[00:09] Simon (~sgarner@apollo.quattro.net.nz) joined #vserver. [00:10] hi simon! [00:12] Bertl the MHZ error disaper after 1 hour of uptime [00:13] yup sure it does ... but the btime patch should eliminate this immediately ... [00:28] blueshoe (~blueshoe@phylogenomics.Berkeley.EDU) left irc: Quit: ... [01:15] JonB (~jon@194.239.210.74) joined #vserver. [01:15] hi jon! [01:16] hey Bertl [01:33] @alex we should discuss the per context sub-capabilites, do you have some time? [01:34] Bertl: i'm interrested [01:34] Bertl: at least in knowing what the sub-capabilities are [01:35] well I suggested it several times, and it seems it isn't only useful but required ... [01:35] now for what it is ... *G* [01:35] you know the capability system? [01:38] not much [01:38] okay you have a linux system at hand? [01:39] yes [01:39] Bertl: i'm sitting infront of one [01:40] less /usr/include/linux/capability.h [01:41] got it [01:41] so basically each process has those capabilities ... [01:42] and the set of capabilities given to a process can be inhereted and modified ... [01:42] cat /proc/self/status [01:42] there you'll find CapInh/CapPrm/CapEff [01:44] if you read the description what CAP_SYS_ADMIN is for ... you will soon find, that this is a lot of different things in one tiny bit ... [01:46] Bertl: thanks for the explination [01:47] now the concept is, to add a separate set of capabilities for the context ... maybe with inheritance later too ... [01:47] but we can do that already? [01:47] what? [01:50] by the way, maybe you should read a little about (Open)Mosix ... you'll find that this propably isn't what you want ... [01:50] good night all [01:50] night alex! [01:51] Bertl: who me ? i have read about open mosix [01:52] but you will never be able to 'migrate' anything away from one node ... [01:52] Bertl: maybe not now, but i see the posibilities, not the current implementation [01:53] well, not with this approach ... [01:53] Bertl: check the mailing list, i just spammed it with 3 emails [01:53] I read them ;) [01:53] Bertl: already ? [01:53] I'm fast ;) [01:53] Bertl: i mean, i just finished typing, and one of them was big [01:54] hmm, maybe that one hasn't reached the list, the the I want to move it all mail I read ... [01:54] Bertl: cant we do separate capabilities for each contexts ? [01:54] goodnight :) [01:54] Bertl: there are more comming [01:54] err, oops. [01:54] hi! [01:54] Bertl: one with a thinkgeek example [01:55] serving (~serving@213.186.190.186) joined #vserver. [01:55] @jon I'm currently reading that one ;) [01:56] Bertl: good [01:56] Bertl: the functionality i describe there is what i want. The mixture of openmosix and vserver might not (currently) give me that, but that is a minor detail [01:57] well, it will work the way you described the TG example ... because all vserver activity is bound to one 'initial' server ... and the 8-Way opteron only is there to do the computational stuff (ala load balancing) [01:58] but it's a far way to actually moving the vserver from one machine to the other without restarting any process ... [01:58] Bertl: i know i cant do it now, but i can dream cant i ? [01:59] Bertl: anyway, i have that test server up and running, and i get these errors: [01:59] shadow (~umka@212.86.233.226) left irc: Quit: to gream [02:00] "/usr/bin/vserver: ulimut: cannot modify max user processes limit_ Invalid argument [02:00] you can dream, and you can use -HS ;) [02:01] Bertl: well, it works on another machine? [02:01] i think [02:01] well, maybe not a >= 2.4.22pre3 kernel then ;) [02:01] Bertl: i suppose not [02:01] this is a kernel issue, has nothing to do with vserver ... [02:02] Bertl: okay [02:02] they changed how setrlimit (ulimit calls that) responds to certain settings ... [02:02] Bertl: okay, it works fine now [02:03] Bertl: i'm not sure if mosix is what is needed, or it is better to make my own approach, time will tell if i code it [02:04] Bertl: i thought about doing that for my 6 month's project that makes me Master of Computer Science [02:04] the first step would be to extend the suspend to disk functionality in such a way that it would work on different systems .. suspend on system A, resume on system B [02:05] then do this just for single processes/process groups ... [02:06] if all this works, you can write a network protocoll to transfer the data from one system to the other ... [02:06] Bertl: why not just extend the openmosix way to allow moving all kind of processes ? [02:06] Bertl: or allowing to suspend single processes ? [02:07] mosix runs via the proxy concept ... all I/O is located on the start node and never moves ... [02:07] in this concept, if the initial node goes down, all processes started there are lost ... [02:08] Bertl: not good enough [02:08] Bertl: i never checked openmosix that much [02:08] Bertl: i figured that they really did just suspend the process to disk, and then moved it and started it again [02:09] matta (matta@69.10.150.254) joined #vserver. [02:09] hi matt! [02:09] hi [02:09] Bertl: well, i suppose i have my 6 month's project [02:09] so i was re-reading over the ckrm thread... [02:10] good .. what did you conclude? [02:10] has anyone started work on porting vserver to 2.6 or work with ckrm? [02:10] ckrm ? [02:10] yes rik did ... [02:10] it's something i'd be interested in testing, 2.6 will be released soon and it includes all of the scalability improvements that you're currently having to rediff [02:10] + much much more.. [02:11] agreed on that ... [02:11] i thought I would have heard about it in here, so besides rik I assume no one else put much effort into it :) [02:11] not yet ... [02:11] the stuff I did (and had to abandon for CKRM) still needs some serious work [02:11] just an indication in what direction we might want to go [02:12] hi rik! [02:12] well it looks like ckrm does some things that no one has even metioned to do with vserver [02:12] a big one is I/O guarantees [02:12] as I/O is the biggest bottleneck for most applications [02:13] http://vserver.13thfloor.at/Stuff/vs-limits.txt [02:13] yeah, i re-read the thread because I sawe that :) [02:13] what is CKRM ? [02:14] JonB: ckrm.sf.net [02:14] Class based Kernel Resource Management ... [02:14] matta: thanks [02:14] it just seems like any more "limits" implemented in 2.4 would have to be re-done in 2.6 [02:14] and 2.6 will start to be the default in distros within a few months i'm sure... (or at least an install option) [02:15] matta: i use 2.6 already [02:15] yes, if we can a) find a ctx:class mapping which seems simple and b) resolve the issues resulting from the context separation ... [02:15] JonB: exactly, wouldn't it be even better to run vservers under 2.6? :) [02:16] matta: i cant, because i need some encryption that only runs on 2.4.21 and earlier [02:16] matta: and no sufficient encryption technology exists on 2.6 [02:16] well my current threshold is 3 people ... as soon as I have 3 testers for 2.6, I'll start the work on it ;) [02:17] i was running 2.6 until arjanv stopped updating the 2.6 rpm's for redhat :) <-- lazy [02:17] Bertl: well, consider me 1 [02:17] as always.. :) [02:17] and a good one too, if I might add ... [02:17] thanks ;) [02:18] well, basically you did 80% of all vserver testing with the new stuff ... [02:19] yeah, i noticed no one responded to the list regarding uptime/reboot [02:20] both small but very useful features [02:20] well, I'm used to getting not much feedback, ... one thing really puzzles me, the kernel_cap_t is 32 bit, but the /proc display shows 64 bits ... [02:20] for the uptime? [02:20] nope, the capabilities ... [02:21] sorry for jumping around ... [02:21] I'm currently working on 3 parts in parallel ... [02:22] one of the nice things about 2.6 and CKRM is that 2.6-vserver probably won't need to do any limits at all [02:22] question I have about ckrm... [02:22] OTOH, for 2.7 I'll probably need to make an "LRM" infarstructure ;) [02:22] Linux Resource Modules or similar [02:22] @rik did you get a feeling how much overhead ckrm introduces? [02:23] does the memory part of it have min/soft/max ? [02:23] Bertl: nope, not a clue [02:23] because unless vservers are allowed to have their own swap area, a soft limit would be very useful. [02:23] matta: min/soft only [02:23] no hard max [02:23] hrm [02:24] what happens when a process (or context) reaches the soft? [02:24] well I'm sure it can be added to ckrm ... [02:24] the whole project is in early development, but it advances very fast ... [02:25] matta: nothing [02:25] is it told to swap out? does it start denying large requests for memory? [02:25] so a min is a guarantee of how much memory the process will always have [02:25] at swapout time the system simply first evicts the pages from resource classes that are over their soft limit [02:25] and the soft is just for show, right now [02:25] nope, not for show [02:25] you WANT soft limits only [02:25] well, there arises another problem. [02:26] because otherwise a class over limit could be slowing down IO for all the other classes, even with enough memory available [02:26] disk I/O [02:26] @rik, @matt what I think you really want is a sof VM and a hard swap limit ;) [02:26] Bertl: yes! [02:26] that is actually a good idea. [02:27] like my thinking with UML is that each virtual servers can have it's own swap area [02:27] so someone purchases a UML server with 32MB of ram and creates a 256MB swap [02:27] then they load up apps and it uses all swap [02:27] and it's constantly swapping in/out [02:28] which eats up the host servers disk i/o [02:28] so i don't know if forcing to swap is always the best idea. [02:29] unless of course there is a per-context I/O weighted scheduler with guarantees [02:29] alekibango (~john@b59.brno.mistral.cz) left irc: Remote host closed the connection [02:29] well you can't have both, a limit to RSS and no swapping, but you can add a penalty for I/O ... so that swapping will become much slower and the overall system behaviour much better ... [02:29] Bertl: do you have any 2.6 patches yet ? [02:30] well, the QHA patches ... but I guess they won't count ... [02:30] qha ? [02:30] and of course the BME patches ... [02:30] Quota Hash Abstraction and Bind Mount Extensions [02:30] okay [02:36] brb [02:36] matta (matta@69.10.150.254) left irc: Quit: Why is the alphabet in that order? Is it because of that song? [02:45] alekibango (~john@b59.brno.mistral.cz) joined #vserver. [02:48] JonB (~jon@194.239.210.74) left irc: Ping timeout: 492 seconds [04:37] alekibango (~john@b59.brno.mistral.cz) left irc: Remote host closed the connection [06:01] Nick change: riel -> surriel [06:19] serving (~serving@213.186.190.186) left irc: Read error: Connection reset by peer [06:55] serving (~serving@213.186.190.186) joined #vserver. [08:15] shadow (~umka@212.86.233.226) joined #vserver. [08:15] morning.. [08:15] morning/night alex ... [08:16] Nick change: Bertl -> Bertl_zZ [08:19] good dream :) [11:33] kestrel_ (~athomas@syd-h43C.adsl.AlwaysONLINE.net.au) joined #vserver. [11:43] hi there [12:42] Simon (~sgarner@apollo.quattro.net.nz) left irc: Quit: so long, and thanks for all the fish [13:19] say-out (~say@212.86.243.154) left irc: Read error: Connection reset by peer [13:31] JonB (~jon@129.142.112.33) joined #vserver. [13:59] serving (~serving@213.186.190.186) left irc: Ping timeout: 493 seconds [15:01] JonB (~jon@129.142.112.33) left irc: Quit: Client exiting [15:15] say-out (~say@212.86.243.154) joined #vserver. [15:18] shadow (~umka@212.86.233.226) left irc: Ping timeout: 493 seconds [15:44] kestrel_ (~athomas@syd-h43C.adsl.AlwaysONLINE.net.au) left irc: Ping timeout: 493 seconds [15:50] serving (~serving@213.186.189.145) joined #vserver. [17:18] shadow (~umka@212.86.233.226) joined #vserver. [19:32] mhepp (~mhepp@r72s22p13.home.nbox.cz) joined #vserver. [20:08] shuri (~ipv6@CroCrodile.HuNter.blacktaboovideo.com) left irc: Ping timeout: 483 seconds [20:37] Petar (~petar@202.6.124.92) joined #vserver. [20:41] Hello! I'm having trouble getting the 2.4.22 patch applied against kernel. Anyone feel up to helping a linux novice? [20:42] ask away [20:43] Hi. Have d/l the split-2.4.22-vs1.00.tar.bz2 [20:44] and unpacked it in the /usr/src/linux-2.4.22 source directory but can't work out how to get 'patch' to apply the diff files created from it/ [20:46] Oh & 2.4.22 kernel is fresh from ftp.kernel.org and running evrything on a stock RH9 box built today if that helps [20:47] you've configured and compiled your own kernel before ? [20:48] Vague rememberance of doing it once with rh7.1 but that was a long time ago. I have a set of instruction here and can do the kernel compile without the patch being applied and installed in /boot. I just have never attempted to /patch/ a kernel source tree first. [21:00] unpack kernel, patch kernel, make menuconfig, make dep, make bzImage, make modules, make modules_install is the order and I have worked through all the steps without doing the patch. Just can't apply the patch without errors. [21:04] You still there? [21:12] ensc (~ircensc@134.109.116.202) joined #vserver. [21:12] hello [21:20] Petar (~petar@202.6.124.92) left irc: [21:58] mhepp (~mhepp@r72s22p13.home.nbox.cz) left irc: Remote host closed the connection [22:05] matta (matta@tektonic.net) joined #vserver. [22:05] interesting... [22:06] i can renice processes in vservers from context 0.. [22:06] yes. you can it. [22:07] weel - i go to bed.. [22:07] good day for all. [22:07] shadow (~umka@212.86.233.226) left irc: Quit: to dreamland.. [22:39] AGoe (~agoeres@80.184.238.42) joined #vserver. [22:40] AGoe (~agoeres@80.184.238.42) left irc: Client Quit [23:31] AGoe (~agoeres@80.184.238.42) joined #vserver. [23:31] AGoe (~agoeres@80.184.238.42) left irc: Client Quit [00:00] --- Tue Nov 11 2003