Server = Fedora 17 running on a MBP4,1 with 4GB RAM and a WDC Scorpio Blue. NFS 
setup as async,rw in /etc/exports and in /etc/sysconfig/nfs I set 
RPCNFSDCOUNT=32. Underlying file system is XFS on LVM on LUKS.

Client = Mac OS X 10.6.8 and 10.7.4, on MBP8,2 with 8GB RAM and a WDC Scorpio 
Black. NFS mount is:
sudo mount_nfs -o resvport,rsize=32768,wsize=32768 
192.168.1.137:/home/chris/Music /Volumes/NFS

Reads from server appear to saturate the server disk at ~70MB/s. For writes, it 
oscillates dramatically, from 0MB/s to 110MB/s. I think what's happening is the 
server caches til RAM is full, and then gets stuck for a bit while it's 
flushing to disk and isn't receiving data for a few seconds. I don't know why 
it doesn't just settle down instead. Could be LUKS encryption, as the CPU in 
the MBP4,1 doesn't support AES-NI.

With NFS set to sync, I get substantially less performance, around 30MB/s. AFP 
is around 34MB/s. In all cases, the files are over 10MB, so random read/writes 
isn't a factor.

I am not using jumbo frames.

The router is a Linksys WRT600N running dd-wrt, in a fairly stock 
configuration. Cables are CAT5e, and one of them is kinda crap. So, the fact I 
can sustain 110MB/s until the server cache is full, without much in the way of 
network optimization and all commodity hardware, is pretty interesting.


Chris Murphy
_______________________________________________
MacOSX-admin mailing list
[email protected]
http://www.omnigroup.com/mailman/listinfo/macosx-admin

Reply via email to