Rick Romero put forth on 1/15/2011 10:34 AM:
Also, if your filesystem is using 4k clusters, aren't you only using 1 random IOPS for a 4k email? It just sounds to me like if you plan 'smarter', anyone can avoid the excessive costs of SSD and get 'end user similar' performance with commodity hardware.
This depends heavily on which filesystem you use and the flow of mails/sec. For instance, if multiple writes of small (<4KB) files (maildir), or multiple writes to a single file (mbox) on an XFS filesystem are pending simultaneously, because XFS uses a delayed allocation scheme, a variable extent size, and because it is optimized for parallel workloads, it can pack many small files into a single 4KB extent (if they add up to 4KB or less) and write them to disk in a single IOP. Likewise, with mbox storage XFS can coalesce multiple writes to the same file extent in a single IOP. Moreover, XFS can take a pending write of say, 37 small files or write fragments to the same file, and if they fit within say, 12KB, it can write all 37 in 3 pipelined IOPS. Using O_DIRECT with mbox files, the IOPS performance can be even greater. However, I don't know if this applies to Dovecot because AFAIK MMAP doesn't work well with O_DIRECT...
***Hay Timo, does/can Dovecot use Linux O_DIRECT for writing the mail files?
Now, here's where load issue comes in. If the pending writes are more than a few seconds apart, you lose this XFS coalesced write advantage. I don't know how other filesystems handle writing multiple small files <4KB.
-- Stan