Re: Load-Average with different backends on different filesystems

[Adam Williams <awilliam@whitemice.org>]

> On one server the load average went up to somewhere between 30 and 50,
> but all other machines (same config, similar amount of LDAP access and
> searches) stayed around 0. The only difference between the machines is
> the filesystem: with xfs, ext2 and ext3 there are no problems, but with
> reiserfs there is no productive use possible (at least with default
> configuration).

So it would seem pretty clear this is a kernel/reiserfs bug, or in fact 
that system is using different library versions (these subtle differences 
are hard to spot).

And, IMHO, your nuts using reiserfs on a production system - but whatever.

> After googling a bit I found out that it is known that the system does
> not perform fast on journaling filesystems and bdb with inproper cache
> settings, but since most accessing operations (max. one write per min)
> are read that should be no problem (especially if it performs well with
> other filesystems).

That it doesn't perform well with "improper cache settings" should be 
self-evident.  It performs JUST FINE on a journalled filesystem,  I doubt 
there is anyone left not using a journaled filesystem.  You can of course, 
tune your filesystems if the system is partitioned properly.

> After looking a while into Doc and FAQ I didn't find anything that
> states that there may be some filesystems that should not be used with
> bdb (but there are some postings in the archive), so maybe this info
> should be included somewhere.

It shouldn't make a huge difference unless the filesystem (driver, 
or structure on disk) is itself broken.

> After switching back to ldbm on this server all things are fine again.
> And hopefully I'll find the time to migrate the machine to xfs...

Good plan.