[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: mail queue's, ext3 data=journal and sync-mount

I decided that this message is better for Debian-ISP, so I replied to the 
list and BCC'd you.  I hope you don't object.

On Sun, 18 Aug 2002 01:20, you wrote:
> I'm having some trouble finding info on this stuff and found a
> knowledgeable-sounding post of yours on debian-isp.  Please ignore if
> this is inappropriate.
> I'm going to setup a debian woody system for qmail (and others) and was
> hoping to secure the qmail queue a bit.  For some reason, sync mounted
> ext2 and 3 is abysmally slow.  How does FFS do it?  anyway, do you know
> if ext3 mounted data=journal (or any other way) is sufficient?

I have not done any serious tests on such things.  If you want to do such 
tests then I recommend my Postal benchmark program (it's in Debian).  But 
don't try the IMAP option - I haven't finished coding it.

Rumour has it that data=journal can actually improve performance in some 
situations.  If a program is writing lots of small files synchronously (quite 
common for a mail server that has one tiny control file for every message, 
and the average message file isn't too big) then journalling the data allows 
for synchronous writes to a small (8M to 32M) region on disk (which is really 
fast) and it'll then be written to it's final destination with the write-back 
caching enabled which allows writes to be ordered for good performance.

I haven't tested this but it all sounds logical.

Without the data=journal option ReiserFS is rumoured to beat Ext3, with 
data=journal ext3 should win.

I don't know much about FFS.

I would be interested in seeing benchmark data.  Also one thing I have been 
thinking of doing is benchmarking Qmail vs Postfix...  ;)

Russell Coker

Reply to: