[OpenAFS] OpenAFS on windows - profile in AFS, who uses it?

Lars Schimmer l.schimmer@cgv.tugraz.at
Mon, 11 Feb 2008 12:15:41 +0100

Hash: SHA1

Jeffrey Altman wrote:
> Lars Schimmer wrote:
>> 1. slowness - Users reports of LONG loading times for 400-1000 MB
>> roaming profiles, up to 20 minutes
> How large is your cache?
> If your cache is smaller than what you are loading all you are doing
> is repeatedly thrashing the contents of the cache.  That will result
> in the worst possible performance.
> OpenAFS must be tuned for your environment.  The defaults 96MB of cache=
> data, 10,000 status cache objects, 5000 volumes, 2500 cells are not
> right for everyone.  They are simply reasonable defaults.
> The max cache size of 32-bit Windows is system dependent but will
> be somewhere between 800MB and 1.2GB.  For 64-bit Windows the limit is
> 512TB.

Ok, I try to setup a 800 MB cache, but on a 2 year old system, with 2 GB
space on C:, it's just a wish, I could only get a 400 MB cache. But I
hope it gets faster.

>> 2. as in computergraphic, we handle 10-400 MB models with our programs.
>> Getting this via the fileserver at 5-6 MB/sec is kinda slow.
> Faster network?

Faster than gigabit?
The fileservers just send data with 10 MB/sec to linux clients, to.

>> The cache on AFS is written as a single file, I would prefer a partiti=
>> as it can't be defragmented
> The cache is a paging file.  As with any file it cannot be defragmented
> while in use.  You can stop the afs client service and defragment the
> file if desired.  The file is also allocated entirely at the same time
> so if there is contiguous space at the time of allocation there will be
> no need to defragment it later.
> You can choose where you want the paging file to be located.  If you
> want to put it on its own partition you can do so.

May I should consider this for new workstations.

>> 3. reliability - some days windows hit a error while saving the profil=
>> and no "readable" info could be found - I assume mostly unicode proble=
> need more details.

Right now such a special "user profile".
Win client 1.5.3009 (fresh installed today, 400 MB cache, Windows XP
SP2) and user profile in AFS space, one RW, 3 RO and a backup volume,
all on OpenAFS fileserver 1.4.6 on debian.
On that special workstation the administrator, my personal user and my
testuser can login and logout without any problems.
That user has a 200 MB user profile, login works fine (I assume out of
local copy) but while logout it waits for at least 10 min, hit a  error
"file could not be copied, network unaccessible" aftr click on OK, it
just tells me "roaming prfile unaccessible" and I could login as other us=
To be more exact: while login, it obtains tokens, user got a valid token
and can browse AFS space and even his profile in AFS space.

After the logout hung, I logged in as admin and did a fs trace -dump,
maybe it can be helpful, where to send it to?

>> 4. the context menu - nearly all workstations has it disabled as it
>> "send the explorer into a timeout" - from 1.5.x on til now for the mos=
>> active win users the context menu is a problem - right click on a file
>> OUT of AFS space keep the system freeze for ~60 sec til the context me=
>> appears. For files in AFS space it appears direct. Til yet I haven't
>> found any solution why it freezes, as in all my test accounts it just
>> doesn't appear, even on workstations on which the other users have tha=
>> problem
> provide a system this can be replicated on that is accessible via
> remote desktop and which has debugging tools for windows installed.
> You are not the first to mention this but no one has ever followed up
> with a system on which the problem can be replicated.  I can't replicat=
> the problem therefore I can't fix it.
> The explorer shell makes the equivalent of the call "fs whereis
> <object>" with the current directory being the folder that is active
> in Explorer.  Does this command when executed from the command prompt
> experience the same delay?

Problem is, it just happens with some users, not with all. I try to get
a user with that problem in which no sensible data is reachable/activated.

And it seems like the fs whereis doesn't show that delay, at least not
for that user in which context menu problem appears.

Lars Schimmer
- --
- -------------------------------------------------------------
TU Graz, Institut f=FCr ComputerGraphik & WissensVisualisierung
Tel: +43 316 873-5405       E-Mail: l.schimmer@cgv.tugraz.at
Fax: +43 316 873-5402       PGP-Key-ID: 0x4A9B1723
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org