IRC chat logs for #ltsp on irc.freenode.net (webchat)


Channel log from 12 June 2010   (all times are UTC)

00:46japerry has joined #ltsp
01:01RiXtEr has quit IRC
01:33alkisg has joined #ltsp
01:33alkisg has joined #ltsp
01:40vagrantc has quit IRC
01:44dobber_ has joined #ltsp
02:03alkisg has quit IRC
02:04alkisg has joined #ltsp
02:07dobber_ has quit IRC
02:08alkisg has quit IRC
02:33bobby_C has joined #ltsp
02:41alkisg has joined #ltsp
03:15leio has quit IRC
03:17leio has joined #ltsp
03:29sch-bot has quit IRC
03:41otavio has quit IRC
03:51bobby_C has quit IRC
04:10alkisg has quit IRC
04:30chupacabra has quit IRC
04:33mikkel has joined #ltsp
04:35komunista has joined #ltsp
05:00F-GT has quit IRC
05:08johnny has left #ltsp
05:09johnny has joined #ltsp
05:17F-GT has joined #ltsp
06:34otavio has joined #ltsp
07:00bobby_C has joined #ltsp
07:32komunista has quit IRC
07:34F-GT has quit IRC
07:46alkisg has joined #ltsp
07:51F-GT has joined #ltsp
08:11komunista has joined #ltsp
08:17satellit_ has joined #ltsp
08:36muttley has joined #ltsp
08:38muttley has left #ltsp
08:55otavio has quit IRC
09:03satellit_ has left #ltsp
09:08F-GT has quit IRC
09:16highvoltage has joined #ltsp
09:24F-GT has joined #ltsp
09:45alexqwesa has quit IRC
09:46alexqwesa has joined #ltsp
09:47chupacabra has joined #ltsp
10:03komunista has quit IRC
10:06otavio has joined #ltsp
10:17japerry has quit IRC
10:17ogra_cmpc has quit IRC
10:18japerry has joined #ltsp
10:18ogra_cmpc has joined #ltsp
10:22Selveste1 has quit IRC
10:52peter_tm has joined #ltsp
12:37johnny has left #ltsp
12:38johnny has joined #ltsp
13:02bobby_C has quit IRC
13:33dobber_ has joined #ltsp
13:35Selveste1 has joined #ltsp
13:37nesusvet has joined #ltsp
13:37
<nesusvet>
exit
13:41
<alkisg>
logout
13:42
<nesusvet>
alkisg: hello
13:42
<alkisg>
hello :)
13:42
<nesusvet>
i am learning a new client irssi )
13:42
and a little confused
13:43
<alkisg>
sure, i was just joking
13:43
Once someone typed the letter "a" by mistake
13:43
The next once wrote "b", the next one "c" etc until we got to "k" or something - just as a joke too
13:43
:)
13:44
<nesusvet>
I'm not very surprised if i will write my secret password )
13:47nesusvet has quit IRC
14:00komunista has joined #ltsp
14:06Egyptian[Home] has joined #ltsp
14:08Egyptian[Home]1 has quit IRC
14:10
<alkisg>
stgraber: I'm preparing to commit the following change in ltsp-init-common: "if the user specified any kind of swap space, either nbd or local swap, *unload* compcache". This should make low end clients usable again without any side effects. Any objections?
14:18
http://pastebin.com/BVsh3JbR
14:19dobber_ has quit IRC
14:22dobber_ has joined #ltsp
14:34
<komunista>
is not better to make possible disable the compcache by lts.conf?
14:35Selveste1 has quit IRC
14:36
<alkisg>
At the time when compcache is loaded, we don't have access to lts.conf
14:36
It's very early in the boot process
14:37
Also at that point we don't know if a swap space *will* be mounted
14:37
(if it isn't, then compcache is OK, as it gives a little more "ram" to the client by compressing it)
14:39
The best thing would be to delay loading compcache after any swap mounting attempts - but then the swap mounting would have to be part of the initramfs, which is itsn't - it's just an initscript.
14:40
So until we can move swap mounting to the initramfs, I don't see any better way.
14:46dobber_ has quit IRC
14:47
<alkisg>
(btw there's a "nocompcache" kernel parameter that can be put in pxelinux.cfg/default)
14:52dobber_ has joined #ltsp
15:08
<komunista>
my idea is about performance, without any test IMHO local swap is better than remote, then if there is enough RAM for small compcache is better to expand swap by nbd, than replace it...
15:09Egyptian[Home] has quit IRC
15:10
<alkisg>
komunista: I didn't get the last part. If you have nbd swapping why would you also use compcache?
15:12
Also, older hard disk have terrible seek times, while a gigabit network is much faster, and the data usually are cached in the server RAM, so there's no hard disk access involved at all.
15:12
I mean that nbd swapping is faster than old hard disks... but anyway that's upon the local admin to decide it from lts.conf.
15:13
<komunista>
I have no GB network :-)
15:13
and no local disk
15:14
and I can be totally wrong about swap, I am only asking :-)
15:14
<NeonLichT>
Try swap on a floppy XDD
15:20
<alkisg>
Well. Before that change (http://pastebin.com/paQZ5zyU) my 64MB client was using 13MB for ramswap, and that had a grave performance hit. Opening firefox took a lot of time because the pixmaps or whatever it's using locally was stressing the CPU (compcache actually ZIPs the data, so it uses a lot ofCPU).
15:20
Even scrolling the firefox window involved compressing/decompressing
15:21
By enabling nbd swap or local swap (I tried both) compcache was removed and those 13MB were freed, so the performance was again acceptable
15:21
Casper also disables compcache for PCs with >512 MB. It's really only needed for PCs booting the live CD with 256 to 512 ram...
15:23Egyptian[Home] has joined #ltsp
15:26vagrantc has joined #ltsp
15:27
<komunista>
I have 256 (64 for VGA) MB RAm and 50 MB compcache
15:27
sometime I run free (by ssh) and I newer see the swap was used (or some kB only)
15:28
hops, 256 MB RAM - 64 MB for VGA card :-)
15:29
<alkisg>
komunista: 256 for a thin client are more than enough, if you don't use localapps
15:30
So even if compcache takes 50MB, your client still won't use any swap space
15:30
(those 50MB will still be unused)
15:30
But the real question is, if you do have a local swap space, why would you also give that 50MB to compcache?
15:31
vagrantc: change of course, disabling compcache when any other swapping is around works better: http://pastebin.com/paQZ5zyU - I'm gonna commit this if noone objects...
15:33
<komunista>
alksig: thanks for explanation
15:35
<vagrantc>
alkisg: i think "modprobe -r ramzswap" should remove everything
15:36
alkisg: also, it should be more able to handle if for some reason /dev/ramzswap* exist, but swapoff fails ... i.e. module loaded by nothing actually added as a swap device
15:37
<alkisg>
vagrantc: swapoff fails ==> wouldn't that "|| true" do the trick?
15:37
<vagrantc>
i also wonder if $swap_devices might be set, but failed to initialize the swap partition... might be better to leave compcache swap devices...
15:38
alkisg: yeah, a || true in both places would be needed, though
15:39
<alkisg>
(btw it'd be nice if we mounted any other swap spaces before even loading compcache, so no rmmod would be necessary, but unfortunately that can't be easily done...)
15:39* alkisg tries modprobe -r...
15:40dobber_ has quit IRC
15:44chupacabra has quit IRC
15:44chupacabra has joined #ltsp
15:49
<alkisg>
Thanks, modprobe -r works fine. Btw, `swapoff "$swap" && remove_ramzswap=true` should work correctly even when swapoff failes...
15:49
<vagrantc>
odd
15:50
<alkisg>
Ah sorry
15:50
my bad
15:51
<vagrantc>
i'd think you'd need: swapoff $foo && bar=baz || true
15:52
<alkisg>
OK, tested, committing...
15:52
<vagrantc>
sad to hear compcache isn't working as well as i had heard...
15:54
<alkisg>
*If* there is any kind of other swapping available, compcache has no reason to take up 25% of the available ram
15:54
The ubuntu live cd only uses compcache for PCs with <512 RAM
15:54
It disables it for newer PCs...
16:02
FInal result: with a default lucid installation, 64MB clients are almost unusable (they're constantly de/compressing ram into that compcache swap). With that commit, and with NBD_RAM_THRESHOLD=100, they're fine.
16:03
(so we could put that threshold by default, but anyway, that's easy to customize locally)
16:10bobby_C has joined #ltsp
16:13
<alkisg>
vagrantc: it seems that I committed the wrong version, is it OK if I uncommit?
16:13
(i.e. rmmod instead of modprobe)
16:14
(I do know how...)
16:16komunista has quit IRC
16:16
<alkisg>
Anyway, I'll just push the small fix
16:24
<vagrantc>
alkisg: nah, just make another commit with the fix
16:24
alkisg: unless you didn't push it yet
16:24
<alkisg>
vagrantc: I did
16:25
(I could still roll them back and repush into one commit, if it'll be better...)
16:25
Thanks though!
16:29
Btw, I tried increasing that "min_ram" thing to 64M. While it usually worked fine, it hanged sometimes when the root switching happened.
16:30
If the client managed to boot, then all OK; but 1 out of 10 times it wouldn't boot because of the root switching.
16:30* alkisg should file a bug to nbd-client about that...
16:30
<vagrantc>
nah, once it's realeased to the wild, best to leave it that way.
16:31
<alkisg>
Anyway, lucid terminals won't boot with 64M and without any kind of swap. So if it can't be nbd swap at that point, temporarily loading compcache is fine :D
16:34
rsyslogd, dbus etc still die with out of memory errors but they get realoaded by upstart a little later when ltsp-init-common gest nbd swapping up and running. Well... 64M is too low anyway, but at least now they boot and they work fine afterwards.
16:43alkisg has quit IRC
17:00Egyptian[Home] has quit IRC
17:09mikkel has quit IRC
17:10bobby_C has quit IRC
17:13Egyptian[Home] has joined #ltsp
17:15highvoltage has quit IRC
19:13alkisg_work has quit IRC
19:38Egyptian[Home] has quit IRC
20:54vagrantc has quit IRC
21:08gymeleou has joined #ltsp
21:45peter_tm has left #ltsp
22:15ogra_cmpc has quit IRC
22:28ogra_cmpc has joined #ltsp
22:47chupacabra has quit IRC
22:51chupacabra has joined #ltsp
22:52alkisg has joined #ltsp
23:41Egyptian[Home] has joined #ltsp