Richard L. Hamilton:
>> Since SPARC can boot over NFS without any local storage at all (and x86 could
>> probably do it with reasonably small flash storage for grub and a miniroot),
>> I can't see going that route very often.  The only use I see for iSCSI 
>> initiator
>> on Solaris is if one already has an iSCSI infrastructure (or has compelling
>> reason to contemplate that rather than an NFS server).   (OTOH, using
   
James Mansion wrote:
> After the best part of 2 decades suffering NFS performance for
> software development I'm a bit sick of it and keen to see if
> something more performant (where the fs is readonly, or private to
> the workstation) is available.  I can see the point of NFS for shared
> areas and home areas.  With a SVN server and an IMAP server, its not
> clear how much shared space is really needed - some, but not much.

Hi James,

The typical complaint with NFS is synchronous writes.  This is different 
for swap over nfs...

Using NFS for swap (1 remote file, no updates on meta-data) has a very different
performance profile compared to development on an NFS mounted source tree.  
(Lots
of different files.)

The sticky bit as mentioned above prevents updates of the swap file metadata, so
all that happens is blocks going back and forth.

The only bottleneck here is that you can't re-use the pages you write to swap 
until
the write is acknowledged by the server.  However, this is asynchronous to your 
application, and should not be noticable as long as the NFS bandwidth can keep 
up with the paging rate.   You should also not be *really* tight on memory, and
latency will not be a problem.

If however, your application suffers on reads,  you'll just have add more 
memory,
or add a local low-latency swap device.  Or you might add more memory to the 
server
and allow it to cache your swap by removing the sticky bit upstream.

I've done some pretty silly things with NFS diskless clients.  You just didn't
want them to boot all at once.  (We staggered the desktops memory test to cope
with that).


Cheers,
Henk

_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to