On 10/12/2012 09:55 AM, Yanko Kaneti wrote:
On Fri, 2012-10-12 at 09:40 +0100, Gordan Bobic wrote:
....
I use iSCSI (ext4 build area on one of the hosts for the packages that
fail self-tests on NFS) and NFS (for everything else) backed by a
reasonably beefy storage box.
Why no just iSCSI ? I am guessing you have some numbers to compare as
you usually do.
I haven't tested it, but I wouldn't expect much difference. NFS is
pretty efficient, and it's designed for that specific mode of operation.
If anything I'd expect it to be faster than iSCSI.
But the main reason I use it is because having shared storage is
convenient for a lot of reasons.
And what about kirkwoods with sata + average-speed spinning hd
vs NFS/ext4 over iSCSI.on a gigabit.
That would require a separate disk per builder, it'd still need some
shared storage for convenience, and the IOPS would never be as good as
what you can provide with a bigger storage box on the network. For
example, my storage box (shared, not dedicated to the build farm) has
13x1TB disks in ZFS RAIDZ2 (using ZoL) and 16GB of RAM. It seamlessly
churns out several times more IOPS than my small build farm can consume,
even during the I/O intensive operations such as extracting src.rpms,
and cleaning up build space.
For cleanup, iSCSI+ext4 might be faster, but ultimately I don't
particularly want to have to buy more disks. Having them all in one box
is convenient and plenty fast enough.
Gordan
_______________________________________________
arm mailing list
arm@lists.fedoraproject.org
https://admin.fedoraproject.org/mailman/listinfo/arm
