[clug] Ideas for future talks (and who wants to talk at next month's meeting?)

[David Price]

Andrew Pollock wrote:
> Care to share your horror stories? I'm a big fan of LVM, and would tend to
> advocate it anywhere (I have plans to give a talk about LVM when I get a
> chance to make some slides). I'm interested to hear about any badness...

The system is a file server with about 400 users' home directories on 
it.  It's running a 2.4 kernel with LVM 1.  It's possible the newer 
version of LVM with the 2.6 kernel might fix the problem, but at this 
stage I think the plan is to move away from LVM altogether, since we 
don't really need it.  Now on to the problem...

Intense disk access to an LVM volume seems to block other access. 
Writing a file on an LVM volume causes client machines to get NFS 
timeouts.  Writing a file to a non-LVM partition on the same RAID has no 
effect.  I think reading has the same issue, but didn't test it.

The end result is that the logs on all the desktops are filled with 
messages saying "kernel: nfs: server .... not responding, still trying", 
then, usually about a second later, "kernel: nfs: server .... OK".  If a 
disk intensive job is run on the server, users' machines can become 
completely unusable for the time the job is running.

It's as though the LVM layer only allows a limited number of concurrent 
accesses.

As I mentioned above, it's possible that LVM 2 fixes this, since I get 
the impression that the way it works is fairly different.

-- 
http://dmprice.com/