More timings populating a mailbox

[Ross Boylan]

On Mon, Dec 25, 2006 at 10:59:36AM -0500, Wesley Craig wrote:
> These are interesting results.  I encourage you to make an entry in  
> the Cyrus Wiki about the work you've been doing.  As an additional  
> filesystem option, we've found that IMAP on reiser improves  
> significantly with the notail option.  Good work.
> 
> :wes
notail is a lot faster: here are some new  timings, with some old ones
for reference
Threads  Mess/s 
  20      171.0  Reiser vanilla
  20      232.1  Reiser + fakesync
  20      199.8  Reiser notail
  20      356.4  Reiser notail + fakesync
  40      514.7  Reiser notail + fakesync (just to check scaling)

  20      185,1  ext3 vanilla
  20      267.3  ext3 noatime
  20      393.2  ext3 noatime + fakesync

Reiser with fakesync has much more variable times than the others.

Reiser also has a nolog option, which I didn't try.  There are quite a
few more knobs to tweak for all the filesystems.

The first time I tried Reiser + notail + fakesync the mailstore got
damaged; I was unable to determine the exact problem (I have a feeling
logging from the chroot I was running in didn't work), or to recover
from it.  I wiped the directories and started over.  No problems on
the 2nd attempt.

ext3 still seems the better choice than Reiser for these tests:
  * No corruption during testing.
  * slightly faster for migration (with fakesync) 
  * faster for safe use (w/o fakesync, though the timings above aren't
  of operations typical of daily use)
  * lower variance of times.

If I have the time, I'll try to update the wiki--can anyone change the
pages?

Ross