Richard Elling wrote:
On Jul 23, 2014, at 4:54 AM, Miles Fidelman via illumos-discuss
<[email protected]> wrote:
Looks like AVS, or maybe OHAC, is in the ballpark - but I can't seem to find
anything in the way of documentation, status, or anything - other than some old
email threads that indicate that both look pretty dead (circa 2011). Sigh....
The SNDR portion of AVS does block-level sync and async replication. It is a
dead
project because such architectures worked well when disks were 9GB, but suck
most
heinously when disks are 4TB. I'll lump DRBD and HAST into that group too.
Have to say - my experience is otherwise, at least w/ 2TB disks. Granted
rebuilds take time, but that's true with any kind of RAID, but for
synchronous writes to a pair of drives - works just fine. Particularly
when you're only synchronizing selected selected volumes within those
drives.
What you will find is that they are easy to setup, easy to fail over to the
remote site,
and impractical to return to the original site.
Actually, it's the setup that's a pain. Failover and recovery work just
fine.
Mind you, I'm not doing multi-site - just multi machine. Avoid the need
for SAN for basic HA of critical VMs.
As Robert says, the only way to handle multi-site redundancy is to implement
above
(application) or below (ZFS). At Coraid, we have the technology to do the
below: wide-area
mirroring. Joyent's customers, and many cloud apps, tend to implement above.
Being
stuck in the middle is a bad place to be.
Love to do everything at the application level - but that falls down for
some critical applications - notably mail and list processing (which is
what our HA cluster is for). It's a lot easier just to set up a VM on
one machine, with a failover VM on another - replicate the spool
directories with DRBD, and it all just works. I'd love to find a way to
do the replication at the application layer, but so far have yet to find
a good model.
For the spool mirroring - without investing in a SAN - DRBD wins big.
There's no other free/cheap solution. Nobody seems to hit the middle
market - small clusters serving small organizations (a department, a
lab, small offices -- a few 1U servers in a rack) - there are LOTs of us
in this segment, but it's not clear there's much money to be made, other
than by commodity hardware vendors. A sweet spot for Supermicro servers
- but they mostly get loaded with open source software. Your stuff is
great, but it's way overkill for us.
--
In theory, there is no difference between theory and practice.
In practice, there is. .... Yogi Berra
-------------------------------------------
illumos-discuss
Archives: https://www.listbox.com/member/archive/182180/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4
Powered by Listbox: http://www.listbox.com
