Subject: Re: [vserver] NFS shares or iSCSI
From: Eugen Leitl <eugen@leitl.org>
Date: Mon, 12 Dec 2011 14:03:13 +0100

On Mon, Dec 12, 2011 at 01:55:45PM +0100, Herbert Poetzl wrote:
> On Mon, Dec 12, 2011 at 12:47:55PM +0100, Eugen Leitl wrote:
> 
> > I would like to put a zfs hybrid box I have for vserver
> > production, to improve IOPS of the underlying disks.
> 
> so some kind of raid/pool setup with ssd/memory caches
> or something like that?

Yeah, it's an 8 GByte RAM 4x 1 TByte Solaris box
with an Intel SSD cache.
 
> > 1) would you use iSCSI or NFS for vserver guests, for
> >    better performance and/or flexibility?
> 
> iSCSI will result in 'block' transfers (similar goes for
> AOE) so that requires a filesystem on the client with all
> the filesystem overhead, but it might provide better
> performance for 'certain' workloads

Which workloads in particular? I'm worried about
workloads like IMAP with Maildir. If thens or
hundreds vserver guests are running on a single
machine there bound to be several users fighting
for spindle access at the same time.
 
> > 2) is just mounting /home e.g. via NFS for each 
> >    vserver guest a better idea?
> 
> than having the entire guest on NFS?

Yes.
 
> also, the 'zfs hybrid box' is a different machine than
> the one running the guests?

Yes. Connected via GBit Ethernet. I can enable jumbo
frames.
 
> please clarify the setup, otherwise it's mostly hand
> waving and probably useless for your setup

Thanks, the questions you've asked are helping me already.
 
-- 
Eugen* Leitl <a href="leitlhttp://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE