On Wed, 17 Nov 2010 08:10:33 -0800, Jonathan B Bayer
<[email protected]> wrote:
> I then did a simple test, copying a 4.3 gig ISO to each volume. I did
> this two times, and timed the second copy; this way the overhead of
> allocating space was eliminated.
>
> The results puzzled me:
>
> NFS: 5:08
> iSCSI: 5:54
>
> I was expecting the iSCSI to be faster, but this seems to indicate
> that it is slower.
>
> What am I missing?
First of all, these do take different code-paths. Inefficiencies in the
code can make for some real differences. NFS performance (unless you're
using NFSv4) is a well hammered out code-path. The state of Linux iSCSI
targets is still in flux. Yes, we've had them for a long time now, but
there are still several packages out there (IETD, SCSG, LIO-Target,
possibly others) and improvements are being made all the time. I don't
know which iSCSI package FreeNAS is using.
Also the initiator software can play a big role in performance as well. As
with the target, NFS has been around a long time and is a battle tested
protocol, where iSCSI is still kinda new. As Rob pointed out, a lot of
server-grade NICs can be converted into a hardware iSCSI HBA with either
BIOS or driver changes, which can provide even faster throughput than
using a software initiator would (or put another way, crappy softare
matters less).
NFS was designed around IP networks from the ground up, where iSCSI is a
port of a completely different signalling protocol to IP and brought some
baggage with it. Chief among them being the desired atomic transfer size.
On SCSI that's generally around 4KB, where the MTU for most Ethernet is
1500B, and that creates overhead that NFS handles differently. On a
correctly tuned network iSCSI should blow past NFS.
--
Law of Probable Dispersal:
Whatever it is that hits the fan will not be evenly distributed.
_______________________________________________
Discuss mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss
This list provided by the League of Professional System Administrators
http://lopsa.org/
