Back to Home / #xen / 2005 / 04 / Prev Day | Next Day
#xen IRC Logs for 2005-04-25

---Logopened Mon Apr 25 00:00:29 2005
00:21@cw is anyone here using glibc-2.3.4-2 with the tls changes from the xen site?
00:51--- ---> Tv [] has joined #xen
01:02--- <<-- DEac- [] has quit (Ping timeout: 480 seconds)
01:13--- ---> tierra [] has joined #xen
01:14--- ---> DEac- [] has joined #xen
01:19--- <<-- drbyte [] has quit (Ping timeout: 480 seconds)
01:26--- ---> drbyte [] has joined #xen
02:41--- <<-- drbyte [] has quit (Ping timeout: 480 seconds)
02:42--- ---> drbyte [] has joined #xen
03:09--- ---> mael [] has joined #xen
03:09mael hi
03:23--- ---> muli_ [] has joined #xen
04:07--- ---> Beaky [] has joined #xen
04:14--- <<-- tierra [] has quit (Quit: Don't hate yourself in the morning -- sleep til noon.)
04:21--- ---> hebutterworth [] has joined #xen
04:22--- ---> athomas [] has joined #xen
04:51--- <<-- elvelind [] has quit (Quit: Lost terminal)
04:53--- <<-- drbyte [] has quit (Ping timeout: 480 seconds)
04:54--- ---> drbyte [] has joined #xen
05:19--- <<-- drbyte [] has quit (Ping timeout: 480 seconds)
06:50--- <<-- hebutterworth [] has quit (Read error: Connection reset by peer)
06:59--- ---> yarihm [] has joined #xen
07:10--- ---> GvG [] has joined #xen
08:36--- User: *** unriel is now known as riel
08:40--- <<-- sunny [] has quit (Quit: bleh)
08:52--- <<-- muli [] has quit (Ping timeout: 480 seconds)
08:58--- <<-- athomas [] has quit (Quit: Leaving)
08:58--- Netsplit <-> quits: DEac-, knewt, Robot101, paavon, jeroney, Shaun, schweeb, HoraPe, joink, Beaky, (+1 more, use /NETSPLIT to show all of them)
08:58--- Netsplit over, joins: schweeb, Beaky, DEac-, jeroney, @Sir_Ahzz, knewt, Robot101, paavon, joink, Shaun (+1 more)
09:01--- ---> rharper [] has joined #xen
09:02--- Netsplit <-> quits: DEac-, knewt, Robot101, paavon, jeroney, Shaun, schweeb, HoraPe, joink, Beaky, (+1 more, use /NETSPLIT to show all of them)
09:03--- Netsplit over, joins: schweeb, Beaky, DEac-, jeroney, @Sir_Ahzz, knewt, Robot101, paavon, joink, Shaun (+1 more)
09:09--- ---> hollis [] has joined #xen
09:21--- ---> athomas [] has joined #xen
09:57--- <<-- hollis [] has quit (Quit: leaving)
10:22--- ---> hollis [] has joined #xen
10:23--- ---> cfreak [] has joined #xen
10:33--- <<-- rharper [] has quit (Quit: Leaving)
10:36--- ---> hbaum [] has joined #xen
10:45--- ---> muli [] has joined #xen
10:47--- ---> rharper [] has joined #xen
10:47--- <<-- Beaky [] has quit (Remote host closed the connection)
10:51--- <<-- HoraPe [~horape@] has quit (Ping timeout: 480 seconds)
10:55--- ---> HoraPe [~horape@] has joined #xen
11:03--- ---> eigood [] has joined #xen
11:09--- <<-- wenchien [] has quit (Ping timeout: 480 seconds)
11:12--- ---> wenchien [] has joined #xen
11:24--- <<-- wenchien [] has quit (Ping timeout: 480 seconds)
11:24--- ---> wenchien [] has joined #xen
11:33* riel looks oddly at vbd.c
11:33riel memset(ret, '\0', 0);
11:38Tv hehe
11:42eigood maybe that was an attempt to get later/earlier code to align on some magic boundary
11:42eigood one would hope gcc is smart and can optimize that away
11:42riel eigood: I don't think this is a "magic boundary" thing
11:52--- <<-- Tv [] has quit (Ping timeout: 480 seconds)
12:06riel neat, kernel-xen0 boots again on fedora rawhide ;)
12:07murb riel: do you get working userspace as well?
12:07riel murb: yes, I just upgraded something with rpm
12:12--- <<-- athomas [] has quit (Quit: Leaving)
12:23riel beautiful, I got "xm create" to oops in set_page_dirty() ;)
13:22--- <<-- GvG [] has quit (Ping timeout: 480 seconds)
13:34--- ---> Tv [] has joined #xen
13:47--- <<-- aliguori [] has quit (Ping timeout: 480 seconds)
13:48--- <<-- DEac- [] has quit (Ping timeout: 480 seconds)
13:56--- ---> itamarjp [] has joined #xen
13:59matta riel: you are updating it to a recent -unstable?
14:01--- ---> DEac- [] has joined #xen
14:02--- ---> GvG [] has joined #xen
14:03riel matta: yes, last night's snapshot
14:04riel xen-2-20050424 and kernel-2.6.11-1.1262_FC4
14:11@cw riel: are you using the glibc with the tls patch?
14:11--- <--- hena [] has left #xen ()
14:13riel dunno ;)
14:13riel what I care about now is fixing the oops on munmaping /proc/xen/privcmd ;)
14:16eigood you're mmaping it?
14:17--- <<-- itamarjp [] has quit (Quit: )
14:17riel eigood: xend is
14:17riel mmap2(NULL, 4096, PROT_WRITE, MAP_SHARED, 17, 0) = 0xb79cc000
14:17riel ioctl(17, SNDCTL_DSP_SPEED or SOUND_PCM_READ_RATE, 0xbfa8dd70) = 0
14:17riel munmap(0xb79cc000, 4096
14:17riel *BOOM*
14:18riel the ioctl number is probably IOCTL_PRIVCMD_MMAP
14:18riel yes, it is
14:32riel *YUCK*
14:32riel so we mmap /proc/xen/privcmd
14:32riel and then we call an IOCTL to actually populate the mapping with pages
14:33riel since we can't handle page faults at run time, and the mapping doesn't get set up at mmap time either
14:36--- ---> tierra [] has joined #xen
14:48--- <<-- cfreak [] has quit (Quit: .)
15:44--- ---> xai [] has joined #xen
15:48--- ---> Lampy [] has joined #xen
15:58--- <<-- GvG [] has quit (Quit: Goodnight)
16:01* riel cries
16:02riel this is a horrid mess
16:08matta eh?
16:08eigood looking at dpkg source?
16:08riel no, drivers/xen/privmsg/privmsg.c
16:08matta is it mainly gcc 4 issues?
16:09riel in order to use an mmaped area through /proc/xen/privmsg, you:
16:09riel 1) open the file
16:09riel 2) mmap it
16:09riel 3) call an IOCTL to populate the mmap !
16:09riel of course, it oopses on munmap
16:09eigood can you even mmap files in /proc?
16:09eigood is that allowed thru the procfs interface?
16:10riel mmap2(NULL, 4096, PROT_WRITE, MAP_SHARED, 17, 0) = 0xb79cc000
16:10riel yes ;)
16:10riel providing the file has an fops->mmap function
16:12eigood so maybe the xen people don't know how to write said function
16:13riel they have that one
16:13riel otherwise the mmap2() above wouldn't have succeeded
16:13eigood besides, this file sounds like it really needs to be in /dev
16:13riel that's besides the point
16:14riel the problem is that it oopses on munmap
16:14eigood so the ioctl isn't doing the same structure mods as mmap
16:15riel it _is_ doing mmap
16:18--- User: *** riel is now known as unriel
17:28--- <<-- Arnald [] has quit (Remote host closed the connection)
17:31--- <--- Lampy [] has left #xen ()
18:15--- ---> amr [~d0ea01e1@] has joined #xen
18:16amr newbie question. Does the default binary install come with an image for domain 0; or must an image be installed to boot domain 0 with the binary files.
18:25@cw amr: dom0 usually ends up being your existing installation
18:25--- <<-- amr [~d0ea01e1@] has quit (Quit: CGI:IRC (EOF))
18:31--- <<-- yarihm [] has quit (Quit: Leaving)
18:39xai is it possible to access physical devices in xen0 from a xenU?
18:40xai for some reason the faq is down.
18:40xai The requested URL /Research/SRG/netos/xen/faq.html was not found on this server.
18:42rharper xai: try:
18:45--- <<-- hollis [] has quit (Quit: leaving)
18:47--- <<-- hbaum [] has quit (Quit: Client exiting)
19:01--- <<-- cw [] has quit (Quit: leaving)
19:10--- <<-- rharper [] has quit (Quit: Leaving)
19:13--- <<-- JViz [] has quit (Quit: YES THEY DESERVE TO DIE, AND I HOPE THEY BURN IN HELL!)
19:15--- ---> drbyte [] has joined #xen
19:20--- ---> JViz [] has joined #xen
19:34--- ---> Robot101_ [] has joined #xen
19:35--- <<-- Robot101 [] has quit (Read error: Connection reset by peer)
21:07--- ---> Llama [] has joined #xen
21:07--- <<-- Llama [] has quit (Quit: )
21:49xai Where are premade images of fedora, slackware, or rhel? if any..
22:24--- ---> aliguori [] has joined #xen
22:25aliguori does anyone here boot their root partitions off of lvm? i've been doing it with devfs but devfs seems to be completely non-functional in the latest version of xen-unstable
22:25aliguori i can't seem to figure out how to do it without devfs in gentoo..
22:40--- <<-- drbyte [] has quit (Ping timeout: 480 seconds)
22:42--- ---> rusty [] has joined #xen
22:47aliguori hey rusty, how's the registry coming along?
22:49--- <<-- tierra [] has quit (Quit: bug, n: A son of a glitch.)
22:59--- <--- VS_ChanLog [] has left #xen (Rotating Logs)
22:59--- ---> VS_ChanLog [] has joined #xen
23:00--- <<-- jeroney [] has quit (Ping timeout: 480 seconds)
23:07matta aliguori: how's the next vm-tools version coming along? :)
23:08caker aliguori: I'm using LVM root, yes
23:08matta LVM for dom0 or domU ?
23:08caker aliguori:, universal initrd for LVM root goodness (dom0)
23:11aliguori matta: lvm for dom0
23:12aliguori caker: oh, thanks, let me give it a try
23:12aliguori matta: good actually, I'm attempting to put out a "preview" release but I'm having trouble testing against the latest xen-unstable :-)
23:19aliguori matta: i've got vm-tools working with -unstable and 2.0.x.. it's missing some a lot of polish and some device support but it's a big step in the right direction
23:19* caker drools for a replacement for xend
23:20caker <-- my latest xend memory usage graph
23:21aliguori caker: nuts
23:21caker I'd move to -unstable, but at the moment it's too ... unstable
23:21aliguori caker: what features of xend do you use?
23:22aliguori what would vm-tools need to do (under 2.0.x) for you to switch?
23:22caker aliguori: that causes the memory increase? I poll xm info ..
23:22--- ---> matta-lt [~matta@] has joined #xen
23:22caker aliguori: ahh -- just the basics, plus bvt tweaking and suspend
23:23aliguori yeah, that's what i figure most users need.
23:23aliguori caker: how much bvt tweaking do you do?
23:23--- <<-- schweeb [] has quit (Read error: Connection reset by peer)
23:23--- ---> schweeb [] has joined #xen
23:24aliguori we've been discussing internally how much to expose for modifying scheduler parameters.. i didn't think most people would do much more than adjust a simply weighting
23:24caker aliguori: none, currently but was hoping to mess with it
23:24aliguori ahh :-)
23:24matta aliguori: I just use the very basics
23:24caker aliguori: that's all I'm after .. basic weight tweaking
23:24matta create, destroy, shutdown, list, info
23:25matta-lt caker: you can read the lists, BVT tuning has came up many oftimes.
23:25aliguori the cambridge guys are working on yet another scheduler
23:25matta-lt it is the replacement for atropos right?
23:25matta-lt i forget the name
23:26matta-lt I think BVT works very well actually
23:26aliguori i'd rather not have to explain how to tweak 3 different schedulers in the docs
23:26matta-lt you just need to understand how it's arcaic settings are and what should be what
23:26aliguori matta-lt: i thought it was a replaced for both atropos and bvt.. honestly, i don't know why you need atropos with a proper bvt scheduler
23:26matta-lt aliguori: RR is easy, i'm pretty suring atropos is going away
23:26aliguori bvt can do, at least the original paper claims, soft real time scheduling
23:26matta-lt yeah
23:27aliguori matta-lt: I don't consider RR to even be a choice :-)
23:27matta-lt using my tuned BVT scheduler I can say nothing bad about it.
23:27aliguori i don't know why it hasn't been removed
23:27matta-lt it's for demonstration only
23:27matta-lt sort of like a template on which to build on...
23:28matta-lt Xen's QoS has really impressed me
23:28matta-lt i've used Virtuozzo which is a commercial VM-type software
23:28matta-lt whivch touts all kinds of QoS
23:28aliguori matta-lt: QoS is what a hypervisor should be all about :-)
23:28matta-lt and on my test server here .. i'd say xen handles more demanding loads gracefully
23:29matta-lt that is tuning a lot though
23:29matta-lt lots of benchmarks to see what performs best
23:29* caker mumbles something about needing disk QoS
23:29--- ---> sunny [] has joined #xen
23:29--- ---> drbyte [] has joined #xen
23:29matta-lt caker: disk QoS works fine
23:29matta-lt if anything setting separate priorities fro each VM rather than round robin would be nice
23:29aliguori I'm quite impressed with BVT.. I think it works well.. I understand the new scheduler is more smp aware though
23:30matta-lt but it's still way better than the disk I/O scheduler in virtuozzo
23:30matta-lt that was where it lagged
23:30matta-lt as far as CPU QoS Virtuozzo was definitely as good as xen
23:30caker matta-lt: so what do you guys do why a node is thrashing like made without any ability compensate?
23:30caker *mad
23:30aliguori matta-lt: i think hardware is going to be the answer for I/O scheduling.. there's a lot of interesting IO virtualization hardware on the horizon
23:31matta-lt have a disk sub-system that is capable of 100MB/s+ helps
23:31--- <<-- schweeb [] has quit (Read error: Operation timed out)
23:32--- ---> schweeb [] has joined #xen
23:32matta-lt aliguori: well, the linux kernels disk I/O schedulers seem to be useful
23:33matta-lt caker: i can't really explain it... we have tons of VM's that are swapping like mad
23:33caker hmm
23:33caker matta-lt: and you just use LVM partitions fed into the domU's right?
23:33caker matta-lt: how much memory to you allocate to dom0?
23:33matta-lt most hosts run at a constant 5MB/s in and/r ut
23:33aliguori matta-lt: do you have hyperthreading enabled?
23:34matta-lt aliguori: I use Opteron's
23:34aliguori oh
23:34aliguori matta-lt: have you tried x86_64 yet? :-)
23:34matta-lt aliguori: no.... am definitely looking forwar to it
23:35matta-lt I guess you IBM guys ae pretty dead set on making sure 32-bit domU's run on a 64-bit xen
23:35aliguori uh... we've certainly discussed it :-)
23:35aliguori there are some other core things that are probably more important though
23:35matta-lt and that was my only thing I wanted, I just want xen 64-bit so I can address >4GB RAM
23:35aliguori like pae
23:36matta-lt aliguori: it was on a doc one of the ibm people sent out
23:36matta-lt the IBM internal requirements or something
23:36matta-lt yeah, PAE might even work out
23:36matta-lt esp if it gets pushed into 2.x
23:36matta-lt PAE has always been a hack though
23:36matta-lt for xen it should work fine though
23:36aliguori matta-lt: oh, mike day's requirements.. those actually weren't just ibm requirements.. the idea was for all the companies involved in xen to get together to develop common requirements
23:37matta-lt my experience with8-16GB RAM ssytems was for oraclesevers
23:37aliguori there was actually an initial requirements doc created by HP I believe a few months ago
23:37matta-lt and PAE each process can only address 4GB RAM
23:37matta-lt er, each process run on a system where PAE is enabled is better
23:38aliguori well, as many have pointed out, there's a ton of PAE hardware out there
23:38aliguori so it's one of those ugly requirements :-/
23:38matta-lt yeah
23:38matta-lt and for xen it would work out
23:38matta-lt my largest VM's are only 512MB
23:38matta-lt so evenif there was a requirement that each VM was limited to 4GB
23:38matta-lt since in xen sense a domain is "sort of" like a process
23:39aliguori initial pae support is going to probably require all domUs to be PAE aware
23:39matta-lt oh
23:39aliguori that kind of sucks, but it's a start
23:40aliguori ideally, you want to shadow the guests so that if a domU uses < 4GB of memory, it doesn't have to be pae aware
23:40matta-lt you'd like it to be less intrusive code?
23:40aliguori well, it sucks to turn on pae if you don't actually use it
23:40matta-lt oh
23:40matta-lt i see
23:40matta-lt so in the kernel config
23:40aliguori yeah
23:40matta-lt each would have to be set to 16GB or higher
23:41matta-lt for it's memory limit setting
23:41matta-lt that doesn't sound too bad, having to change the source of the xenU kernel sucks since it is more to keep in sync with the linux kernel releases
23:42matta-lt hrm, but I guess PAE support would reside entirely in arch
23:42aliguori eh
23:42aliguori xen doesn't separate arch stuff very well
23:42aliguori pae support requires changes in arch, but then also changes in a bunch of the userspace stuff
23:43matta-lt aliguori: how is vm-tools using the cpu_weight variable in the config files?
23:43matta-lt same method?
23:43aliguori matta-lt: as xend, yes
23:43aliguori but that's changing.
23:43aliguori :-)
23:43matta-lt I have been maintaining my own patch...
23:44aliguori yeah, unfortunately, we can't take patches... stupid lawyers.
23:44matta-lt it's a very minor, perhaps just me
23:44aliguori hopefully, we'll work things out and get vm-tools into the xen tree
23:44matta-lt but I changed where CPU weight is calculated by the value / 10
23:44matta-lt I canged 10 to 1000
23:44aliguori huh
23:44matta-lt when / 10 only values 1-10 are actually useful
23:44aliguori i didn't even know it did that at all
23:45* aliguori looks in cvs
23:45matta-lt and the xm bvt command will only let you go as 'low' as 0
23:45matta-lt whereas a weight of let's say 20
23:45matta-lt would be bvt .5
23:45matta-lt equiv in dom0
23:45matta-lt and you always want dom0 to not be starved as it maintains the disks and network
23:46aliguori oh, see, i don't think weight is the best thing for bvt
23:46aliguori b/c you really want to assign two values
23:46matta-lt well, that's what it is....
23:46aliguori weight, and then responsiveness i guess
23:46matta-lt that's why it warp
23:47aliguori yeah
23:47matta-lt tuning warp and it's vales are important
23:47matta-lt as i've found
23:47aliguori yup
23:47matta-lt perhaps stuff like that is why I think Xen QoS rocks
23:47aliguori :-)
23:47matta-lt it took a lot of research to figure out what the hell those values all mean
23:48aliguori yeah, xen is still in the early adopter phase :-)
23:48matta-lt anyhow, changing to 1000 vs 10 basically makes values of 0 - 1000 useable
23:48aliguori sure, makes sense
23:48matta-lt and means you can give dom0 max weight with 0
23:49matta-lt hrm, maybe that is 0-100 usable
23:49aliguori i hope they put more info out about the new scheduler
23:49aliguori let me try this initrd, brb
23:49--- <<-- aliguori [] has quit (Quit: Leaving)
23:53--- ---> aliguori [] has joined #xen
23:53aliguori :-/ no dice
23:54caker aliguori: you tweaked the grub variables?
23:54caker aliguori: you might need to make some dev nodes inside the initrd .. I think it currently only has hd*
23:54aliguori stupid user error.. wrong kernel config
23:55aliguori ramdisk too small.. xen's default config is pretty crappy
23:55caker ahh
23:56aliguori i have my own config but apparently i didn't cp it over after doing an mrproper
23:56caker add ramdisk_size=10240 to your xeno module line
23:57caker module /vmlinuz-testing root=/dev/ram0 lvm2root=/dev/vg1/root elevator=cfq ramdisk_size=10240
23:57aliguori caker: i need to compile in support for my hardware anyway
23:57caker for example
23:57aliguori otherwise i'm gonna be using pio :-)
23:57caker nod
23:57aliguori which i don't really want :-)
23:57aliguori caker: so do you use lvm for dom0?
23:57caker aliguori: yup
23:57aliguori i've been really happy with this setup.. i just wish i could get rid of the boot partition
23:58matta-lt i don't quite see the benefit of LVM for dom0
23:58matta-lt my dom0's are braindead... I just give 10GB and some swap and they're fine
23:58matta-lt and use LVM for domU's
23:58aliguori matta-lt: i have multiple dom0's that i develop on.. i've had to resize a few times already
23:59matta-lt ahh....
23:59matta-lt yeah, for development that makes sense
23:59aliguori in production, yeah, i probably wouldn't use lvm for dom0
23:59--- <<-- lilo [] has quit (Quit: brb)
---Logclosed Tue Apr 26 00:00:46 2005