Our mail store is on a LeftHand SAN, which we bought this summer. The speed is pretty good, even on just a GigE network, and it's certainly a helluva lot cheaper than FC stuff. Downsides include the lack of an integrated fencing device for failover (most FC switches are fencing devices), and the general lack of clue in LeftHand's frontline support. Upsides include all of the cash that's still in our pocketses. Luckily, I don't think that a kernel-level implementation of iSCSI will be a big drawback, since Cyrus IMAP (and IMAP servers in general) aren't processor-bound. We'll run into I/O problems long before we run out of oomph to drive iSCSI and Cyrus IMAP. Chris St. Pierre Unix Systems Administrator Nebraska Wesleyan University On Sat, 29 Jul 2006, Greg A. Woods wrote: >At Wed, 26 Jul 2006 16:20:57 -0500, >Greg Harris wrote: >> >> On 7/26/06 3:33 PM, "Greg A. Woods" <woods-cyrus@xxxxxxxxx> wrote: >> >> > Using a SCSI host interface isn't going to be nearly so flexible as >> > using a Fibre Channel one, especially in the longer run (e.g. if you >> > ever want to add more storage, more storage controllers, share them >> > amongst multiple hosts, add backup devices and also share them between >> > hosts, etc., etc., etc.). >> >> If your really looking for flexibility, check out Left Hand Networks >> solution. Their claim is that they can beat FC stuff. Not having the need >> for speed at this point, I can't testify personally for how fast, but the >> solution looks like an awesome concept with lots of real world app. > >Left Hand Networks just sell an "open" iSCSI based SAN implementation. > >While iSCSI might be a more flexible solution at some time in the >future, it's still relatively bleeding edge, especially in terms of >wide-spread adoption and wide-spread installed base and existing >industry experience. > >I think Fibre Channel will still give one far more options for the >immediate future, including lots of options for low-end and used gear >too. > >Right now I think beating FC in the performance department depends on >buiding dedicated 10-GigE networks and using rather high-end processors >as iSCSI is typically still implemented deep in the OS, not yet in smart >controllers that simply make it look like a more traditional storage >device thus off-loading all the protocol handling to a dedicated control >processor. In other words the very things that supposedly make iSCSI >more flexible are also the same things which can easily hobble its >performance if one doesn't design one's infrastructure very carefully. > >-- > Greg A. Woods > >H:+1 416 218-0098 W:+1 416 489-5852 x122 VE3TCP RoboHack <woods@xxxxxxxxxxx> >Planix, Inc. <woods@xxxxxxxxxx> Secrets of the Weird <woods@xxxxxxxxx> >---- >Cyrus Home Page: http://asg.web.cmu.edu/cyrus >Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu >List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html > ---- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
