On Thu, 2003-10-23 at 10:13, Greg Whynott wrote:
> I am putting together an Courier IMAP server and would like to squeeze
> all the performance I can from the array as there will almost 200
> persons collecting mail from the machine. As you may know Courier
> stores all mail as individual files, averaging in our environment
> between 2 and 4K each.
> The RAID 5 array is across 7 disks on a FC controller. Would anyone
> have a recommended set of arguments to pass mkfs.xfs while creating the
> the md raid was created using a chunk size of 32, you think this is ok
> or should it be smaller?
> I was going to use this line to create the xfs fs:
> mkfs.xfs -f -b size=2048k -d agcount=16 /dev/md0
-b size specifies the basic block size of the filesystem, it is
in bytes, I do not think 2048k is going to work. Why choose 2K?
If you are going to switch from the default of 512 then the best
bet is 4096.
If you are specifying an external log, then you will need to specify
its size, you might want to go above the default in any case.
Specifying an agcount is probably not a good move, right now an
allocation group is limited to 4G bytes, I would let it do its
own thing. If it picks up the stripe infor from md correctly, it
will round robin the start of the allocation groups across the
spindles. This spreads the metadata I/O load, there are some blocks
at the starts of allocation groups which are hit pretty frequently.
> also I am considering putting the metadata log on another device.
> any thoughts on this would be really appreciated.
> thanks for your time,
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx