porkbelt has quit IRC
[05:25:28] *** porkbelt has joined #smartos
[05:27:28] *** prasm has joined #smartos
[05:36:49] *** dabb has quit IRC
[05:39:22] *** willmurnane has quit IRC
[05:45:42] *** prasm has quit IRC
[05:46:48] *** dabb has joined #smartos
[05:50:37] *** darjeeling has joined #smartos
[05:54:20] *** prasm has joined #smartos
[05:57:23] *** porkbelt has quit IRC
[05:59:34] *** wolfeidau has quit IRC
[05:59:43] *** wolfeidau has joined #smartos
[06:05:49] *** denizr has quit IRC
[06:07:41] <trentster> nahamu: he may have updated it already
[06:09:48] *** sachinsharma has joined #smartos
[06:10:14] *** jamesd2 has quit IRC
[06:20:26] *** Licenser has quit IRC
[06:21:50] *** Licenser has joined #smartos
[06:25:45] *** krisa has quit IRC
[06:36:48] *** estibi_ has joined #smartos
[06:39:56] *** estibi has quit IRC
[06:43:11] *** notmatt has joined #smartos
[06:45:45] *** geoffnorthcott has joined #smartos
[06:46:36] *** szaydel has quit IRC
[06:46:42] *** potatosalad has quit IRC
[06:50:34] *** deedubs has joined #smartos
[06:53:52] *** Raynos has joined #smartos
[06:54:58] *** gregburd has joined #smartos
[06:57:44] *** darjeeling has quit IRC
[07:00:11] *** nefilim has quit IRC
[07:02:08] *** wolfeidau has quit IRC
[07:04:44] *** hungryblank has joined #smartos
[07:14:48] *** ipalreadytaken has joined #smartos
[07:16:30] *** asonge has joined #smartos
[07:19:29] *** wolfeidau has joined #smartos
[07:19:34] *** porkbelt has joined #smartos
[07:29:16] <MerlinDMC> morning
[07:29:44] <Licenser>  morning MerlinDMC
[07:29:45] <MerlinDMC> trentster + nahamu, the new imgadm does work fine for me on datasets.at - without code changes
[07:30:43] *** arekinath has quit IRC
[07:31:04] *** arekinath has joined #smartos
[07:31:05] *** arekinath has joined #smartos
[07:31:14] <MerlinDMC> nahamu, the new one seems to use the /ping endpoint to determine if the server is capable of imgapi or just dsapi ... if you implement the ping thing it should work (assuming that all the manifests are OK because imgadm v2 is validating them before it converts them to the new format)
[07:39:49] *** darjeeling has joined #smartos
[07:42:35] *** des2 has quit IRC
[07:59:15] *** Webhostbudd_ has joined #smartos
[08:03:14] *** Webhostbudd has quit IRC
[08:15:48] *** alucardX has joined #smartos
[08:28:19] *** geoffnorthcott is now known as Miraz
[08:32:00] *** papertigers has quit IRC
[08:32:01] *** bahamas10 has quit IRC
[08:34:22] *** bahamas10 has joined #smartos
[08:34:33] *** papertigers has joined #smartos
[08:40:00] *** bluezenix has joined #smartos
[08:48:53] *** alcir has joined #smartos
[08:50:48] *** bens1 has joined #smartos
[08:54:56] *** ChrisPartridge has quit IRC
[08:56:57] *** bluezenix has quit IRC
[08:57:39] *** darjeeling has quit IRC
[09:10:21] *** zug has quit IRC
[09:13:13] *** estibi_ is now known as estibi
[09:14:24] *** bens1 has quit IRC
[09:20:27] *** texarcana has quit IRC
[09:21:39] *** texarcana has joined #smartos
[09:24:08] *** ipalreadytaken has quit IRC
[09:25:15] *** disharmony has quit IRC
[09:27:22] *** chh has joined #smartos
[09:37:44] *** bluezenix has joined #smartos
[09:37:50] *** bluezenix1 has joined #smartos
[09:40:54] *** darjeeling has joined #smartos
[09:43:16] *** abhishekvasisht1 has joined #smartos
[09:45:00] *** abhishekvasisht has quit IRC
[09:51:52] *** KermitTheFragger has joined #smartos
[09:52:18] *** ChrisPartridge has joined #smartos
[10:22:49] <alcir> what about Percona XtraDB Cluster
[10:22:49] *** notmatt has quit IRC
[10:24:38] *** ipalreadytaken has joined #smartos
[10:28:44] *** ipalreadytaken has quit IRC
[10:29:51] *** bens1 has joined #smartos
[10:32:46] *** cheeseplus has quit IRC
[10:37:38] *** Miraz has quit IRC
[10:37:45] *** Miraz has joined #smartos
[10:39:36] *** stetho has joined #smartos
[10:51:13] *** darjeeling has quit IRC
[10:56:19] *** avrntsv has joined #smartos
[11:08:16] *** trentster1 has joined #smartos
[11:10:29] *** trentster has quit IRC
[11:16:04] *** trentster has joined #smartos
[11:17:14] *** trentster1 has quit IRC
[11:25:48] *** ipalreadytaken has joined #smartos
[11:30:14] *** ipalreadytaken has quit IRC
[11:48:59] *** darjeeling has joined #smartos
[11:49:22] *** bens1 has quit IRC
[11:50:33] *** bens1 has joined #smartos
[11:58:25] *** xenol has quit IRC
[11:58:45] *** xenol has joined #smartos
[12:00:26] *** darjeeling has quit IRC
[12:07:02] *** jamesd2 has joined #smartos
[12:09:24] *** bahamas10 has quit IRC
[12:09:57] *** papertigers has quit IRC
[12:11:50] *** bahamas10 has joined #smartos
[12:12:12] *** papertigers has joined #smartos
[12:12:44] *** jamesd2 has quit IRC
[12:17:20] *** chh has quit IRC
[12:17:51] *** chh has joined #smartos
[12:26:15] *** ipalreadytaken has joined #smartos
[12:26:38] *** sachinsharma has quit IRC
[12:31:25] *** ipalreadytaken has quit IRC
[12:31:35] *** wramthun has quit IRC
[12:33:25] *** wramthun has joined #smartos
[12:41:58] *** darjeeling has joined #smartos
[12:54:21] *** papertigers has quit IRC
[12:55:13] *** bahamas10 has quit IRC
[12:55:30] <Licenser> hmm I'm trying to compile rsyslog with ssl support and it works great but when trying to start it - it does not seem to see the gcrypt library for some odd reason any hints what I may be doing wrong?
[12:56:09] <jperkin> what's the actual error?
[12:56:46] *** papertigers has joined #smartos
[12:57:20] *** bahamas10 has joined #smartos
[12:57:42] <Licenser> jperkin: 2013-02-25T09:27:53.098036+00:00 build rsyslogd-2066: could not load module '/opt/local/lib/rsyslog/lmnsd_gtls.so', dlopen: ld.so.1: rsyslogd: fatal: relocation error: file /opt/local/lib/rsyslog/lmnsd_gtls.so: symbol gcry_control: referenced symbol not found
[12:58:20] <Licenser> it seems that gcry_control is provided by the lib: /opt/local/lib/libgcrypt.so:000000000001b7a0 T gcry_control
[12:58:46] <Licenser> and I sadly am not an expert at linking it always escaped me -.-
[12:59:29] <jperkin> ok so you need to ensure that the build for lmnsd_gtls.so has -lgcrypt
[13:00:11] <jperkin> if it's an autoconf-based package you can normally just add it to LIBS, or edit the Makefile and add it to lmnsd_gtls_la_LDFLAGS or so
[13:00:26] <Licenser> bulding it with pkg-src :)
[13:00:35] <jperkin> ok, even better ;)
[13:00:39] <Licenser> which does not make it easyer for me ^^ it's a monster of a system
[13:00:46] <Licenser> I will make a pull request if I get this done :P
[13:00:59] <jperkin> simplest fix is to just add 'LDFLAGS+= -lgcrypt' to the pkgsrc Makefile
[13:01:13] <jperkin> not ideal as that will get added to all objects, but should at least solve the problem
[13:01:33] <MerlinDMC> isn't rsyslog already used at the platform?
[13:01:46] <Licenser> that looks easy
[13:03:00] <Licenser> one day I'll get the hang of pkgsrc
[13:03:15] <Licenser> jperkin btw are you guys interested in a pull request for nginx-1.3.13?
[13:05:10] <jperkin> Licenser: too late: http://mail-index.netbsd.org/pkgsrc-changes/2013/02/22/msg085161.html :)
[13:05:43] <Licenser> hah
[13:06:06] <Licenser> well I still learned something building it :P
[13:08:12] <Licenser> man jperkin  if I ever get a chance I buy you a beer!
[13:08:13] <Licenser> it worked
[13:08:49] <Licenser> or tea, or coffee, or coke, or clubmate, or pretty much whaterver ;)
[13:09:00] <jperkin> pkgsrccon in berlin in a few weeks? ;)
[13:09:10] <Licenser> you're there?
[13:09:13] <jperkin> yup
[13:09:16] <Licenser> awesome!
[13:09:27] * Licenser will try to be too!
[13:10:02] <MerlinDMC> hmm ... saturday
[13:10:07] *** chh has quit IRC
[13:10:19] *** chh has joined #smartos
[13:11:26] <MerlinDMC> maybe i get some spare time as well and try to be not to lazy for the travel
[13:25:39] <Licenser> :( it's not wokring
[13:26:41] *** ipalreadytaken has joined #smartos
[13:30:54] *** ira has joined #smartos
[13:33:26] <Licenser> mostly because I was stupid (and the logging of the logger sucks)
[13:33:41] *** cwo has joined #smartos
[13:34:44] *** cwo has quit IRC
[13:34:55] *** cwo has joined #smartos
[13:35:28] <ChrisPartridge> Is there an article that addresses vlans between multiple servers? e.g. zones on 2 seperate pysical servers, on a single vlan?
[13:45:40] *** khushildep has joined #smartos
[13:46:29] *** leecallen has quit IRC
[13:46:57] *** leecallen has joined #smartos
[13:47:02] *** bluezenix1 has quit IRC
[13:59:00] *** Licenser has quit IRC
[14:00:32] *** Licenser has joined #smartos
[14:11:39] <Licenser> jperkin you guys are interested in the SSL enabled rsyslogd?
[14:17:13] <khushildep> Anyone noticed odd performance with Sun JDK?
[14:17:20] *** abhishekvasisht1 has quit IRC
[14:19:44] <yofuh> what is "odd performance"?
[14:19:44] *** jamesd2 has joined #smartos
[14:21:02] <Alasdairr> Yes, if you have it installed and visit certain websites your computer gets hacked
[14:21:08] <Alasdairr> But that's probably not what you meant ;-)
[14:21:47] <jesse_> applets are so... actually, there never were a decade when applets were in!=)
[14:21:52] <jesse_> was
[14:22:20] <jesse_> and yes, I'm interested to hear about off performance, too
[14:22:33] <jesse_> odd, too
[14:23:30] *** bluezenix has joined #smartos
[14:25:16] <jperkin> Licenser: fixes for it, sure.
[14:25:32] <Licenser> not technically fixes :) it just wasn't enabled by default
[14:26:07] <Licenser> pretty much 7 lines changed (not even sure if a pull request is worth it)
[14:26:23] <Licenser> propably a simple diff would do the trick
[14:26:25] <jperkin> yeh but I thought PKG_OPTIONS.rsyslog+=gnutls didn't work without additional LDFLAGS?
[14:26:44] <jperkin> right, patches are always preferable actually as we can't really handle pull requests
[14:27:17] <Licenser> ah okay
[14:27:37] <Licenser> no it needed the LD flags and one addiotnal included .mk file (for gnupg)
[14:28:13] <jperkin> gnupg? that doesn't sound right
[14:29:04] <jesse_> does gnutls still call randomly exit() when it has problems with some ssl connection?=)
[14:29:06] <Licenser> it requires it for some minor tasks - if you don't have it the build fails
[14:29:18] <Licenser> jesse_ no clue ^^
[14:29:20] <Triluch> can i delegate dataset after zone is created?
[14:29:39] <MerlinDMC> Triluch, delegate_dataset can be set on create only
[14:30:22] <jesse_> unsupported; you can still delegate afterwards with zonecfg
[14:30:47] *** szaydel has joined #smartos
[14:32:05] <Licenser> jperkin sorry I was totally wrong it was libgcrypt
[14:32:23] <Licenser> no clue why I thought it was gnupg
[14:38:07] *** khushildep has quit IRC
[14:42:20] <Licenser> jperkin https://github.com/joyent/pkgsrc-wip/issues/2 :) enjoy
[14:52:38] <alcir> vrrpadm must be issued inside the zone?
[15:10:32] *** jim80net has joined #smartos
[15:13:32] *** phus has quit IRC
[15:14:35] <alcir> by the way there are problems in the vrrp
[15:15:25] <alcir> if I ping an ip handled via vrrp and the master is a zone on another server
[15:16:03] <alcir> and the backup is on the server from wich I run the ping
[15:16:15] <alcir> I got no answers ...
[15:19:53] *** denizr has joined #smartos
[15:22:27] *** chorrell has joined #smartos
[15:24:35] <alcir> if the gz think that the vrrp mac address is handled by herself
[15:24:47] <alcir> there is no way to send packet outside
[15:24:51] <alcir> ...?
[15:27:32] <nahamu> MerlinDMC: does dsapid store the iamge files in couchdb?
[15:28:27] <MerlinDMC> nahamu, my thingy there ... yes
[15:28:55] <MerlinDMC> and I basically proxy download requests via nginx to couchdb on datasets.at ... no nodejs used for file streaming
[15:28:56] <nahamu> (I find it funny that imgadm can talk to datasets.at, datasets.at can talk to datasets.shalman.org, but imgadm can't talk to datasets.shalman.org)
[15:29:07] <MerlinDMC> :)
[15:29:33] <MerlinDMC> if you want we can dig there a little ... but right now i'm in the office
[15:30:31] <MerlinDMC> will be back home in about 3h 30m ... depends on if i get the first train :P
[15:30:54] <nahamu> If you have the time later, I certainly would appreciate the help. :)
[15:31:44] <nahamu> I wonder if it's the content type header I'm sending that's doing it
[15:31:55] <nahamu> could also be nginx doing something funky, I suppose.
[15:32:42] *** cwo has quit IRC
[15:33:00] *** enmand has joined #smartos
[15:33:41] <MerlinDMC> I'll try to get my macbook on the desk
[15:38:29] <alcir> at the same time I cannot ping to the vrrp ip from the zone in backup state
[15:42:14] <alcir> this vrrp seems unuseful
[15:46:44] <Alasdairr> VRRP needs L3PROTECT disabled
[15:46:58] <Alasdairr> you probably have disabled it already
[15:47:04] <Alasdairr> but thought i'd mention it in case not
[15:47:13] <alcir> L3PROTECT
[15:47:15] <alcir> uh
[15:48:08] *** th has joined #smartos
[15:48:58] <alcir> how to disable it and what is it?
[15:49:00] <alcir> :-)
[15:50:55] *** colstrom has joined #smartos
[15:51:07] <alcir> http://wiki.smartos.org/pages/viewpage.action?pageId=755858&navigatingVersions=true
[15:53:05] *** colstrom has quit IRC
[15:53:22] <alcir> by the way it doesn't work
[15:54:05] *** colstrom has joined #smartos
[15:55:50] <Alasdairr> what doesn't work?
[15:56:14] <Alasdairr> oh that wiki documentation sucks
[15:57:23] <alcir> I disabled l3PROTECT
[15:57:27] <alcir> I think
[15:57:38] *** tonyarkles has joined #smartos
[15:57:58] <Alasdairr> from our wiki: http://paste.ec/?40658263643ffa72#p7Q0nD1ooctbSzYkaNqamrIFyUxzd3Qu08DYBreR8mc=
[15:58:12] <th> mdb -k <<< ::memstat  tells me that > 80GB (63%) of memory are used for "ZFS File Data"; i guess that is the reason, why vmadm complains that it does not have enough memory to create a new 16GB vm.  is there a way to reduce the "ZFS File Data" memory, or should that happen automatically?
[15:58:16] <Alasdairr> obviously "mac" is the mac of your nic
[15:58:20] <alcir> but I cannot ping the vrrp ip from the gz where is hosted the zone having the vrrp backup instance
[15:58:45] <Alasdairr> EBRAINNOPARSE
[15:59:13] <alcir> allow_ip_spoofing on the primary nic
[15:59:21] <alcir> I mean not the vrrp nic
[15:59:24] <alcir> or both?
[16:02:46] *** denizr has quit IRC
[16:02:53] <MerlinDMC> th, probably this could be useful: https://www.illumos.org/issues/1974
[16:03:03] *** denizr has joined #smartos
[16:04:32] *** while1eq1_away has joined #smartos
[16:07:17] <fortytwo_> th: you can probably limit the zfs ARC memory usage with the "set zfs:zfs_arc_max = number_of_bytes" in /etc/system iguess
[16:07:22] <fortytwo_> or similar
[16:07:23] <th> MerlinDMC: i see - so the proposal is to force releasing some pages by allocating some memory
[16:07:54] <Alasdairr> MerlinDMC: I'm surprised it hasn't been looked at
[16:08:18] *** neophenix has joined #smartos
[16:08:28] <th> fortytwo_: sure thing. but it still makes sense to let arc use the avail memory once the VMs allocated what they need.
[16:08:52] <fortytwo_> indeed
[16:08:59] <th> is vmadm-create to blame here for not recognising "ZFS File Data" as usable memory?
[16:10:12] <szaydel> vmadm must be doing a memory check to determine whether there is sufficient memory to actually run the VM.
[16:10:47] <szaydel> It is interesting to me, since just creating a VM does not mean it will be running, nor does it mean that others won't be shut down to yield additional memory.
[16:10:51] <th> i agree. i dont want to overcommit. but 63% (~80GB) of memory on zfs caching is nice to have, but not if it prevents a 16GB VM to be created.
[16:13:02] <th> the perl alloc thing eats up "Free (freelist)" (still 11GB in there) first, so it does not reduce "ZFS File Data" for me
[16:16:08] *** tonyarkles has quit IRC
[16:22:08] *** khushildep has joined #smartos
[16:22:42] *** potatosalad has joined #smartos
[16:22:55] <khushildep> Hi folks - sorry dropped off earlier. THe problem I'm see is a lot of vm_xalloc mutex locks - which is odd. I don't see the same on a linux box which is even odder...
[16:25:41] <th> my workaround for now: + payload.available_MiB = payload.available_MiB + 10000;
[16:25:45] <th> vm successfully created
[16:30:43] *** tonyarkles has joined #smartos
[16:45:59] *** nefilim has joined #smartos
[16:49:20] *** alucardX has quit IRC
[16:53:26] *** sachinsharma has joined #smartos
[17:07:24] *** Azbruh has quit IRC
[17:09:07] *** Azbruh has joined #smartos
[17:10:15] <EMH_Mark3> anyone else having performance issues with virtio drives/network adapters in linux kvms? I can't seem to get more than 100MB/s on either
[17:10:53] <Alasdairr> yeah, everyone
[17:11:05] *** axonpoetsphere has joined #smartos
[17:13:29] *** sachinsharma has quit IRC
[17:14:46] *** axonpoet has quit IRC
[17:14:46] *** axonpoetsphere is now known as axonpoet
[17:15:03] <elijah-mbp> EMH_Mark3: do you have txtimer bits in your vm config?
[17:15:26] <elijah-mbp> 100mbit seems artificially low - you should be able to get 700-800M with txtimer set, easily
[17:15:41] *** khushildep has quit IRC
[17:15:45] <elijah-mbp> no txtimer, it tends to swim along around 300-600mbit in my experience.
[17:15:47] *** khushildep_ has joined #smartos
[17:15:56] <elijah-mbp> remember that with a virtualized NIC, its all on cpu.
[17:16:03] <elijah-mbp> *make sure it has plenty of CPU*.
[17:17:43] <nahamu> elijah-mbp: what are the txtimer bits?
[17:18:13] <nahamu> is there something better to set it to than the default?
[17:18:30] <nahamu> or it is mostly a latency vs bandwidth tradeoff?
[17:19:10] *** CarlosC has joined #smartos
[17:19:25] <elijah-mbp> at one point it was SOP for us to add an attribute for virtio-txtimer to get better performance.  most people didn't 'need' it, but it was helpful.  hang a sec and i'll paste the command in.
[17:20:39] <elijah-mbp> zoneadm -z YOURUUIDGOESHERE 'add attr; set name=virtio-txtimer; set type=string; set value=100000; end'
[17:20:52] *** avrntsv has quit IRC
[17:21:42] <elijah-mbp> what that does, exactly, i am SO not the right person to ask.  :p  it frobs the timers for virtio - vague.
[17:21:54] <elijah-mbp> it does make it GoFaster, though.
[17:22:06] <elijah-mbp> or so my iperf testing always showed.
[17:22:26] <elijah-mbp> i sort of expected that to show up in vmadm or something's defaults by now - it may do it for you and i just don't know it.
[17:22:50] <elijah-mbp> (skew between what we run in production and what you guys get in open source releases (which in this case, are probably BETTER...)
[17:23:36] <elijah-mbp> if you literally can't get it to do more than 100mbit with that, and no amount of free CPU given to it helps, then odds are good that something is broken.
[17:24:31] <elijah-mbp> EMH_Mark3: is the pokey network on a gigabit or a 10G line?
[17:30:05] <jesse_> jperkin, trunk dataset has the repo twice in /opt/local/etc/pkgin/repositories.conf (at least mine has, and I don't think I've added anything to the file)
[17:30:10] *** Cpt-Oblivious has joined #smartos
[17:30:28] *** khushildep_ has quit IRC
[17:31:22] <jperkin> jesse_: yeh, same bug as the 2012Q4 multiarch one, I'll fix on the next iteration
[17:31:36] <jesse_> jperkin, jsut a fyi=)
[17:31:40] <jperkin> thanks
[17:33:41] *** cheesepl_ has joined #smartos
[17:34:00] *** chh has quit IRC
[17:38:48] *** alcir has left #smartos
[17:40:28] <nahamu> elijah-mbp: rmustacc might know what it does more accurately
[17:41:04] <elijah-mbp> yeah, robert or brendan are probably your guys if you care about VM virtio-net speed :)
[17:41:12] <nahamu> currently looks like the default in SmartOS is 100000, so your SOP is to double the timeout
[17:41:49] <nahamu> which I suspect means it spools up more bits before the timer fires so it improves your bandwidth (probably at the price of a small latency increase).
[17:42:20] *** khushildep has joined #smartos
[17:46:28] <EMH_Mark3> elijah-mbp: sorry, got pulled away by something. the 'pokey network' is a vnic over virtio. but I'm also getting low speeds on the virtio hard drive (/dev/vda). and finally I'm getting 100MB/s, not 100Mbit/s :)
[17:46:58] <elijah-mbp> 100MB/sec is lots better than 100mbit :)
[17:47:08] <EMH_Mark3> about 8 times better, approximately.
[17:47:25] <EMH_Mark3> but still way lower than raw zfs pool speed
[17:47:54] <jesse_> 100MBps is not that bad on a GE with small packets?
[17:48:56] *** szaydel has quit IRC
[17:51:14] *** ryancnelson has joined #smartos
[17:54:24] <elijah-mbp> EMH_Mark3: dunno about the vda issues - maybe post bonnie results from in-VM and raw ZFS-in-the GZ to smartos-discuss and see which engineers pop their heads up?
[17:54:56] <EMH_Mark3> is virtio-txtimer only applicable for network adapters?
[17:55:49] <elijah-mbp> think so, yeah.
[17:55:55] <EMH_Mark3> bummer.
[17:56:03] <elijah-mbp> txtimer = transmit timer i think
[17:56:29] <EMH_Mark3> it does seem like cpu usage is getting maxed out when doing heavy i/o from either /dev/vda or virtio net adapter
[17:56:48] <EMH_Mark3> (usage in vm that is)
[18:00:03] *** szaydel has joined #smartos
[18:01:22] <rmustacc> EMH_Mark3: If you're doing a lot of disk I/O do you have a slog?
[18:03:50] <rmustacc> And I would focus on something simple like using dd or something you write that does direct I/O instead of something like bonnie.
[18:06:52] *** ira has quit IRC
[18:09:36] <cheesepl_> before I punt and just fire up a kvm been able to get therubyracer gem to compile in a zone?
[18:09:52] <cheesepl_> has anyone been able*
[18:10:11] <rmustacc> I haven't tried, but if you have a build log of a failure, then we can help.
[18:10:26] <cheesepl_> I do but it's 50MB ;)
[18:10:47] <cheesepl_> https://gist.github.com/cheeseplus/5090943
[18:11:02] *** alfonsodev has joined #smartos
[18:11:16] <rmustacc> Okay.
[18:11:25] <rmustacc> Do you have the full line the compiler used?
[18:11:26] <cheesepl_> that is the head/tail of it, it fails to link against libv8 (although the libv8 gem installs fine on it's own)
[18:12:17] <rmustacc> It looks like someone isn't passing in -l lines to certain libraries.
[18:12:23] <rmustacc> Hence the missing things like printf.
[18:12:35] <rmustacc> But without seeing the full line that was used to compile it, it's hard to know why.
[18:13:07] <cheesepl_> rmustacc: I'll see if I can find that line
[18:14:24] <rmustacc> cheesepl_: Thanks.
[18:14:44] *** dap has joined #smartos
[18:14:46] <rmustacc> I'm not as familiar with building ruby gems, but I know how to build things in general, so hopefully I can provide some help.
[18:15:02] <cheesepl_> excellent, I'm in the same boat with gems
[18:16:18] <Alasdairr> ruby gems, python eggs, perl cpan modules and php pecl modules are cesspits of gccisms and assumptions
[18:16:46] <rmustacc> So it's like everything else then. ;)
[18:16:46] <Alasdairr> Like /bin/sh being bash and what not
[18:16:52] <Alasdairr> Well, pretty much :-)
[18:17:08] <jesse_> Alasdairr, I like to call those linuxisms=)
[18:17:26] <Alasdairr> well thankfully mac os x has helped a lot
[18:17:29] *** wolstena has quit IRC
[18:17:45] <Alasdairr> hipster devs tend to try to get stuff to work on their macbooks too
[18:17:58] <jesse_> ...never thought of that
[18:18:18] <Alasdairr> unfortunately /bin/sh on macos is bash
[18:18:19] <Alasdairr> :-)
[18:18:43] <jesse_> doesn't bash disable some features if it's invoked as 'sh'?
[18:19:04] <jesse_> (now, the other pain, #!/bin/bash ...)
[18:19:40] <rmustacc> Certainly not all of them.
[18:19:54] <rmustacc> A good example is the use of the local keyword in functions.
[18:24:54] <nahamu> It would be great if computers had an arm on them and bash when invoked as sh would still run correctly when you use bash-isms, but at least smack you for doing it.
[18:27:53] <jesse_> jperkin, omg, zsh in trunk? *wave* =)
[18:29:47] <ryancnelson> *mutters "hipster" under his breath*
[18:31:39] <jesse_> I thought bash was the hipster shell
[18:31:49] <jesse_> newer, bigger, with less features
[18:31:52] <jesse_> fixie of shells
[18:31:53] <cheesepl_> rmustacc: here is the makefile https://gist.github.com/cheeseplus/5092151
[18:32:01] <ryancnelson> bash is "newer"?
[18:32:02] <richlowe> zsh is the "used to be a hipster in 2001" shell
[18:32:09] <jperkin> bash is way too stable and old to be hipster
[18:32:12] <cheesepl_> ha, truth
[18:32:16] <jesse_> oh, I used it waaaaay before that...
[18:32:18] *** cheesepl_ is now known as cheeseplus
[18:32:21] * jesse_ is not helping himself
[18:32:32] <jperkin> surely there must be a haskell shell by now for hipsters
[18:32:47] <jperkin> completely useless, takes days to build and is a pig to debug, but hey, it's niche
[18:33:55] <ryancnelson> i'm sure, however, over in the #solaris channel, they're calling those crazy usb-booting, virtualizing kids in #smartos "hipsters", too :)
[18:34:04] <jesse_> zsh is just the best interactive shell. Shell scripts I write for /bin/sh
[18:39:13] *** alfonsodev has quit IRC
[18:39:16] *** cheeseplus has quit IRC
[18:39:39] <Alasdairr> The #solaris channel is trapped in corporate land where using anything other than RHEL or Solaris 8/9/10 is not gonna happen
[18:39:58] <Alasdairr> I also get the feeling a lot of corporate places are going Solaris 10 -> RHEL rather than be gouged by Oracle
[18:40:53] <Alasdairr> I hope one day SDC starts to penetrate the corporate environment
[18:41:31] *** cheeseplus has joined #smartos
[18:41:37] <ryancnelson> me too.
[18:43:00] <chris---> i just got pricing on a pair of half rack superclusters
[18:43:04] <chris---> the horror
[18:44:04] <nahamu> I live too far on the bleeding edge for SDC at the moment...
[18:44:57] <nahamu> I'm hoping SDC 7 will have enough of the SmartOS features and enough wiggle room for the crazy security stuff we do that I can stop using my own hacky management tools.
[18:46:08] <ryancnelson> i meet a TON of solaris people who like to architect rube-goldberg machines and thrive on the tweaker ndd recipes in text files they've been dragging around for decades.  No offense, but "wiggle room" == "unsupportable crazy pants"
[18:47:12] <ryancnelson> i'd rather open-source more chunks of sdc, and let you run with that on your own
[18:47:30] <Alasdairr> Can SDC do nic bonding now?
[18:47:33] <ryancnelson> yes
[18:47:36] <Alasdairr> yay
[18:47:41] <ryancnelson> i'm using it right now
[18:48:16] <Alasdairr> might have to re-evaulate it at some point
[18:49:06] <EMH_Mark3> rmustacc: was using dd with bs=1M to do reads, and was reading from a sparse file (so zfs/rawdisk performance isn't the issue)
[18:50:36] *** ipalreadytaken has quit IRC
[18:50:52] *** ipalreadytaken has joined #smartos
[18:51:43] <nahamu> ryancnelson: I don't mind it being unsupported. It's unsupported *now* :)
[18:52:04] <rmustacc> cheesepl_: The makfile itself isn't the most useful as I personally can't really evaluate that into the compile line.
[18:52:32] <cheeseplus> rmustacc: figured as much, trying to find where it actually writes/outputs that magic line
[18:52:41] <ryancnelson> nahamu: you don't get to see the fun support tickets...
[18:52:52] <nahamu> ryancnelson: touche'
[18:53:17] <ryancnelson> "i sewed the head of a cow onto my headnode… when i try to milk it, i'm not getting the performance i expect.  when can brendan fly here and look at it?"
[18:53:25] <rmustacc> cheeseplus: Try passing V=1
[18:53:30] <rmustacc> In the gmake line.
[18:53:36] <rmustacc> That's a semi-standard way.
[18:54:30] <nahamu> ryancnelson: The only reason they sewed the cow head onto the headnode in the first place was to get Brendan to visit. Everyone loves Brendan...
[18:55:19] <nahamu> But I completely understand your immediate response to my desire for enough rope to hang myself... :)
[18:55:35] <wesolows> having brendan visit fucked-up installs is fine as long as we're billing at his professional services rate of $100k a day
[18:56:19] <wesolows> un-hanging people who've hung themselves shouldn't be cheap
[18:58:38] <nahamu> wesolows: agreed.
[18:59:04] <cheeseplus> rmustacc: have to run now but will try after lunch, thanks for the pointers
[19:03:02] *** while1eq1_away has quit IRC
[19:13:28] *** KermitTheFragger has quit IRC
[19:16:00] *** nefilim has quit IRC
[19:16:17] *** nefilim has joined #smartos
[19:21:34] <MerlinDMC> hello
[19:22:40] <nahamu> hi MerlinDMC
[19:28:42] *** chh has joined #smartos
[19:31:54] *** wolstena has joined #smartos
[19:44:36] *** khushildep has quit IRC
[19:55:43] *** JT-EC has quit IRC
[19:55:53] *** JT-EC has joined #smartos
[19:57:14] *** noahmehl has joined #smartos
[19:58:01] <joshie> sigh, I come on irc to escape hipster talk and mutilated cow/server hybrids. Oh well
[19:58:03] *** bluezenix has quit IRC
[19:59:58] *** chorrell has quit IRC
[20:05:33] *** fsteinelX has joined #smartos
[20:09:35] *** ctable has joined #smartos
[20:10:34] *** pringlescan has joined #smartos
[20:11:40] <pringlescan> Hello all, you guys were very helpful yesterday with troubleshooting a Dell R710 server. I just got a new box (non-dell) and it's also rebooting after disk initialization. I'm just wondering, why do I have so much trouble with LSI RAID cards and normal hardware? Is this normal to have boots fail out of the box?
[20:12:01] <wesolows> well, it depends what the problem is.
[20:12:34] <wesolows> in general we discourage using HW RAID controllers, but we do have quite a few of the R710s in our data centres so obviously some of them do work
[20:13:16] <wesolows> if you want a system guaranteed to boot and work out of the box, see http://download.joyent.com/pub/manufacturing.  That's our manufacturing database, the instructions we give our own vendors to build our systems.  We know they work if you buy exactly that BOM.
[20:13:43] <wesolows> But more on point, since I don't know what problem you're having with this new system, it's hard to say.
[20:14:00] <wesolows> I think LeftWing is doing some work to make the SmartOS setup process easier to use and to debug when shit breaks.
[20:14:25] <pringlescan> I'm going to try with the -K switch, I was seeing if I could disable C-states on this machine. Why do you discourage hardware raid cards? because of ZFS?
[20:14:28] <richlowe> wesolows: haha
[20:14:59] <wesolows> Partly because their function is redundant, but also because ZFS is observable; firmware on an HBA is not.
[20:15:03] <chris---> are you doing raid with the lsi card or is it just passing the disks through
[20:15:15] <wesolows> We can debug and fix ZFS.  We cannot help at all if a RAID controller is not working.
[20:15:18] <pringlescan> chris—, 2 RAID 10s with 4 drives and 1 hot spare on each
[20:15:33] <pringlescan> (SSD + spinning disks)
[20:15:45] <chris---> stop that :)
[20:16:03] *** ryancnelson has quit IRC
[20:16:30] <pringlescan> I'm not opposed to being given suggestions; or even being told what to do, but unless there's a manpage for it, I need an explanation ;-)
[20:16:54] *** wramthun has quit IRC
[20:17:21] *** wramthun has joined #smartos
[20:19:10] <chris---> the installer will do single disk, mirror a pair, raidz if more than a pair. if you want fancier than that you have to do some work outside the installer
[20:19:35] <chris---> not sure i've run across docs for such things
[20:19:59] <pringlescan> so, you're saying ZFS is good enough to not use NV-ram or battery backup, so you shouldn't use a HW raid card?
[20:20:04] <chris---> but if your lsi is one that will let the drive passthrough if you dont configure raid, your in good shape
[20:20:09] <chris---> yes
[20:20:32] <pringlescan> I'm sure it will, but I need a convincing reason why, I would imagine performance is better with the card?
[20:20:58] <wesolows> pringlescan: correct
[20:21:02] <chris---> what wesolows said
[20:21:30] <wesolows> why would performance be better?  these ASICs have maybe a 533 MHz PPC CPU in them.
[20:21:32] <chris---> your not abstracted a layer away from it
[20:21:42] <wesolows> your sandy bridge CPU is probably 20x more powerful
[20:22:17] <pringlescan> so instead of doing a hot spare, I can use all of my drives with ZFS?...
[20:22:24] <chris---> you can hot spare in zfs
[20:22:27] <chris---> if you want
[20:22:32] <wesolows> you can, but the redundancy aspects aren't any different
[20:22:32] <chris---> just not out of the installer
[20:22:32] <pringlescan> just not from the installer?
[20:22:43] <wesolows> if you want a spare with HW RAID, you probably still want one with ZFS too
[20:23:01] <wesolows> hopefully that will change soon.  disklayout and mkzpool will create pools with spares by default.
[20:23:18] <pringlescan> the installer will help me set up a zfs pool (i don't know the lingo yet) with multiple disks?
[20:23:36] <wesolows> if by help you mean obstruct and generally annoy, yes
[20:23:43] <wesolows> but yes it will create a pool
[20:24:11] <chris---> i've done some work on making the installer better, but not ready to share
[20:24:18] <pringlescan> I have the open solaris book on my desk but I need to get these in a datacenter in about 48 hours… I'm willing to learn but I need to hit the ground running and hopefully use a sane config from the get go, or semi-sane.
[20:24:53] <chris---> you have 6 drives?
[20:26:22] <chris---> looks like the last stuff i shared is still an open pull request
[20:26:43] <wesolows> LeftWing is working in that area, and I pointed him at your work.
[20:26:53] <chris---> maybe i'll obsolite that before its pulled
[20:27:01] <wesolows> We're aware of it.
[20:27:16] <wesolows> apparently the entire setup script is basically crap though
[20:27:17] <chris---> yeah i know its been seen, not being critical
[20:27:21] <pringlescan> I have 10 drives, 5 SSDs, 5 spinning disks
[20:27:45] <pringlescan> if there are step-by-step instructions I can follow them I'm used to slackware and gentoo and legos
[20:27:53] <chris---> i dont know my way around git enough to know how to build on what i've done before its pulled in for sure
[20:27:57] <chris---> yet
[20:28:19] <pringlescan> did you do it on a feature branch?
[20:29:05] <chris---> i'm in no state to figure that out at the moment either :)
[20:29:18] <chris---> worry about your drives :)
[20:29:28] <wesolows> pringlescan: assuming your disks are visible (and not being glommed into some RAID array), you would just type in the names of the disks you want in the pool during setup.
[20:29:55] <chris---> suppose i should check with leftwing and see what they are doing
[20:30:01] <LeftWing> chris---: I've taken a look in there this week.
[20:30:09] <chris---> speaking of
[20:30:10] <LeftWing> It's not especially great.
[20:30:12] <wesolows> This is shitty, because they'll probably be WWNs.  However, it's worth noting that the SSDs will have different looking WWNs from the spinning disks, so you can use that to distinguish them.  This is because they'll have different vendor IDs
[20:30:22] <LeftWing> And it does annoying things, like: clearing the screen on a serial console.
[20:30:26] <LeftWing> (Which I've just fixed.)
[20:30:58] <pringlescan> :-( it looks like it doesn't support pass through and they have to been in an array. unless deleting all of the arrays exposes them as bare drives; which I can test.
[20:30:59] <chris---> i already dont like some of the formatting i did
[20:31:21] <chris---> that would be how it would pass through
[20:31:35] <chris---> no raid config, here are the drives
[20:31:42] <Forced> not always
[20:31:50] <Forced> I wish, though
[20:31:50] *** khushildep has joined #smartos
[20:31:51] <chris---> what i've seen from the lsi i have anyway
[20:32:11] <wesolows> there are so many different LSI ASICs and firmware loads that it's impossible to generalise.
[20:32:14] <Forced> I have some that don't let you do it all, and I have others that need to "specify" pass-through
[20:32:22] <Forced> so true
[20:32:27] <wesolows> At most, the behaviour of a specific PCI ID and revision may be known.
[20:32:34] <chris---> ah, so i've just played in a small pool then
[20:32:46] <rmustacc> Most Dell PERCs that have nvram do not.
[20:32:56] <pringlescan> This box has two E5-2670's and 128gb of ram, I can't wait to get it running
[20:33:04] <wesolows> I mean, you can get a 2308 with IT firmware on it from the factory, and that card will never do RAID of any kind.
[20:33:07] *** tonyarkles has quit IRC
[20:33:07] <pringlescan> rmustacc, this is a different box, that one is actually behaving relatively speaking
[20:33:34] <wesolows> pringlescan: if you have that kind of money, just go buy a working HBA and throw that RAID controller away.
[20:34:01] <wesolows> you can get one for $200 or less.  Trivial next to the cost of such a system.
[20:34:15] <wesolows> and next time just spec it with that instead, and you'll actually save money.
[20:34:21] <chris---> i have a pile of lsi sas 9211-8i cards, and thats how they behave
[20:34:32] <chris---> no raid? here are the drives
[20:34:41] <wesolows> those are IR by default
[20:34:42] <chris---> so i'll limit my comment to that then :)
[20:35:00] <wesolows> IR works by default, but it's still not as good as IT
[20:35:12] <chris---> they are the same as in my sunoracle boxes have in them
[20:35:26] <chris---> i havent tried to do IT fw
[20:35:27] *** wolstena has left #smartos
[20:35:32] <wesolows> Sun is very conflicted about RAID.
[20:36:01] <wesolows> They make a lot of money selling ZFS systems with PMC controllers in them, while their commodity server group keeps pushing RAID with GNU/Linux or Windows like it's 2002.
[20:36:02] <pringlescan> What's an HBA? It was grant money we had to use… I would have just used the R710
[20:36:27] <wesolows> an HBA is the thing that has PCIe on one end and SAS on the other
[20:36:45] <wesolows> yours also has a bunch of shit in the middle that you don't need :-)
[20:36:51] <Forced> http://en.wikipedia.org/wiki/Host_Bus_Adapter
[20:37:06] <pringlescan> blah, still getting kernel panics, I see ata_attach and pciide_init, etc so I guess it's storage related
[20:37:26] <wesolows> kernel panics are not exactly common
[20:37:34] <pringlescan> for everyone but me
[20:37:43] <wesolows> is there a stack trace?
[20:37:53] <pringlescan> not that I can dump, I can take a photo of what I can see
[20:38:07] <pringlescan> I'm in kmdb can I do something from there?
[20:38:12] <wesolows> yes.  $C
[20:38:27] <pringlescan> ok what am I looking for
[20:38:32] <pringlescan> it's pretty long
[20:38:39] <wesolows> what's below cmntrap?
[20:39:56] <pringlescan> after the debug_enter and trap+ it refers to apix_intx_enable, *_enable_vector, _setup_io_intr, apix_addspl, apix_add_avintr, and some pci_enable
[20:40:05] <pringlescan> wildcard being apix
[20:40:46] <pringlescan> futher down I see pci-ide`pciide_intr_ops  a bunch of ata and ddi entries after that
[20:40:48] <wesolows> rmustacc will be interested in that
[20:41:04] <wesolows> have you enabled the X2APIC on this system?
[20:41:10] <pringlescan> I can take it in two photos
[20:41:35] *** sorgenti has joined #smartos
[20:41:36] <pringlescan> it's as configured from the vendor, I'm unsure, I booted with defaults and a -k flag
[20:42:24] <pringlescan> I have power savings turned off though… I didn't see any APIC options
[20:42:46] *** denizr has quit IRC
[20:44:53] <richlowe> That's the same crash as trentster's
[20:45:18] <wesolows> APIC != ACPI
[20:45:38] <wesolows> the confusing similarity (and brokenness of both) is sad
[20:45:39] <pringlescan> Advanced Power Interrupt Controller? just a guess
[20:45:44] <wesolows> programmable
[20:45:45] <pringlescan> what happened to APM?
[20:45:45] <richlowe> Programmable
[20:45:55] <pringlescan> Kidding, don't answer that.
[20:45:57] <richlowe> The full panic stack would let me bucket it at "same as trentster's"
[20:46:05] <wesolows> APM lives on in the form of the SMM trigger I/O port
[20:46:08] <pringlescan> I'm getting that ready Richlowe
[20:46:57] <richlowe> maybe two victims will wake robert up, and he can be confused instead of me :)
[20:47:59] <richlowe> wesolows: you guys have a dump device by default, right?
[20:48:11] <richlowe> so pringlescan can just let it dump, then savecore and cut/paste from mdb, rather than uploading photos?
[20:48:43] <wesolows> yes, we should, if he's gotten far enough
[20:49:23] <pringlescan> how do I do that?
[20:49:27] <pringlescan> mdb?
[20:49:37] <wesolows> you qould $q to make it dump and reboot
[20:49:39] <pringlescan> I still have the kernel debugger open
[20:49:40] <wesolows> *would
[20:49:44] <pringlescan> where will it save to?
[20:49:48] <pringlescan> it said no dump device
[20:49:53] <richlowe> drat
[20:50:00] <wesolows> then it won't, because you're not set up or not booted far enough
[20:50:01] <pringlescan> it's said no dump device on both systems
[20:50:18] <pringlescan> can I set it up manually or something preferable to photos?
[20:50:33] <richlowe> well, what was happening with trentster is we're trying to setup interrupts on two pci-ide devices at once, and one just crapped on the other (probably)
[20:50:54] <richlowe> the panic-stack is probably good enough to say it's at least similar, ::stacks -c ghd_register | ::findstack -v should be enough to be reasonably sure it's the same
[20:51:27] <pringlescan> there is a ghd in there I saw
[20:51:59] <rmustacc> richlowe: I'll try and spend some time so I can be confused with you.
[20:52:11] <pringlescan> ata`ghd_register+0x178(fffffffff7faa4b3, ffffff21ecad8d98, ffffff21d35712a8, 0, ffffff21ecad8c80, ffffffff7fa86e0)
[20:52:23] <pringlescan> I'm not guarenteeing the number of f's but other things should be correct
[20:52:35] <richlowe> the command line I mention above should show you two threads under there, if so it is the same as trentster's almost certainly
[20:52:36] *** fsteinelX has quit IRC
[20:52:38] <richlowe> and we have his dump
[20:52:47] <richlowe> (but I, at least, am not quite sure _why_ we're screwing up, just vaguely sure of how)
[20:54:14] <pringlescan> richlowe, that spit out a bunch
[20:54:21] <pringlescan> it looks like the same output as $c
[20:55:04] <pringlescan> photos are uploading
[20:55:30] <pringlescan> http://imgur.com/a/Sj3Av
[20:57:33] <pringlescan> given that, are there any bios settings I could possibly adjust? I'm thinking of turning the onboard SATA off if it's not, I hoped the vendor would but they didn't and I don't have an optical drive
[20:59:56] <richlowe> in grub, edit the kernel$ line
[20:59:59] <richlowe> add -Bdisable-apix=true
[21:00:08] <richlowe> should work around it
[21:00:24] <pringlescan> is that bad to use permanently?
[21:01:47] *** Cpt-Oblivious has quit IRC
[21:01:58] <pringlescan> so ZFS wouldn't even like a RAID controller because it can't work right on one disk
[21:02:09] *** khushildep has quit IRC
[21:02:45] <pringlescan> richlowe, I'm trying that now
[21:04:03] <pringlescan> I feel like I'm going to end up fsck'd for running this in production, for not only a lack of knowledge, but hardware that will only boot begrudgingly
[21:05:02] <pringlescan> the setup can't see any of my disks when they're not in arrays, and I didn't see any options in the LSI web bios to enable pass through :-(
[21:05:08] *** tonyarkles has joined #smartos
[21:05:22] <Forced> what LSI card do you have?
[21:07:30] <pringlescan> Forced, checking doesn't say on the invoice
[21:07:57] <pringlescan> Forced… 9271-8i
[21:08:02] *** alfonsodev has joined #smartos
[21:08:38] <pringlescan> with a BBU
[21:10:56] <Forced> darn, I'm not seeing any mention of pass-through in any form :(
[21:11:24] <pringlescan> so if I can't benefit from ZFS in this respect, I basically lose the benefits  other than hot snapshots, right?
[21:11:41] <Forced> well I think it has other features
[21:11:56] <Forced> but these guys can tell you more, I don't use it much
[21:13:13] <szaydel> pringlescan: Generally speaking, I think with ZFS you want as little abstraction between the disks and the filesystem as possible. I am sure I am stating the obvious.
[21:13:54] <Forced> looks like you would lose the "self healing"
[21:13:58] <Forced> http://en.wikipedia.org/wiki/Zfs
[21:15:07] <wesolows> you lose all observability into the state of disks
[21:15:16] <Forced> that too....
[21:15:17] <wesolows> you lose all observability into the operation of RAID
[21:15:32] *** denizr has joined #smartos
[21:15:37] <wesolows> you lose all ability to recover from corruption affecting a single disk
[21:15:51] <pringlescan> at the ZFS level, I have to let my RAID card do all of that
[21:15:55] <e^ipi> mpt cards are literally $10
[21:16:14] <wesolows> you will be unable to use the system tools such as zfs replace and fmadm and will instead have to use binary-only (and inferior, frankly) tools from LSI to manage your system
[21:16:19] <e^ipi> and they work *awesome*
[21:16:27] <wesolows> e^ipi: but you don't want mpt.  You want mptsas.
[21:16:29] <szaydel> It almost seems like combining RAID cards with ZFS actually reduces significantly benefits of both.
[21:16:35] <wesolows> mpt is pure evil.  just ask Greg Price.
[21:16:37] <szaydel> At least this has been my perception.
[21:17:01] <pringlescan> Is it worth losing major face to not use the RAID card and BBU? I know it's "fundamentally wrong" but...
[21:17:08] <e^ipi> wesolows: right, they're still $10-20
[21:17:23] <e^ipi> some cheezy sas1068e card from ebay.
[21:17:28] <wesolows> there is exactly one benefit to HW RAID, and it applies only if you use RAID5 or RAID6 and care about getting large numbers of low-latency small-block write IOPS without a slog.
[21:17:37] <wesolows> 1068e is mpt.  evil.
[21:17:42] <wesolows> 9211-8i.
[21:17:53] <wesolows> they're not $10, but they work a lot better
[21:17:54] <pringlescan> I thought RAID10 was the best for databases… but maybe I misread or 0+1
[21:18:00] <wesolows> pringlescan: exactly.
[21:18:02] <chris---> i think i paid $160 for my 9211-8i
[21:18:11] <wesolows> which is why there's absolutely no benefit to HW RAID for you.
[21:18:15] <wesolows> it's strictly worse.
[21:18:33] <pringlescan> all the books on DBA I read said to have a battery backed RAID card...
[21:18:54] <wesolows> that applies if you are on a system without consistent filesystems.  Such as Windows or GNU/Linux.
[21:18:57] <pringlescan> but in a datacenter with UPS in a server with redundant power supplies… and ZFS it seems pointless
[21:19:05] <szaydel> pringlescan: mirroring, which delivers more spindles is good for databases, to be sure. This will change to a degree with fast SSDs.
[21:19:09] <wesolows> ZFS is consistent without a UPS or battery.
[21:19:14] <pringlescan> onboard sata is worse than onboard raid right?
[21:19:20] <pringlescan> ***the LSI raid
[21:19:53] <wesolows> worse in what way?  It's better for observability, but probably slower and certainly less scalable.  It also forces you to use SATA disks, which are never better than and rarely as good as SAS disks.
[21:20:06] <wesolows> most AHCI controllers also have dubious hotplug support.
[21:20:12] <wesolows> or none at all.
[21:20:15] *** ipalreadytaken has quit IRC
[21:20:35] <pringlescan> I didn't have a big storage budget, the disks I looked at that came in SAS and SATA, the SATA performed better in 9/10 benchmarks, but it was just an entry level Seagate drive
[21:20:49] <wesolows> I should just create a charitable foundation that ships someone an SMCI 2308-IT card each day.
[21:21:09] <wesolows> performed better according to what?
[21:21:25] <Miraz> have you tried comparing a database performance on ZFS whilst using an SSD Zil?
[21:21:28] <pringlescan> the standard battery of disk benchmarks you'd expect from a tom's hardware style benchmark
[21:21:39] <pringlescan> Zil?
[21:22:28] <pringlescan> http://www.supermicro.com/products/motherboard/xeon/c600/x9dri-f.cfm that's my board
[21:22:33] <chris---> sometimes i wonder if the docs for dbas assume your working on stuff in your garage
[21:22:55] <nahamu> wesolows: they make a standalone card?
[21:23:12] <pringlescan> it says it has AHCI SATA and SCU SATA
[21:23:14] <Miraz> You can designate disks in your zfs pool to act as a write cache - so if you add an ssd (or two) then zfs will use them to cache writes to the normal disks
[21:23:15] <chris---> they were completely sold that ASM was a must for single nodes, cause it did away with hot spots! and they were on luns from an hitachi uspv
[21:23:34] <chris---> "this is all made up, move along"
[21:24:40] <pringlescan> I've never used it, but I'm guessing IPMI with KVM support would help with kernel dumps? or serial access
[21:24:43] <Miraz> Zil = zfs intent log
[21:25:05] <wesolows> pringlescan: you know, if you'd gotten the x9drd-7jln4f instead, you'd have an onboard 2308 with IT firmware, and this would be Just Working
[21:25:33] <pringlescan> I didn't pick it, I was told on IRC being capable of building systems didn't mean I could stick hardware into a rackmount chassis and make a server
[21:25:34] <wesolows> the SCU is completely useless; we have no driver for it and it appears to be a dead-end in terms of its future.
[21:25:54] <pringlescan> this is a super micro, I'm going to have to disagree, I could of slapped drives and ram in this… but I couldn't have gotten wholesale pricing
[21:25:54] <wesolows> heh
[21:26:29] <wesolows> I'm just noting that you are a couple of characters away from using the exact systems Joyent uses.
[21:27:00] <wesolows> and if you were using those, this would all work perfectly with no futzing.
[21:27:44] <pringlescan> I might try to return the LSI card… it's bent...
[21:27:51] <pringlescan> i know it works, but it's clearly bent
[21:27:54] <wesolows> sounds like that's probably the issue. ;-)
[21:28:11] <wesolows> maybe they'll give you an AOC-S2308L-L8E instead if you ask nicely
[21:28:18] <wesolows> (that's what you want, by the way)
[21:28:49] <pringlescan> isn't that way cheaper than the LSI card?
[21:29:03] <wesolows> it's a card made by SMCI with an LSI ASIC on it
[21:29:11] <wesolows> let me see what a decent price is...
[21:29:25] <wesolows> $175
[21:29:35] <wesolows> probably less than half what the 9271 costs
[21:29:48] <pringlescan> so they might swap it out… brb
[21:29:53] <wesolows> they just might.
[21:32:20] <szaydel> pringlescan: I think one reason that might dissuade one from going with SMCI vs. LSI is the support. They do not have world class support, and while LSI is not best that there is, I think they are much better. It may not be worth it in your case, but if you are going to buy hundreds, it should factor in.
[21:32:21] <pringlescan> and you guys think my system would boot with that board?
[21:32:38] <pringlescan> nah, just one...
[21:39:50] *** porkbelt has quit IRC
[21:42:59] <pringlescan> performance-wise is zfs with an hba going to work faster than linux on a hw raid for databases?
[21:49:40] *** bluezenix has joined #smartos
[21:51:11] *** rbrown_ has joined #smartos
[21:52:56] <szaydel> pringlescan: I think the answer is it depends on a lot of things.
[21:53:16] <szaydel> ZFS is not the fastest thing, unless you design it to be fast.
[21:53:34] <szaydel> You can achieve same with a much more reckless EXT3 for example.
[21:53:48] <szaydel> Disable barriers, etc. and pert. will be awesome.
[21:54:24] <szaydel> There are tradeoffs however to think about, i.e. data consistency, reliability, atomicity of transactions, caching, etc.
[21:55:03] <szaydel> Again, some of these factors may come at a price that you will not pay, seemingly with other approaches, but at a higher risk, potentially.
[21:57:34] *** stetho has quit IRC
[22:00:27] <wesolows> szaydel: SMCI's support is poor, to be sure, but LSI is not going to do shit for you unless you are using their drivers on one of their supported OSs.
[22:02:20] <szaydel> wesolows: I think it really depends when it comes to LSI. For companies that use a lot of their gear, like we do, since we sell our systems with their components, I think they are much more eager to help. Though in other cases they may be rigid yes. I still think even with their rigidity they are better than SM.
[22:02:27] *** ipalreadytaken has joined #smartos
[22:02:35] <pringlescan> i'm never going to hear the end of it if I switch this card out probably
[22:02:45] <pringlescan> except I can blame it on the fact that it's bent
[22:02:55] <wesolows> We got absolutely no help from them at Sun, when we were moving thousands of their ASICs a year.
[22:03:19] <szaydel> wesolows: That's disappointing to hear.
[22:03:40] <chris---> "card was cattywampus, had to go"
[22:03:53] <szaydel> Though, what seemed to be true also is that Sun was not great about engaging with other companies at the right levels.
[22:04:02] <szaydel> At least, that was my experience while at USPS.
[22:04:11] <szaydel> And we had a shit-ton of Sun gear.
[22:04:48] <pringlescan> If you pay $10k for a server, you don't expect a card to be in there crooked, they knew it was going ground from CA to PA, is it unreasonable to ask for a replacement?
[22:05:51] <pringlescan> naturally nobody is picking up, they're probably all busy playing simcity
[22:06:15] *** gkyildirim has joined #smartos
[22:06:20] <szaydel> pringlescan: More than reasonable. I would say this is not acceptable at all. We are a small company and we never think twice when something we ship arrives in this manner at the customer's footstep.
[22:08:23] <pringlescan> It's net 30 so they have to fix it anyway I suppose if they want to get paid
[22:11:48] <szaydel> That's the right way of thinking pringlescan.
[22:12:53] <pringlescan> Sarcasm? They were out to lunch but called back, so that was nice. He said I definitely wanted to do an HBA with ZFS and to send a photo along, so they seem willing to work with me.
[22:13:54] <szaydel> pringlescan: No sarcasm. At all.
[22:14:26] <gkyildirim> Hi. Is there a package for zabbix-*? I'd ask for joyent zone.
[22:15:07] <pringlescan> Maybe I can get them to send me two HBAs and I can replace the PERC too… it should still be less than the one LSI
[22:15:10] *** while1eq1_away has joined #smartos
[22:17:22] <pringlescan> how is it that HBA differs from an internal port? is the hba directly to the pci express bus?
[22:18:21] <while1eq1_away> hello, im on my headnode but when I go to 'imgadm' it seems the command is missing...
[22:18:47] <szaydel> pringlescan: It is PCIe I believe.
[22:19:28] <pringlescan> PCIe = PCI Express, no?
[22:26:58] *** densone has joined #smartos
[22:27:34] <densone> Someone recently sent me a link to an smf generator and I forgot what it was called.
[22:27:50] <wesolows> smfgen
[22:27:59] <densone> hrm
[22:28:00] *** while1eq1_away has quit IRC
[22:28:08] <densone> It had a catchier name.
[22:30:47] <nahamu> manifold
[22:31:15] <brendang> pringlescan: if you are benchmarking ZFS (mentioned earlier I think), then fio might be worth a try. I've been using it a lot recently on SmartOS.
[22:31:31] <brendang> for exmaple: ./fio --runtime=60 --time_based --clocksource=clock_gettime --name=randread --numjobs=8 --rw=randread --random_distribution=pareto:0.9 --bs=8k --size=10g --filename=fio.tmp  # file system random I/O, 10 Gbytes, 8 threads
[22:31:49] *** ryancnelson has joined #smartos
[22:32:08] <szaydel> brendang: Do you prefer it to vdbench, or otherwise?
[22:32:12] <brendang> what's neat about it is that it can do non-uniform random distributions. that fio command was based on profiling real world cloud databases.
[22:32:16] <brendang> szaydel: yes
[22:32:26] *** while1eq1_away has joined #smartos
[22:32:32] <szaydel> Very interesting.
[22:32:44] *** rbrown_ has quit IRC
[22:32:47] <pringlescan> brendang, right now I'm at the kernel panic stage, I'm excited to reach the ZFS benchmarking stage ;-)
[22:33:02] <brendang> in fact, if anyone was interested, here's a set I have been using. I should make this into a blog post I guess:
[22:33:05] <brendang> ./fio --runtime=300 --time_based --clocksource=clock_gettime --name=randread --numjobs=8 --rw=randread --random_distribution=pareto:0.9 --bs=8k --size=10g --filename=fio.tmp  # throw-away: 5 min warm-up
[22:33:09] <brendang> ./fio --runtime=60 --time_based --clocksource=clock_gettime --name=randread --numjobs=1 --rw=randread --random_distribution=pareto:0.9 --bs=8k --size=10g --filename=fio.tmp  # file system random I/O, 10 Gbytes, 1 thread
[22:33:14] <brendang> ./fio --runtime=60 --time_based --clocksource=clock_gettime --name=randread --numjobs=8 --rw=randread --random_distribution=pareto:0.9 --bs=8k --size=10g --filename=fio.tmp  # file system random I/O, 10 Gbytes, 8 threads
[22:33:18] <szaydel> +1 about blog post!!!
[22:33:32] <nahamu> +1
[22:33:48] <wesolows> ARC test?
[22:35:04] <jesse_> pkgsrc needs a fio package first?=)
[22:35:05] <brendang> the "--clocksource=clock_gettime" is a work around for a solaris bug, which I spoke oven twitter with the author (Jens Axobe) about
[22:35:27] <pringlescan> brendang, does that relate to the high resolution timer bug?
[22:35:27] <brendang> yes, this is ultimately an ARC test, but in a realistic fashion.  the parety distribution resembles inverse exponential
[22:35:40] <brendang> so you get some disk I/O, but a higher ARC hit rate
[22:35:51] <brendang> s/parety/pareto
[22:36:23] <brendang> the parameter (0.9) was picked by visual comparison to dtrace-based working set analysis on a real database.
[22:36:50] <while1eq1_away> brendang: What do you reccomend for benching the CPU?
[22:37:55] *** bens1 has quit IRC
[22:38:29] <brendang> while1eq1_away: I start by rolling my own nop-loop assembly to get a baseline. then switch to more memory I/O intensive workloads. depends if you want single or multithreaded.  sysbench or 7zip are probably ok
[22:38:42] <while1eq1_away> thanks
[22:39:07] <pringlescan> is unixbench still relevant as a system benchmark?
[22:39:30] <brendang> not really
[22:39:33] *** scsbcn38 has joined #smartos
[22:39:45] <scsbcn38> good night
[22:39:46] <brendang> I still have the byte magazine where unixbench was introduced loaded in a tab somewhere
[22:40:04] <pringlescan> wasn't byte a commodore magazine? o_O
[22:40:17] <ryancnelson> oh, you crazy kids
[22:40:37] <scsbcn38> anybody knows can store /etc/system.conf parameters in smartos? tnx
[22:40:49] <ryancnelson> no, you can't.
[22:41:24] <ryancnelson> you'll need to change the platform to make that persistent.
[22:41:31] <scsbcn38> buah
[22:41:33] <ryancnelson> why are you changing /etc/system.conf?
[22:41:53] <scsbcn38> for make persistent some parameters for zfs like arc and others
[22:42:00] <brendang> http://archive.org/details/byte-magazine - love the ads...
[22:42:46] <szaydel> scsbcn38: Just create a transient service that runs on startup.
[22:42:57] <szaydel> Well, for some things it might be too late.
[22:43:47] <pringlescan> so just a recap, I need HBA cards, I have a 12 port sata backplane that goes to two SAS it looks like, 4 SSDs, 4 HDDs and 1 hot spare for each, what's going to have the best hardware support and use the best driver… I keep hearing about IT firmware
[22:43:57] <scsbcn38> example: set zfs:zfs_arc_max = 32212254720
[22:44:06] <rmustacc> Why do you need to set max arc?
[22:44:22] <scsbcn38> I think that is not possible in transicient service no?
[22:44:32] <rmustacc> I interpret that it's a bug that it's not releasing arc correctly.
[22:44:33] <scsbcn38> I want limit max arc to 1GB
[22:44:36] <rmustacc> Why?
[22:44:52] <rmustacc> You want to have unused dram?
[22:45:06] <ryancnelson> you hate read performance?
[22:45:30] <scsbcn38> I hate it ;) .
[22:45:31] <rmustacc> If the problem is that zfs isn't releasing arc correctly on SmartOS, then we care about that.
[22:45:34] <szaydel> ryancnelson: That's harsh!
[22:45:36] <nahamu> there is still a bug somewhere. I haven't filed a report yet, but if you do something that really fills up the ARC, sometimes SmartOS will refuse to create a new VM.
[22:45:48] <rmustacc> That sounds like a bug with vmadm.
[22:45:51] <nahamu> rmustacc: yes
[22:45:56] <rmustacc> Not a reason to restrict your arc.
[22:46:06] <nahamu> if the VM already exists, it will launch just fine.
[22:47:17] <scsbcn38> thanks all for your comments
[22:47:17] <trentster> nahamu: I concur have seen that bug as well it rears its head every now and again and refuses to create vm's. If you wait for 20 minutes and try again it generally works.
[22:49:42] *** tonyarkles has quit IRC
[22:52:34] *** densone has quit IRC
[22:54:34] *** scsbcn38 has quit IRC
[22:54:35] *** ChrisPartridge has quit IRC
[22:55:26] *** wolfeidau has quit IRC
[22:57:25] *** densone has joined #smartos
[22:57:48] *** ira has joined #smartos
[22:58:40] <nahamu> I just use the workaround perl invocation from the bug report on illumos.org
[22:59:15] *** gkyildirim has left #smartos
[23:03:28] *** sorgenti has quit IRC
[23:05:23] *** bluezenix has quit IRC
[23:06:10] <nahamu> (this one https://www.illumos.org/issues/1974 )
[23:07:14] <nahamu> rmustacc: Is that bug technically fixed in SmartOS, but the patch hasn't been pushed back up to illumos?
[23:07:31] <pringlescan> This card was suggested: AOC-S2308L-L8E but they're not very common, any other suggestions?
[23:07:34] <rmustacc> I don't know.
[23:10:08] *** CarlosC has quit IRC
[23:12:06] <pringlescan> Alright everyone, thanks for the help, I'm going to try to cross-ship an HBA, wish me luck.
[23:12:09] *** pringlescan has quit IRC
[23:14:12] *** wolfeidau has joined #smartos
[23:16:59] *** sorgenti has joined #smartos
[23:24:48] *** bluezenix has joined #smartos
[23:28:41] *** ipalreadytaken has quit IRC
[23:32:35] *** chh has quit IRC
[23:37:32] *** robinbow1s is now known as robinbowes
[23:44:07] *** while1eq1_away has quit IRC
[23:45:20] *** densone has quit IRC
[23:45:41] *** ctable has quit IRC
[23:47:02] *** tonyarkles has joined #smartos
[23:47:12] *** denizr has quit IRC
[23:50:04] *** des2 has joined #smartos
top

   March 5, 2013  
< | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | >