Back to Home / #uml / 2007 / 10 / Prev Day | Next Day
#uml IRC Logs for 2007-10-25

---Logopened Thu Oct 25 00:00:41 2007
00:14|-|mgross [] has joined #uml
01:11|-|mgross [] has quit [Ping timeout: 480 seconds]
01:13|-|balbir [~balbir@] has joined #uml
01:23|-|balbir [~balbir@] has quit [Ping timeout: 480 seconds]
01:35|-|balbir [~balbir@] has joined #uml
02:07|-|balbir [~balbir@] has quit [Ping timeout: 480 seconds]
02:33|-|balbir [~balbir@] has joined #uml
03:41|-|balbir [~balbir@] has quit [Ping timeout: 480 seconds]
03:54|-|balbir [~balbir@] has joined #uml
04:10|-|balbir [~balbir@] has quit [Read error: Operation timed out]
04:23|-|balbir [~balbir@] has joined #uml
05:32|-|balbir [~balbir@] has quit [Ping timeout: 480 seconds]
05:46|-|tyler29 [] has joined #uml
05:50|-|balbir [~balbir@] has joined #uml
05:51|-|darious12 [] has quit [Quit: Konversation terminated!]
05:54|-|darious12 [] has joined #uml
05:57|-|darious12 [] has quit [Remote host closed the connection]
06:17|-|tyler29 [] has quit [Ping timeout: 480 seconds]
06:21|-|tyler29 [] has joined #uml
06:41|-|tyler29 [] has quit [Ping timeout: 480 seconds]
06:54|-|tyler29 [] has joined #uml
07:18|-|aroscha [] has joined #uml
07:29|-|DomQ [] has joined #uml
07:44|-|tyler29 [] has quit [Ping timeout: 480 seconds]
07:58|-|tyler29 [] has joined #uml
08:12<Magotari-#uml->>Hello, DomQ.
08:31|-|balbir [~balbir@] has quit [Read error: Operation timed out]
08:46|-|balbir [~balbir@] has joined #uml
09:12|-|dang [] has joined #uml
09:20|-|tyler29 [] has quit [Ping timeout: 480 seconds]
09:37|-|tyler29 [] has joined #uml
09:44|-|Magotari [] has quit [Ping timeout: 480 seconds]
09:45|-|balbir [~balbir@] has quit [Read error: Operation timed out]
09:48|-|Magotari [] has joined #uml
09:58|-|balbir [~balbir@] has joined #uml
10:20|-|aroscha [] has quit [Ping timeout: 480 seconds]
10:42|-|balbir [~balbir@] has quit [Ping timeout: 480 seconds]
10:49|-|arun [] has joined #uml
10:55|-|balbir [~balbir@] has joined #uml
11:06|-|ram [] has quit [Ping timeout: 480 seconds]
11:06|-|balbir [~balbir@] has quit [Read error: Operation timed out]
11:07|-|hfb [] has joined #uml
11:19|-|balbir [~balbir@] has joined #uml
12:06|-|jdike [] has joined #uml
12:06<jdike-#uml->>Hi guys
12:06<Magotari-#uml->>Hello, jdike.
12:08|-|DomQ [] has quit [Quit: Si c'est comme ├ža moi je me *CASSE*]
12:08<Magotari-#uml->>I have found something in the code today. At the start of UML it seems to check for xmm cpu flag. But there is no such cpu flag possible on 32bit, not anymore anyway. Maybe sse is the right flag?
12:08<Magotari-#uml->>However, the code which is connected to the thing does not seem to do anything anyway.
12:09<Magotari-#uml->>A variable gets set, and I don't think it ever gets used.
12:12<jdike-#uml->>yeah, things like that need to be fixed
12:12<Magotari-#uml->>I am compile testing the UML with the bit removed.
12:12<Magotari-#uml->>Looks pretty good so far, but the compile is not done.
12:12<Magotari-#uml->>I will send in a patch shortly, but a question.
12:13<Magotari-#uml->>Would you rather I base it on 24-rc1 or 23-mm1?
12:13<Magotari-#uml->>Someone did some style changes in rc1, for some reason.
12:13<jdike-#uml->>mm1 would be better
12:13|-|balbir [~balbir@] has quit [Read error: Operation timed out]
12:13<jdike-#uml->>that contains everything in rc1
12:14<jdike-#uml->>I've been making style changes as I go
12:14<Magotari-#uml->>Right. Expect a patch in a few hours once I do some testing to make sure I did not break anything. Just in case.
12:14<jdike-#uml->>BTW, reading /proc/cpuinfo isn't the right way to look for CPU capabilities
12:15<Magotari-#uml->>Yeah, and you do it for cmov too, I think.
12:15<jdike-#uml->>the right thing to do is actually try to use the instruction in question and see what happens
12:15<Magotari-#uml->>I see.
12:15<Magotari-#uml->>Hmm... I will mess around with that. For now though, the removal thing.
12:15<jdike-#uml->>the ptrace testing code does this reasonably well
12:15<jdike-#uml->>check for functionality by actually trying to use it
12:15<Magotari-#uml->>Oh, and there is a SCSI build problem too, but I hear that SCSI is badly broken so...
12:16<Magotari-#uml->>I chalk it as not a fault of uml.
12:26|-|balbir [~balbir@] has joined #uml
12:47|-|dserban [] has joined #uml
12:49<dserban-#uml->>I hate asking this in a specific channel, but after looking at the virtualizers comparison chart on wikipedia, I walked away feeling a little dumber. I want to know if UML is right for my needs. Mainly ISP virtual servers type stuff with rapid deployment as well as security encapsulation... :) I'm sure I'll hear a resounding yes, but I can't figure the difference between lguest/xen/uml/kvm ... there are just too many.
12:53<dang-#uml->>UML is nice for that kind of thing, because of it's flexibility.
12:53<dang-#uml->>Want a new server? Start one.
12:53<dang-#uml->>On the other hand, it lacks things like resource guarantees that xen has.
12:55<dang-#uml->>So, if you don't have some kind of SLA with resource guarantees, I'd say try UML first. It will be easier to set up and configure.
12:55<dang-#uml->>Much easier.
12:58<dserban-#uml->>ok but you say there are no resource guarantees. Which means that a malicious (or stupid) user can bring the whole box down to a crawl?
12:58<newbie-#uml->>Ok, my saga continues... making some progress with my fc3 root filesystem. In the last chapter (ie yesterday) I nailed the problem down to one line of rc.sysinit where kmodule was being called. I decided to simply comment it out (modules should load as needed anyway so no loss in not pre-loading them). Now I am seeing an error about the filesystem being read-onlty and that a lock in /var/lock/subsys/ can not be removed. I shutdown the UML, mounted the im
12:58<dang-#uml->>dserban: No, that's not true...
12:58<dang-#uml->>There are resource limits, but not guarantees.
12:59<dang-#uml->>A uml has a limit, for example, on the amount of ram or disk it can use.
12:59<dang-#uml->>But it's not guaranteed that amount of ram.
12:59<dang-#uml->>It could be in swap on the host.
12:59<dserban-#uml->>Ahh. I see what you mean. Same goes for CPU? No guarantee since the host may be hogging it?
13:00<dang-#uml->>Also, UML has no CPU guarantees. The host scheduler guarantees fairness among all processes, but you don't get guarantees for x% of cpu.
13:00<jdike-#uml->>Same goes for everything, pretty much
13:00<jdike-#uml->>upper limits, but no lower ones
13:00<dserban-#uml->>right so a process running on a guest can theoretically bring the system down?
13:01<dang-#uml->>No, that should be safe.
13:01<dang-#uml->>It can use all the CPU, if nothing else is running on the host.
13:01<jdike-#uml->>However, if there are guarantees provided by the host, UML can take advantage of them
13:02<dserban-#uml->>I've had things like mysql bring the loadaverge of a system in the 90 zone.. :( heh. so things like that I'd like to avoid. If something is coded badly, i'd like the other guests to have a fighting chance to still serve what needs to be served.
13:02<jdike-#uml->>i.e. you can bind a UML to a CPU, bind everything else off that CPU, and that UML has a dedicated CPU
13:02<dserban-#uml->>dang, I see.
13:02<dserban-#uml->>jdike, that would only work on a multi-processor system correct?
13:02<jdike-#uml->>of course
13:03<dserban-#uml->>Just being clear ;)
13:03<jdike-#uml->>a single UML can add only one to the host load average
13:03<jdike-#uml->>regardless of what's happening inside it
13:03<jdike-#uml->>except that the disk I/O threads can be in disk wait, which contributes to the load ave
13:03<dang-#uml->>And it should compete equally for CPU time with other uml's
13:03<dserban-#uml->>very nice!
13:04<dang-#uml->>(and other processes on the host, of course)
13:04<dang-#uml->>Of course, the load average *inside* the uml could be 90, but that's a different issue.
13:04<dserban-#uml->>ok, next question... wth happened to the uml website? ;) All the info's hidden in the "old" site...
13:05<dserban-#uml->>Not really a question, since google can find my howtos for me :)
13:05<dserban-#uml->>dang, I understand now.
13:05<jdike-#uml->>there's a link to it
13:05<jdike-#uml->>but it's an unorganized mass of stuff
13:06<dang-#uml->>dserban: Buy the book. :)
13:06<dserban-#uml->>Ok, I'm gonna run off and do some reading on uml.
13:06<dserban-#uml->>dang, I hate paper.
13:06<dserban-#uml->>I'd buy the pdf :)
13:06<dang-#uml->>dserban: I'm not a huge fan of it myself, but it's a hugely useful book.
13:06<dang-#uml->>(paper, that is)
13:07<dserban-#uml->>so lguest and kvm are otherwise a useless option? (I'm hoping for an unbiased answer)... uml seems to be better documented and somewhat simpler to setup? :o
13:07<dserban-#uml->>ahh... i gotta run. bbiab
13:07<dang-#uml->>lguest and kvm are pretty new...
13:08<jdike-#uml->>lguest and kvm are not production-ready at this point
13:08<dang-#uml->>I've used both, but neither is mature enough (IMO) for commercial deployment.
13:08<dang-#uml->>Right, what he said. :)
13:11<Magotari-#uml->>From experience: kvm ain't so horrible, but lguest is very immature. Neither got me what I wanted.
13:12<jdike-#uml->>kvm or something like it is the future
13:12<jdike-#uml->>at least when you have the hardware support
13:13<Magotari-#uml->>Sadly so. *sigh* I really like the idea of UML.
13:13<dang-#uml->>My problem with kvm was it needing all of qemu...
13:13<dang-#uml->>I like UML being self-contained.
13:14<dang-#uml->>kvm has a way to go to catch vmware-player...
13:14<dang-#uml->>(If I want a whole virtual machine, of course)
13:14<jdike-#uml->>Magotari, you don't lose UML
13:14<Magotari-#uml->>I find UML elegant and simple to work with. No tedious setups, hypervisors, closed source or special hardware.
13:14<jdike-#uml->>UML can run as a kvm guest, and it'll act just like it does now
13:14<jdike-#uml->>just faster, hopefully
13:15<jdike-#uml->>that's the nice thing about kvm's model
13:15<Magotari-#uml->>I don't see how adding a layer of abstraction could help, but ok...
13:15<dang-#uml->>:) kvm is just kernel level access to hardware virtualization.
13:16<dang-#uml->>Lots of things *can* run on it.
13:16<Magotari-#uml->>Gotcha now. Thanks for the explanation.
13:16<dang-#uml->>Right now, tho, only a modified qemu *does* run on it.
13:16<Magotari-#uml->>I thought kvm reffered to the modified qemu.
13:16<Magotari-#uml->>Hmm... Maybe kqemu too that way?
13:16<jdike-#uml->>I'm trying to think of other, "normal" apps that could use hardware virtualization
13:16<Magotari-#uml->>For all us lame brothers without super cpus?
13:16<jdike-#uml->>like could emacs or apache use their own page tables?
13:16<dang-#uml->>kqemu will eventually b replaced by kvm, I suspect.
13:17<jdike-#uml->>then you have normal UML
13:17<dang-#uml->>Basically, it would work like this: modprobe kvm; run uml
13:18<Magotari-#uml->>Yeah, I get that part now. But would it be possible to use kqemu instead of kvm? It would need changes in the code, but would it be a theoretical possibility?
13:18<dang-#uml->>jdike: You think uml would need to be compiled differently for the two?
13:18<dang-#uml->>Magotari: I don't think there's an advantage.
13:18<jdike-#uml->>dunno yet
13:18<dang-#uml->>UML already does what kqemu allows.
13:18<dang-#uml->>It does it by modifying the OS (UML in this case)
13:19<dang-#uml->>You only need kqemu to run unmodified OSs virtually.
13:19<jdike-#uml->>my first thought is yes, but possibly it could be built to detect the hardware and behave accordingly
13:19<dang-#uml->>That'd certainly be nicer...
13:19<dang-#uml->>Of course, it becomes less of an issue as time goes on.
13:21|-|mgross [] has joined #uml
13:24|-|mgross [] has quit [Remote host closed the connection]
13:25|-|mgross [] has joined #uml
13:36[~]jdike #uml untangles a header file mess#uml-> untangles a header file mess
13:41|-|tyler29 [] has quit [Ping timeout: 480 seconds]
13:42|-|anderiv_ changed nick to anderiv
13:53<newbie-#uml->>my saga update: seems Anacron loads ok but that's when services stop loading - filesystem still read-only - most puzzling - can login via console and mount says / is mounted as rw which is STRANGE INDEED
13:53<newbie-#uml->>this sort of problem feels vaguely familiar - just forget where I saw it last
13:54<newbie-#uml->>and how to fix it :-(
13:55<jdike-#uml->>can you modify files?
13:56<jdike-#uml->>and do you really need this particular FC3 fs?
13:56<jdike-#uml->>there are newer ones which probably work better
13:58|-|tyler29 [] has joined #uml
13:58<newbie-#uml->>I can't create or remove files.
13:58<newbie-#uml->>I think I do need it - I need our setup - ie: BOTH 32bit and 64bit libs need to be installed.
14:00<Magotari-#uml->>jdike: Does adding rw to the commandline work in UML too?
14:05<newbie-#uml->>jdike: I could go at this from another angle - I could start with bare minimum fc3 then get a list of our rpms and add them. I could also consider kickstart (the way we usually setup machines) but I have not been able to find the easiest way to get netwoking AND routing all setup from the command line when launching UML
14:05<jdike-#uml->>that works
14:06<jdike-#uml->>that's the usual way to bootstrap an RPM-based filesystem
14:06<newbie-#uml->>which? kickstart?
14:06<jdike-#uml->>as for networking and routing, the distro should set everything up for you given an ifcfg-eth0 file
14:06<jdike-#uml->>install a minimal number of RPMs, then boot it and upgrade/install more
14:06<newbie-#uml->>ideally it would be dhcp but uml won't see the real network without bridging or other assistance
14:07<newbie-#uml->>I still don't know if I can launch kickstart AFTER a minimal install
14:07<jdike-#uml->>uml_net will set up the host
14:07<newbie-#uml->>hmm it will?
14:10|-|balbir [~balbir@] has quit [Remote host closed the connection]
14:10<jdike-#uml->>if you put eth0= on the UML command line without naming a tap device, the driver will invoke uml_net to allocate one and set up the host
14:11<jdike-#uml->>you don't want to do this if you're concerned about security, but for playing around, it's fine
14:11<jdike-#uml->>and it will tell you exactly what it does so you can copy it if you want to allocate the tap device for it
14:11<newbie-#uml->>I will need dhcp too OR a way to specify the default routing
14:14<Magotari-#uml->>dhcp does not pass into UML by default.
14:14<Magotari-#uml->>You would need a forwarding program on the host.
14:14<Magotari-#uml->>I tried, I failed. Nowadays just going with routing adjusting on the guest.
14:15<newbie-#uml->>I'm looking for the simplest path to getting this going. Is there a way to specify routing and netmask on the command line? Then I can use a static IP on the command line?
14:15<Magotari-#uml->>uml_net will set up the host for you, so all you really need is to do an ifconfig ethn up <ip> and route add default gw...
14:15<jdike-#uml->>you can't do it on the UML command line
14:16<jdike-#uml->>routing and stuff is userspace responsibility
14:17<newbie-#uml->>:( hmmm ok
14:18<newbie-#uml->>let me call my local kickstart expert and see if I can invoke it after a minimal install
14:21<newbie-#uml->>ok, she will be visiting me in a bit.
14:21<newbie-#uml->>Anyone know if ks is a stardard boot option in the linux kernel or unique to the fedora/rehat kernels?
14:21<newbie-#uml->>ks=activate kickstart
14:22<Magotari-#uml->>rh afaik
14:22<newbie-#uml->>that's what I thought
14:22<newbie-#uml->>ahh perhaps I could boot the guest from an iso image of the standard install CD?
14:23<newbie-#uml->>I tied that with the kickstart iso but it quickly wanted the network and failed since it was not configured
14:27<newbie-#uml->>Would booting from a cdrom iso be as simple as setting ubda=fc3cdrom.iso?
14:32<dang-#uml->>It should be...
14:32<Magotari-#uml->>initrd problems
14:32<dang-#uml->>ubd can be anything that can be loop mounted, I believe...
14:33<dang-#uml->>(assuming you have iso9660 in your UML kernel, of course)
14:33<jdike-#uml->>ubd can be any file at all
14:33<dang-#uml->>Well true, but it has to have a filesystem in it to be useful. :)
14:33<jdike-#uml->>anything that does open/read/write/seek/close
14:34<dang-#uml->>It shows up as a block device, I know. But he was talking about booting from it.
14:34<dang-#uml->>So it has to have a filesysteem.
14:34<dang-#uml->>er filesystem.
14:34<jdike-#uml->>config ubdb=my-file; UML# dd if=/dev/ubdb of=my-file is a good way to get data into UML
14:34<jdike-#uml->>for booting, yes
14:35<Magotari-#uml->>You need an initrd in your commandline if you want to boot from a cd most of the time.
14:35<dang-#uml->>Depends on the CD... and how you've compiled your UML.
14:51<newbie-#uml->>I have both an iso of the image (which failed to boot - so I guess no iso support in my uml guest kernel) and now (just created) an ext3 version
14:51<newbie-#uml->>predictably it wants to know initrd's location
14:52<newbie-#uml->>I have found it in /isolinux/initrd.img and am about to try adding that to the command line
14:53<newbie-#uml->>well - no go - it actually wants to know where init is NOT initrd and I don't know enough about isolinux boot processes to tell it
14:53<newbie-#uml->><sigh - back to google I guess>
14:53|-|baroni [] has joined #uml
14:55<dang-#uml->>Init is probably inside the initrd...
14:55<dang-#uml->>It's usually /isolinux on an isolinux setup, if I remember correctly...
15:17|-|tyler29 [] has quit [Ping timeout: 480 seconds]
15:30|-|tyler29 [] has joined #uml
15:41|-|netdrake_ [] has joined #uml
15:41|-|netdrake [netdrake@2001:5c0:904b:ffff::1] has quit [Read error: Connection reset by peer]
16:19<Magotari-#uml->>jdike: I have a question. Would detecting a SIGILL be sufficient to determine the non-existance of cmov? If there isn't more to it, then I think I can do it. I need to learn signal handling anyway.
16:19<jdike-#uml->>maybe SIGSEGV
16:20<Magotari-#uml->>Aha. I think I can live with that. I will try.
16:39<newbie-#uml->>dang: yes, it is inside the initd, now I just need to figure out 1) if that initrd is actually being mounted and 2) if so, at what point
17:04|-|dang [] has quit [Quit: Leaving.]
17:24|-|hfb [] has quit [Quit: Leaving]
17:47|-|jdike [] has quit [Quit: Leaving]
17:59|-|tyler29 [] has quit [Ping timeout: 480 seconds]
18:01|-|silug [~steve@] has quit [Read error: Connection reset by peer]
18:06|-|baroni [] has quit [Quit: Saindo]
18:12|-|tyler29 [] has joined #uml
19:09|-|mgross [] has quit [Quit: Leaving]
19:18|-|silug [~steve@] has joined #uml
20:04|-|tyler29 [] has quit [Ping timeout: 480 seconds]
20:23|-|dang [] has joined #uml
21:08|-|mgross [] has joined #uml
23:20|-|iamtravis [~travis@] has joined #uml
23:21|-|iamtravis [~travis@] has quit []
23:59|-|VS_ChanLog [] has left #uml [Rotating Logs]
23:59|-|VS_ChanLog [] has joined #uml
---Logclosed Fri Oct 26 00:00:35 2007