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


Channel log from 30 May 2010   (all times are UTC)

00:44alkisg has quit IRC
01:26alkisg has joined #ltsp
01:39F-GT has quit IRC
01:55F-GT has joined #ltsp
02:33alkisg has quit IRC
02:39alkisg has joined #ltsp
02:40alkisg has quit IRC
04:46slidesinger has quit IRC
04:49slidesinger has joined #ltsp
05:53alkisg has joined #ltsp
05:55pmatulis has joined #ltsp
06:17alkisg has quit IRC
06:38cyberorg has quit IRC
06:38sep has quit IRC
06:39Selveste1 has quit IRC
06:40sep has joined #ltsp
06:40cyberorg has joined #ltsp
07:34BWMerlin has quit IRC
08:33RiXtEr has joined #ltsp
08:35RiXtEr has joined #ltsp
09:27pmatulis has quit IRC
09:35alkisg has joined #ltsp
09:37ogra has quit IRC
09:38ogra has joined #ltsp
10:23dba has left #ltsp
10:27shamino_ has joined #ltsp
10:56pmatulis has joined #ltsp
10:59Barbosa has joined #ltsp
11:00Barbosa_ has quit IRC
11:27komunista has joined #ltsp
12:07LedHed has quit IRC
12:21NeonLicht is now known as b0T38
12:22b0T38 is now known as NeonLichT
12:22vagrantc has joined #ltsp
12:27slidesinger has quit IRC
12:29slidesinger has joined #ltsp
12:58RiX2 has joined #ltsp
13:01RiX2 has quit IRC
13:06slidesinger has quit IRC
13:12slidesinger has joined #ltsp
13:24slidesinger has quit IRC
13:35pmatulis has quit IRC
13:37slidesinger has joined #ltsp
13:50pmatulis has joined #ltsp
14:15komunista has quit IRC
14:36t0mmii_ has joined #ltsp
14:37t0mmii has quit IRC
14:48
<alkisg>
stgraber: I noticed vagrantc has some code in ltsp-init-common which puts the nbd-client processes to sendsigs.d which in turn prevents nbd-client from getting killed on shutdown. So I tested removing the upstart ltsp-poweroff/reboot jobs and running `poweroff` manually and I didn't notice any problems at all. Maybe the whole -fp thing, along with the upstart jobs, aren't really needed with sendsigs.d?
15:02bobby_C has joined #ltsp
15:15RiX2 has joined #ltsp
15:19
<rhodan>
If I use SquashFS over NBD as my root file system, should I change the default block size?
15:27
<johnny>
why?
15:27
why would you bother?
15:27
i'm not sure what our settings are
15:27
might wanna check the code
15:29
<rhodan>
johnny: Because squashfs has a default block size of 128k.
15:30
If my OS wants to read many small files, that could produce network overhead.
15:32
<johnny>
rhodan, read the code to see what we set i guess
15:33
<vagrantc>
rhodan: are you using fat clients?
15:34
rhodan: otherwise, it won't really matter much
15:34
<rhodan>
Diskless clients.
15:34
<vagrantc>
as everything happens on the server
15:34
rhodan: thin or fat?
15:34
<johnny>
that's thin then
15:34
yeah.. you rarely read after booting rhodan .
15:34
<rhodan>
You guys have a different nomenclature.
15:34
<vagrantc>
yes
15:34
rhodan: are you using LTSP?
15:35
<rhodan>
The clients have a network-mounted root.
15:35
vagrantc: No ;)
15:35
<vagrantc>
aha
15:35
rhodan: are they running applications locally, or on the server?
15:35
<rhodan>
Locally.
15:35
Fat-clients=local disk for me.
15:36
<vagrantc>
ah, then it might matter
15:36
rhodan: yeah, i know
15:36
<rhodan>
Thinclients are terminal server clients for me.
15:36
<vagrantc>
or rather, i've heard that term used
15:37
<rhodan>
That squashfs creation is damn fast.
15:37
<vagrantc>
in the end, you just gotta describe your scenario without using ambiguous terms
15:38
<rhodan>
I boot regular, performant desktop computers with PXE and let them access a read-only root file system and have writes put to a tmpfs with UnionFS.
15:39
> 752M rootfs-0.1.squash
15:39
Oh wow.
15:39
I could fit that into RAM entirely.
15:39
OTOH, it's only XFCE with OO.o and GIMP for now.
15:42RiX2 has quit IRC
15:43dmarkey_ is now known as dmarkey
15:55
<alkisg>
rhodan: did you manually setup the tmpfs/unionfs overlay?
15:55
<rhodan>
Yes.
15:56
Why?
15:56
<alkisg>
What directories did you use? The whole root?
15:56
I'm just interested to see how that goes, as I tried something similar last summer which didn't really go well.
15:58
<rhodan>
alkisg: http://nopaste.info/e75bfb191e.html
15:58
The kernel sources and portage DB aren't needed on the clients.
16:04
<alkisg>
Thanks - but I meant the directories for the tmpfs/unionfs overlay, e.g. ltsp uses these: "/var/lib/xkb /var/log /var/spool /var/tmp /tmp /var/run /var/cache/hald"
16:05
<vagrantc>
more than that
16:05
least on debian
16:06
<alkisg>
Yes this isn't the full list...
16:07
<rhodan>
Why not just overlay /var?
16:07
<vagrantc>
because it's not simply an overlay
16:07
in all cases
16:07
<rhodan>
I plan to get rid of /var/log and use syslog-ng over network.
16:07
<vagrantc>
for overlays, why not just overlay /
16:08
<rhodan>
vagrantc: I imagine that would fuck up /sys /dev and /proc
16:08
<vagrantc>
only if it's done after they're mounted
16:08
<rhodan>
OK.
16:08
So what do you mean?
16:09
Exclude /var/run from the squashfs?
16:09
<vagrantc>
i.e. if you mount / first as an overlay, and later mount directories on top of it, the overlay won't interfere
16:09
<rhodan>
Yes, I see your point.
16:09
<vagrantc>
no need to exclude anything from the squashfs, really...
16:09
<rhodan>
But why not have your intramfs use mount --move to save a few seconds?
16:09
Why not?
16:09
<vagrantc>
NBD only transfers the bits that are actually used
16:10
<rhodan>
I know that.
16:10
<alkisg>
I wonder what problems are preventing us to use the "copy on write" mount option of nbd
16:10
<vagrantc>
alkisg: i've filed a few bugs about it
16:10
<alkisg>
This way no tmpfs/unionfs would be needed
16:10
<rhodan>
Do you know how much time it takes to compress a portage tree?
16:10
<vagrantc>
rhodan: thankfully, no :)
16:10
<alkisg>
vagrantc: are they difficult to fix? any links?
16:10
<rhodan>
I do. It's 10 minutes with pigz on a dual core.
16:11
Also, I can fit my squashfs into my server's RAM.
16:11
<vagrantc>
alkisg: http://bugs.debian.org/470963 is the main one
16:12bobby_C has quit IRC
16:12
<alkisg>
vagrantc: that seems easy enough
16:12
<rhodan>
If I want to modify my chroot and be able to revert to an earlier state, what do I use?
16:13
ATM I'm doing it with compressed tarballs that I store away.
16:13
<alkisg>
...and also btw a negotiation phase could be added where the nbd-client asks the nbd-server for a specific image (so that many images can be exported on the same port)
16:14
vagrantc: other than that, nbd-server -c works ok? Then we could even use it now, by making a symlink from /tmp/xxx to /opt/ltsp/images/i386.img before running nbd-server?
16:15
<vagrantc>
alkisg: i think it worked alright...
16:15
alkisg: you've got your classic /tmp problems with that approach, but sure.
16:15
alkisg: if you did mktemp -d and then used a subdir, you'd be alright, i guess.
16:16
<alkisg>
I guess nbdrootd could do that, a seperate dir for each client based on its ip or mac address or something
16:16
<vagrantc>
alkisg: at some point, it checked if the base file was writeable, but i think that got fixed.
16:17
alkisg: why not just a random one every time?
16:17
<alkisg>
So that if a client crashes, it'll reuse the same dir/space...
16:18
But sure a random dir would also do.
16:18
<vagrantc>
ah, the NBD reconnect issue
16:18
yeah, filesystem loosing all changes on a reconnect would be bad.
16:19
<alkisg>
I wonder if also a dynamic swap file could be generated on that nbd copy-on-write root...
16:30
<vagrantc>
so as not to require a separate swap file?
16:31
er, nbdswapd
16:31
<alkisg>
Yup
16:31
And also to be dynamic
16:47pmatulis has quit IRC
17:03alkisg has quit IRC
17:04zamba has quit IRC
17:06zamba has joined #ltsp
17:14zamba has quit IRC
17:22hookdr_999 has joined #ltsp
17:25
<hookdr_999>
any fedora 13 users
17:28pmatulis has joined #ltsp
17:42zamba has joined #ltsp
18:37jhutchins_lt has joined #ltsp
18:49Selveste1 has joined #ltsp
19:06vagrantc has quit IRC
19:53pmatulis has quit IRC
20:04sbalneav has quit IRC
21:09Selveste1 has quit IRC
21:09Selveste1 has joined #ltsp
22:17ogra_cmpc has quit IRC
22:24jammcq has joined #ltsp
22:24
<jammcq>
hello all
22:30ogra_cmpc has joined #ltsp
22:30
<jammcq>
ogra_cmpc: ?
22:43alkisg has joined #ltsp
23:06hookdr_999 has quit IRC
23:32monteslu has quit IRC
23:34alkisg has quit IRC
23:43map7 has quit IRC