**** BEGIN LOGGING AT Fri Jan 26 17:07:02 2007 Jan 26 16:50:06 benr's uploading the slides for tonights talk Jan 26 16:51:53 hahahaha Jan 26 16:52:01 brendan gregg you nut. Jan 26 16:52:06 roflao Jan 26 16:52:19 okok I'm calling Jan 26 16:52:20 "Hi GREG!!!" Jan 26 16:52:30 'I'm brendan gregg and i'm an alcoholic' Jan 26 16:52:44 [the group is doing introductions] Jan 26 16:52:48 It's a holiday here... I must have better things to do on a beautiful friday afternoon Jan 26 16:53:05 * boyd wonders if he will just be able to read the gman transcript :) Jan 26 16:53:55 all sorts of dudes present - from sun seems to be josh berkus, steve lau, michelle olson, alan dubuff, alan burlison, gary pennington, .. Jan 26 16:54:10 george wilson from zfs Jan 26 16:54:15 [if this isn't useful, i can stop] Jan 26 16:54:30 I thought stevel was from Singapore Jan 26 16:54:32 mark mabeal [?] Jan 26 16:54:45 boyd, no, he lives in sf with 2 beagles Jan 26 16:54:50 [and obviously his wife] Jan 26 16:54:57 .. and a ruined piano Jan 26 16:55:01 :) Jan 26 16:55:03 haha Jan 26 16:56:05 there's no way to reduce java's ram usage, is there? I still don't quite get why firefox can manage to run in 30 megs, yet sun studio takes 370mb with no loaded projects/files, and another instance of java that apparently starts by default takes up another 80mb of ram. Jan 26 16:56:29 you can control the heap size Jan 26 16:56:55 will that actually help? Jan 26 16:57:22 well most everything lives in the heap afaik Jan 26 16:57:40 the main reason for the mem hunger of java is that it starts for every thing a new vm Jan 26 16:58:15 I never understood why it's made that way Jan 26 16:58:23 Auralis: still though, why does NetBeans take up 3x the ram of VS2005 to display an empty window? Jan 26 16:58:30 I just don't get whats actually *in* that ram Jan 26 16:58:49 magic beans Jan 26 16:58:50 did they ever give out a url for the slides? Jan 26 16:59:12 i missed it if they did (but them i'm doing a few other things too) Jan 26 16:59:20 hrm, benr didn't upload the slides Jan 26 17:00:30 We need a live IRCer at the meeting Jan 26 17:00:36 yeah Jan 26 17:00:47 joyent, widely considered ruby on rails experts Jan 26 17:00:55 everyone is currently a director at the company Jan 26 17:01:04 That' Jan 26 17:01:13 That's so they don't need to pay dental :) Jan 26 17:01:13 involved in both nevada and opensolaris [community] Jan 26 17:01:28 incubator for innovation - try to get smart people working with smart people Jan 26 17:01:42 try to be open with secrets Jan 26 17:02:07 providing 5 products - shared business hosting, joint connector [web 2.0], accelerator hosting [containers] and something else i can't remember Jan 26 17:02:15 * yongsun__ (n=yongsun@219.239.107.200) has joined #opensolaris Jan 26 17:02:17 haha Jan 26 17:02:35 shared business hosting on dell machines [boos], ufs Jan 26 17:03:14 aww, octave won't build with studio Jan 26 17:03:15 :( Jan 26 17:04:01 lots of traditional rules of an isp are being broken because they want customers to run fun and cool stuff Jan 26 17:04:20 joyent connector, web 2.0 shared calendaring, email and contacts Jan 26 17:04:22 e^ipi: Sparc or x86? Jan 26 17:04:53 * boyd notes that e^ipi has a shorter nick every time he looks Jan 26 17:05:01 implemented in ruby on rails, currently deployed on 5 containers [zones] on 4 systems 2 t1000's, 2 x4100's Jan 26 17:05:06 x86 Jan 26 17:05:17 nothing is done in the global zone Jan 26 17:05:18 Gman: hows the shared calendaring done? Jan 26 17:05:19 it's not going to get shorter than this Jan 26 17:05:25 soon it'll be -1 Jan 26 17:05:29 heh Jan 26 17:05:30 dlg, dunno :) Jan 26 17:05:33 You mean i Jan 26 17:05:42 err 1 Jan 26 17:05:44 that's root(-1) Jan 26 17:06:09 * Chile` has quit (Remote closed the connection) Jan 26 17:06:10 Yeah, but I figure the distinction:pixel ratio is better Jan 26 17:06:21 and then i'll just be . Jan 26 17:06:24 why not nfsv4 - no big community adoption right now, only one guy blogging away, no obvious advantages, nfsv3 is well understood Jan 26 17:06:43 maybe i should move this onto another channel ;) Jan 26 17:06:44 Interesting... I thought it benchmarked better Jan 26 17:06:51 i thought so too Jan 26 17:06:58 let's try cranking down these optimizations Jan 26 17:07:07 #opensolaris-ossvug Jan 26 17:07:08 of course i can understand the troubleshooting Jan 26 17:07:31 i stumbled upon /var/run/nfs4_domain (or whatever it's called) Jan 26 17:08:27 but, you'd think just the inclusion of compound operations would help a lot Jan 26 17:08:53 and delegations Jan 26 17:09:00 * yongsun_ has quit (Connection reset by peer) Jan 26 17:11:07 i wonder if we should ask them to repeat the audience questions Jan 26 17:11:35 Do you have a way to ask that? Jan 26 17:11:50 other than interrupting, no Jan 26 17:12:14 Live semi-transcripting in #opensolaris-ossvug from Gman Jan 26 17:07:02 * Now talking on #opensolaris-ossvug Jan 26 17:07:02 * brown.freenode.net sets mode +n #opensolaris-ossvug Jan 26 17:07:02 * brown.freenode.net sets mode +s #opensolaris-ossvug Jan 26 17:07:19 * boyd (i=brontita@ppp157-138.static.internode.on.net) has joined #opensolaris-ossvug Jan 26 17:08:17 strong space Jan 26 17:08:21 one of 2 storage products Jan 26 17:08:27 global storage, accessible anywhere Jan 26 17:08:30 scp, https, Jan 26 17:08:35 file browser [ruby on rails] Jan 26 17:08:39 suited for backup and archiving Jan 26 17:08:46 buy a strong space account, start an rsync Jan 26 17:08:51 5gb to 100gb Jan 26 17:08:55 willing to give you whatever you want Jan 26 17:08:58 container on a thumper Jan 26 17:09:08 so much storage available - happy to give 1tb of storage Jan 26 17:09:12 one command typed and done! Jan 26 17:09:26 bingo disk, other storage product Jan 26 17:09:32 collaborative web storage Jan 26 17:09:33 webdav Jan 26 17:09:39 store you stuff in different ways Jan 26 17:09:47 native support in almost all file browsers Jan 26 17:09:53 osx, windows, nautilus, konquerer, etc.. Jan 26 17:10:09 25gb to 100gb accounts Jan 26 17:10:23 container on a thumper Jan 26 17:10:25 same machine! Jan 26 17:10:31 Heh Jan 26 17:10:35 accellerators - interesting product, zones Jan 26 17:10:38 on demand containers Jan 26 17:10:53 struggles with what to call it - virtual dedicated? Jan 26 17:11:15 standard solaris containers - sell them in flavours Jan 26 17:11:24 1/4, 1/8, 1/16th Jan 26 17:11:30 capacity on demand Jan 26 17:11:38 implemented on sunfire 4100's Jan 26 17:11:45 architecture Jan 26 17:11:51 2 datacenters that you'll be interested in Jan 26 17:12:02 level 3 facilities - one northern california, one southern Jan 26 17:12:08 tier 1 backbone provider Jan 26 17:12:25 internet connectivity is absolutely phenominal Jan 26 17:12:40 * jbk (i=jason@highhopes.madhack.com) has joined #opensolaris-ossvug Jan 26 17:12:56 unfortunately level 3 has a lot of power problems :) Jan 26 17:13:03 switching env is copper based ethernet Jan 26 17:13:08 deploy on 3 standardized configs Jan 26 17:13:42 [insert something about sunfire and ruby on rails not having great threading] Jan 26 17:13:50 deploying on t1000 requires a lot of forethought Jan 26 17:14:03 people just want it to work really really fast Jan 26 17:14:12 notice application is dogging because app isn't threaded Jan 26 17:14:20 not necessarily ready to plan for this forethought Jan 26 17:14:21 heh Jan 26 17:14:33 were planning on selling niagara containers Jan 26 17:14:34 RoR has only Ruby's green thread. Not OS threads. Jan 26 17:14:39 had trouble early on, and put plans on hold Jan 26 17:14:53 sunfire x4100 Jan 26 17:14:59 predominant server atm Jan 26 17:15:09 [insert hardware stuff] Jan 26 17:15:14 excellent server, love it Jan 26 17:15:32 favourite server/storage for web 2.0 server is thumper x4500 Jan 26 17:15:34 max configuration Jan 26 17:15:53 useable capacity of 18tb Jan 26 17:15:58 flexible/wonderful Jan 26 17:16:05 in some sort of raid config [missed that bit] Jan 26 17:16:17 raid-z6 Jan 26 17:16:20 lifetime management is why they've gone for sun Jan 26 17:16:29 excellent onboard networking Jan 26 17:16:36 4 gb ethernet on board the machine Jan 26 17:16:52 lots of advantages to having a single hardware/software vendor Jan 26 17:17:01 networking: Jan 26 17:17:03 as long as it's not IBM :) Jan 26 17:17:07 need to be creative! Jan 26 17:17:17 1 of 4 is public [1gb is more than you need!] Jan 26 17:17:19 1 is private Jan 26 17:17:26 * Auralis (n=barbie@port-212-202-201-12.dynamic.qsc.de) has joined #opensolaris-ossvug Jan 26 17:17:37 aggregate [thanks to dladm] 2 ports for 2gb storage network Jan 26 17:17:47 os is standardized on nevada b43 Jan 26 17:17:50 need stability Jan 26 17:18:00 [need something to hold on for quite some time] Jan 26 17:18:02 quite stable Jan 26 17:18:07 handful of minor problems with b43 Jan 26 17:18:14 biggest problem due to a single character typo Jan 26 17:18:17 Heh... show me a linux distro where they pre-releases are chosen because they "need stability" Jan 26 17:18:17 cost us a lot of money! Jan 26 17:18:28 all systems are jumpstarted Jan 26 17:18:49 [wasn't a semi-colon, colon sockfs, or something] Jan 26 17:18:56 shows up as nfs problem seemingly Jan 26 17:19:07 we are migrating Jan 26 17:19:20 need to pick stable releases, hold and deploy, then pick another stable release Jan 26 17:19:24 now going for b56 next Jan 26 17:19:28 utopian paradise Jan 26 17:19:34 it's called duckhorn Jan 26 17:20:09 dtrace, sfm, fma fairly obvious Jan 26 17:20:18 great appreciation for solaris observability Jan 26 17:20:24 really handy for debugging oracle Jan 26 17:20:36 benr is now at a different company, not familiar with customers apps Jan 26 17:20:52 nice to have the observability tools Jan 26 17:21:08 ptools, truss, dtrace, mdb right there at the fingertips Jan 26 17:21:39 solaris does define carrier grade Jan 26 17:21:51 we think it's the dev platform of choice Jan 26 17:21:55 people say linux Jan 26 17:21:59 we think it's going to change Jan 26 17:22:06 perfect for integration of zones and zfs Jan 26 17:22:14 solaris 10 has played a lot of catchup to the type Jan 26 17:22:17 s/type/hype Jan 26 17:22:34 haha Jan 26 17:22:47 a lot of things we need aren't in the updates Jan 26 17:22:51 (marketing not matching reality) Jan 26 17:23:02 clones, migratability, ... Jan 26 17:23:09 b43 gave it to us way before an update Jan 26 17:23:39 not a fully supported version of solaris Jan 26 17:23:41 never will be Jan 26 17:23:44 unless sun supports nevada Jan 26 17:23:47 we're totally open to that Jan 26 17:23:51 we'll see if that happens Jan 26 17:24:06 last person we want to call is 1800-FOR-SUN Jan 26 17:24:11 shoot out an email to the opensolaris lists Jan 26 17:24:18 actually its 800-USA-4SUN :) Jan 26 17:24:32 (yes i have it memorized :() Jan 26 17:24:35 last problem 3 days 24 hours a day awake Jan 26 17:24:38 lots of data Jan 26 17:24:42 no sun support could have helped Jan 26 17:25:01 finally had enough data, and guessed the zfs guys would definitely guess that he knew what it was Jan 26 17:25:13 jim mauro replied - had some ideas what it was, but was about to get on a plane Jan 26 17:25:22 bryan cantrill followed up offering to help up Jan 26 17:25:29 within an hour had 10 engineers helping Jan 26 17:25:35 can't beat that type of support Jan 26 17:25:58 that mail - 'zfs and nfs - plea for help' Jan 26 17:26:00 check out that thread Jan 26 17:26:08 he was desperate and scared - really nasty problem Jan 26 17:26:16 resolved within 3 hours Jan 26 17:26:19 permanently Jan 26 17:26:34 opensolaris community really rocked Jan 26 17:27:00 send that type of mail out tothelinux kernel mailing list Jan 26 17:27:08 you might get 20-40 mails in the first hour telling you that you suck Jan 26 17:27:14 you don't find the people who wrote the code Jan 26 17:27:58 so... Jan 26 17:28:02 zones and joyent connector Jan 26 17:28:09 implmented on t1000's for mail Jan 26 17:28:15 postgres on x4100 Jan 26 17:28:19 and rails applications Jan 26 17:28:37 eserver is a connector for flexibility Jan 26 17:28:41 not employ anything on global zone Jan 26 17:28:58 why would you run anything real on a global zone Jan 26 17:29:22 one of hte problems we had with the connector product was that the devs were tired of working on staging systems different to production Jan 26 17:29:41 zones provide an unique solution for this Jan 26 17:29:48 containerize the deployment env Jan 26 17:29:51 and clone them Jan 26 17:29:54 on the exact same system Jan 26 17:30:01 and just resource them Jan 26 17:30:05 on the same piece of metal Jan 26 17:30:08 works well for us Jan 26 17:30:18 gets rid of excuses, provides a lot of utilization, and easy! Jan 26 17:30:29 zones and consolidation - bingo disk and strong space on a single thumper Jan 26 17:30:38 everything is local on its own container Jan 26 17:30:50 zfs added functionality of adding data set onto zone Jan 26 17:30:52 stuff Jan 26 17:30:59 and import it and use it as a local file system Jan 26 17:31:32 brings in another advantage, zfs compression Jan 26 17:31:58 out customers love it Jan 26 17:32:25 [brb] Jan 26 17:33:15 back! Jan 26 17:33:29 wb Jan 26 17:33:48 no disk has every failed Jan 26 17:33:54 benr says fma is awesome Jan 26 17:33:59 other director is a bit worried ;) Jan 26 17:34:09 thinking about taking backup thumper and deploying it too Jan 26 17:34:12 really nervous when that happens Jan 26 17:34:18 [questions being asked] Jan 26 17:35:20 [don't know what hte questions are...brrr.] Jan 26 17:35:38 zones and hosting Jan 26 17:35:41 x4100 Jan 26 17:35:42 2 disks Jan 26 17:35:50 os on one disk, zpool on the other Jan 26 17:35:57 snapshots and rsync of backups Jan 26 17:36:00 onto thumper Jan 26 17:36:06 create master template of a zone Jan 26 17:36:09 for speed Jan 26 17:36:15 clone it off as many times as you want Jan 26 17:36:26 changing ip and passwords to create customer container Jan 26 17:36:51 typical zone root is about 5gb Jan 26 17:36:58 provide nfs storage on thumper after that Jan 26 17:40:33 thumper, zfs and nfs Jan 26 17:40:41 typical size of shared container is about 25gb Jan 26 17:41:02 flexible to give more Jan 26 17:41:39 create storage by hand on the thumper - will create hooks in time Jan 26 17:41:45 very few people appreciate libzfs Jan 26 17:41:49 meaning to write ruby bindings for it Jan 26 17:41:53 cool things can be done Jan 26 17:42:06 one commadn to create, set a quota, compression, all inherited Jan 26 17:42:08 simple to manage Jan 26 17:42:19 use zfs snapshots on thumpers to allow them to be accessible to customeres Jan 26 17:43:19 do a lot of benchmarking on zfs Jan 26 17:43:22 all looking beautiful Jan 26 17:43:29 was getting 100megs sequential Jan 26 17:43:36 untarring something small would take a long time Jan 26 17:43:40 thanks to dtrace found it Jan 26 17:43:47 disabled [some config] and it helped Jan 26 17:43:55 zfs tmp log'? Jan 26 17:43:58 zil Jan 26 17:44:01 intent log Jan 26 17:44:02 yeah, that's it Jan 26 17:44:10 thanks jbk :) Jan 26 17:44:16 <-- big fan of zfs, annoyed he can't use it at work yet Jan 26 17:44:35 should qualify this - over nfs is slow, local device on a thumper is blazing fast Jan 26 17:44:42 i probably annoyed bill moore to death when i met him at lisa on 05 -- i was quite excited about it Jan 26 17:45:06 why nfs isn't enough: Jan 26 17:45:11 easy and accepted Jan 26 17:45:19 a number of databases have issues [lock issues] Jan 26 17:45:37 snapshots can be seen but not controlled in any sense Jan 26 17:45:48 when dealing with zones particularly Jan 26 17:45:51 users want access to zfs Jan 26 17:46:12 didn't expect people tocome to joyent because of zfs Jan 26 17:46:17 more like dtrace, zones, ... Jan 26 17:46:23 can't really provide it as well as he wanted Jan 26 17:46:28 zones and nfs is an absolute no no Jan 26 17:46:34 dan price said 'don't do it' Jan 26 17:46:42 dp gave a great explanation of why Jan 26 17:46:46 here's what we're building up to: Jan 26 17:46:51 iSCSI Jan 26 17:46:53 future Jan 26 17:47:00 builds a lot of flexibility Jan 26 17:47:02 with zones, magic! Jan 26 17:47:13 take zones, stored centrally, abstracted from the metal Jan 26 17:47:23 zfs file systems can be accessed within zones - as dataset or managed by user Jan 26 17:47:25 'have fun!' Jan 26 17:47:36 virtualization is double edged sword Jan 26 17:47:50 horrific maintenance, very scary proposition Jan 26 17:48:04 take each zones, thin provisioned zpool, iscsi target Jan 26 17:48:18 advantage, abstract running zone from bare metal Jan 26 17:48:32 when system goes away, throw target to another system, import Jan 26 17:49:41 solaris does aggregation very very well Jan 26 17:50:39 switch providers - on storage network, early on realized there was a problem, on solaris was easy to do link aggregation (truncing) Jan 26 17:51:09 [i think there was a question from the audience that i didn't pick up] Jan 26 17:51:39 something about force 10? Jan 26 17:51:57 when you do start doing interesting things like this, and work out the kinks Jan 26 17:51:59 i think it's a switch Jan 26 17:52:00 you still have to manage it Jan 26 17:52:12 you really need good backup solutions Jan 26 17:52:18 something close to real time Jan 26 17:52:25 with zfs you create 1000's of filesystems Jan 26 17:52:37 in the old days you had 40 engineers Jan 26 17:52:52 not so old :) Jan 26 17:52:53 jamesd on his 60gb filesystem has 100's of them seemingly Jan 26 17:53:00 management gets harder Jan 26 17:53:18 have security internal to their network Jan 26 17:54:02 should make clear, this currently isn't in production yet Jan 26 17:54:09 personally use for benr's containers Jan 26 17:54:11 test env Jan 26 17:54:19 this is where we can go and the sorts of things we can do Jan 26 17:54:44 if we can do it, and fast, then a lot of technology can be scripted into a dataserver Jan 26 17:54:55 all about possibility Jan 26 17:54:58 really exciting about opensolaris Jan 26 17:55:04 wanna try doing this on linux or another os? Jan 26 17:55:16 get a little creative, a little drunk, and come up with something creative Jan 26 17:55:29 resource management, or lack thereof Jan 26 17:55:37 marketing is really misleading here [my opinions,and other people?] Jan 26 17:55:55 customers have largely been led to believe you can cap cpu, memory Jan 26 17:55:59 and do all these things from global zone Jan 26 17:56:03 cap cpu = pay for idle Jan 26 17:56:05 bit of the reality Jan 26 17:56:11 memory can be capped, using rcapd Jan 26 17:56:14 do it from within a zone Jan 26 17:56:19 problem when you have a 'hostile env' Jan 26 17:56:37 taking anyone, and slapping cuffs on them, until you give them a key and they become dangerous Jan 26 17:57:06 resource controls are accessible fromwithin thezones Jan 26 17:57:10 'the gridzone problem' Jan 26 17:57:20 only competitor doing similar stuff to joyent with containers Jan 26 17:57:27 sell off really small containers Jan 26 17:57:46 64mb containers for $10 Jan 26 17:57:50 [smells like rcapd] Jan 26 17:57:58 really no way of doing that, maybe they know something i don't Jan 26 17:58:05 benr goes to buy an account Jan 26 17:58:12 svcadm disable rcapd Jan 26 17:58:22 take one of brendang's scripts and allocates memory Jan 26 17:58:27 turns off and never logs in again Jan 26 17:58:37 they have small machines, little memory and no customers Jan 26 17:58:39 just not realistic Jan 26 17:59:10 need to convince users to play by the rules Jan 26 17:59:28 when you do run out of memory, something is wrong Jan 26 17:59:39 consume lots of memory, lots of swap Jan 26 17:59:45 lots of different problems there Jan 26 17:59:55 or threads :) Jan 26 18:00:13 or segkpm Jan 26 18:00:35 cpu is worse Jan 26 18:00:59 capping number of threads is great, but want to cap the amount of time they use Jan 26 18:02:13 so a partial solution is to either play into the problem [things are 'burstable'] Jan 26 18:02:31 unfortunately there's a lot of visibility to that Jan 26 18:02:47 even though they can only see their processes running, they see the load average Jan 26 18:02:55 call it 'pay for idle' Jan 26 18:03:16 especially those who pay for part of a machine Jan 26 18:03:23 if system is going 100% cpu Jan 26 18:03:35 they complain they're paying for a resource, and want that resource Jan 26 18:03:39 don't want anyone else using it Jan 26 18:03:51 traditional sun answer is 'you don't need it, it'll be there when you want it' Jan 26 18:03:57 minority of customers Jan 26 18:04:10 one of the ways to deal with this, you use pcep? Jan 26 18:04:29 the amusing thing is a lot of this is stuff i tried to sell at work Jan 26 18:04:38 just for internal stuff Jan 26 18:05:01 use fairshare scheduler to pull back Jan 26 18:05:04 hard to upsell Jan 26 18:05:08 people always have the cpu they need Jan 26 18:05:45 enter duckhorn - dan price is currently working on it Jan 26 18:06:32 gives us a lot of what we're waiting for Jan 26 18:06:40 gerry is the man Jan 26 18:06:43 [whoever he is] Jan 26 18:07:15 don't see any pages for it on opensolaris.org Jan 26 18:07:21 makes zones what they should have been Jan 26 18:07:25 resource control into the real world Jan 26 18:07:28 rcapd in the global zone Jan 26 18:07:31 hard cap memory Jan 26 18:07:41 resource control for swap Jan 26 18:08:15 jbk, might have misheard :/ Jan 26 18:08:28 i heard it as duckhorn too Jan 26 18:08:38 hard to use, people won't use them Jan 26 18:08:39 dead simple Jan 26 18:08:43 I think it's related to this: http://www.opensolaris.org/os/community/on/flag-days/pages/2006121401/ Jan 26 18:08:48 duckwater? Jan 26 18:08:56 Which is flagged for b5=] Jan 26 18:08:58 b56 Jan 26 18:09:07 duckwater is someting else Jan 26 18:09:12 yeah, that looks like it boyd Jan 26 18:09:13 name service stuff Jan 26 18:09:20 resource control management in solaris is elegant Jan 26 18:09:23 problem is how people look at it Jan 26 18:09:25 * boyd has to go back to real life Jan 26 18:10:13 moving towards real [hard and soft] resource management Jan 26 18:10:31 other projects watching closely - Jan 26 18:12:03 duckhorn Jan 26 18:12:06 iscsi target Jan 26 18:12:21 [much more powerful when integrated into zfs] Jan 26 18:13:08 enhanced smf profiles Jan 26 18:13:36 crossbow Jan 26 18:14:23 slightly worried about crossbow - careful about how it's used Jan 26 18:14:30 don't want people snooping on an internal network Jan 26 18:14:52 duckwater - supplied name services Jan 26 18:14:57 i want the public libdlpi that comes w/ crossbow, so i can finish up my one project Jan 26 18:15:07 solaris is extremely powerful, way more complex than we all admit Jan 26 18:15:11 resource management is complex Jan 26 18:15:19 security management is powerful, but non-intuitive Jan 26 18:15:40 open ourselves up to larger problems of management with things like zfs Jan 26 18:16:28 zfs - small file performance, large numbers of file systems cause very long boot times Jan 26 18:16:46 really cool to ditch ufs without journaling, long time to fsck Jan 26 18:16:54 just as long to boot zfs :( Jan 26 18:17:03 that has changed ,but in some cases you should be careful Jan 26 18:17:14 every so often take the thing down so you know how long it's going to take back up Jan 26 18:17:21 10 minute outage took one hour recently! Jan 26 18:18:09 20 minutes down to shutdown Jan 26 18:18:25 booting it up took about 10-15 mins to remount things Jan 26 18:20:05 surprised zfs doesn't prompt you for confirmation Jan 26 18:20:11 easy to make small typo Jan 26 18:20:14 fixed with zpool history Jan 26 18:20:22 [old slide] Jan 26 18:20:31 zfs replication isn't recursive, snapshot creation is Jan 26 18:20:50 currently using rsync because of the recursion problems Jan 26 18:21:03 debateable - snapshot management was seamlessly integrated Jan 26 18:21:31 all utilities written in bourne shell scripts - incredible Jan 26 18:21:47 hahahahahaha Jan 26 18:21:51 lol Jan 26 18:22:00 * jbk holds his tongue Jan 26 18:22:03 something about benr hating semantic :) Jan 26 18:23:46 performance slowdowns due to compression on zfs - benr has no numbers, but doesn't believe they are too significant Jan 26 18:24:24 ok for me to post this log later on my blog? Jan 26 18:24:36 sure Jan 26 18:25:31 sure Jan 26 18:27:56 most serious problem is that upgrades are a disaster Jan 26 18:28:08 joyents has great wonderful things Jan 26 18:28:18 at some point, he's going to have to upgrade from b43 Jan 26 18:28:28 scarey problem Jan 26 18:29:02 live upgrade isn't exactly live - it does take a while, that's a problem Jan 26 18:29:13 [back and forth between people in the room now] Jan 26 18:29:35 encourage everyone to use slash archives / webarchives Jan 26 18:29:42 flash Jan 26 18:29:47 flash, yes ;) Jan 26 18:29:56 makes more sense ;) Jan 26 18:30:09 * jbk is quite familiar with flars :) Jan 26 18:30:28 i'm a desktop developer - all this stuff isn't stuff i usually care much about Jan 26 18:30:36 <-- sysadmin Jan 26 18:30:51 upgrade cycle with zones are horrible Jan 26 18:30:56 install sucks Jan 26 18:31:01 he's never done an hp-ux upgrade :) Jan 26 18:31:06 or install Jan 26 18:31:39 everyone is serious about solaris don't use install, they use jumpstart Jan 26 18:31:41 though caiman looks interesting Jan 26 18:31:44 and even jumpstart needs help Jan 26 18:32:10 zulu - zone upgrade, live upgrade Jan 26 18:32:18 only helps you when you do initial system install Jan 26 18:32:25 another thing is asanti? Jan 26 18:32:34 dunno Jan 26 18:32:34 no public arc case or information about it Jan 26 18:32:37 not heard of it Jan 26 18:32:46 haha Jan 26 18:33:20 even without installer, ability to move zones around is a problem Jan 26 18:33:42 benr gets bfu'd machine with duckhorn, imports the zones and boots up Jan 26 18:33:44 and it works Jan 26 18:33:58 bfu upgrade gives merge problems, acr Jan 26 18:34:08 migrating real customers in the next month Jan 26 18:34:21 will find every possible problem, and write his own script/acr to do it Jan 26 18:38:09 every time they have a crash, do a full analysis of the crash Jan 26 18:38:12 had very few Jan 26 18:38:18 amazing we're running on cutting edge code Jan 26 18:38:56 great to be able to roll your own patches and deploy them Jan 26 18:39:44 heh.. well idr's are kina like that Jan 26 18:41:33 use 'uptime' - useful utility to figure out what was going on previously Jan 26 18:41:47 know exactly what the machine is doing - commercial product Jan 26 18:41:59 isn't very expensive, but powerful Jan 26 18:43:35 going to be growing over time - we know who we're looking at Jan 26 18:43:50 if you're interested in joining the team, send benr your name, and blog too Jan 26 18:43:59 interested in the movers and shakers Jan 26 18:44:06 heh i don't know how many thing i could blog about from work :) Jan 26 18:44:10 or who'd be interested Jan 26 18:44:49 working on sun systems for years and years, never lost a system Jan 26 18:47:01 vnic? :) Jan 26 18:47:48 (crossbow) Jan 26 18:48:17 jbk, bingo :) Jan 26 18:48:46 [question was how to monitor bandwidth in zones] Jan 26 18:53:16 stop reading man pages/documentation and read the arc cases Jan 26 18:53:23 if they're publically accessable Jan 26 18:53:34 :) Jan 26 18:54:03 contribution is really scary in terms of code Jan 26 18:54:10 contribute other stuff - governance Jan 26 18:54:22 at the start linux was relatively easy to contribute Jan 26 18:54:26 opensolaris is pretty mature Jan 26 18:54:31 hard to find stuff that needs fixing Jan 26 18:58:00 or old dell junk :) Jan 26 18:58:39 i know some people that work at hosting providers (that use linux), some of their issues were amusing Jan 26 18:59:20 I wish i could hear the other half of the discussion about lights out management Jan 26 18:59:25 yeah, joyent proud of their systems Jan 26 18:59:28 don't offer junk Jan 26 19:00:05 i think i'm going to head off now, so feel free to add - i'll post this log later ;) Jan 26 19:00:14 heh ok.. Jan 26 19:00:21 bed sounds not bad Jan 26 19:00:24 nighters Jan 26 19:00:24 joyent will never use intel Jan 26 19:00:25 :) Jan 26 19:00:28 Auralis, night! Jan 26 19:00:43 i'll probably need to do that in a bit.. Jan 26 19:01:07 damn time zones :) Jan 26 19:01:51 ok, i hung up - later all! Jan 26 19:02:11 * Auralis (n=barbie@port-212-202-201-12.dynamic.qsc.de) has left #opensolaris-ossvug ("Leaving") Jan 26 19:03:22 one of the issues with impi -- lack of a console log like on the sparc loms/sp's Jan 26 19:03:27 err ipmi Jan 26 19:04:42 praise for the seregenti system controllers Jan 26 19:06:16 talking about the netra, so perhaps the standard he didn't like in them is some obscure telephony standard Jan 26 19:08:04 puts os on one disk, zpool on the other, have to ahve zpool to make zone creation go really fast Jan 26 19:08:17 no on os due to hassle of making zfs root right now Jan 26 19:09:02 put most of the stuff in non-global zones (which are backed up) Jan 26 19:09:14 makes it easy to rebuild a box Jan 26 19:09:57 *applause* Jan 26 19:10:25 next month: disgtingushed engineer karl(?) talking about infiniband Jan 26 19:10:28 and something else i didn't get Jan 26 19:11:36 well.. off to bed.. Jan 26 19:11:39 * jbk (i=jason@highhopes.madhack.com) has left #opensolaris-ossvug Jan 26 19:44:39 * You are now known as GmanAFK