Native Replication

[mncheng]

Hello all,

I'm looking for a solution to implement replication cross over multiple country, and read the following statement in the overview document, is it ready?

--
Future: Native Replication. Zimbra clustering can be characterized as “single copy” failover in the sense that, from perspective of ZCS, there is only a single disk image to worry about (RAID mirroring is hidden to Zimbra). The Zimbra Collaboration Suite will in the future also provides “dual copy” failover that does not rely on underlying shared disk. ZCS native replication is likely the better fit for failover across a WAN, such as for multi-site disaster recovery wherein shared network attached storage may be infeasible. Zimbra replication is also well suited to commodity server and storage hardware.
--

Many Thanks.

[Krishopper]

Nope, Zimbra doesn't replicate anything between servers at this point. I believe Bug 11423 is the enhancement request. If that's what you're looking for, vote for it.

[ewilen]

That and Bug 42231 - HA/DR through Log Shipping

In the meantime, I've posted a bunch of links and ideas toward accomplishing something like the planned feature, at Setting up cold backup server

BTW, where did that text come from? Was it this PDF: http://www.01.com/architectural_overview.pdf

Note the date: 2005. Native replication also mentioned here as an enhancement planned for 2006.

I'm not carping, but if I were in marketing at VMWare/Zimbra/01, I wouldn't want those old things lying around.

[FritzBrause]

Depends on your IT Infrastructure. Mirroring your SAN- or SAS- storage would be the easiest way.

Another way is using high-tech like VMware ESX with Fault Tolerance:

http://www.vmware.com/products/fault-tolerance

https://www.vmware.com/files/pdf/per..._tolerance.pdf

If this is not an option for you you might want to look into using a DRBD- Setup. There are already some threads about setting up Zimbra Servers with DRBD:

[SOLVED] Zimbra on DRBD

The official DRBD Site is DRBD:What is DRBD

If you're not concerned about the delay, DRBD is working great - even big distances are no problem. PST (San Francisco) to Europe (Berlin) RTT is about 150-200 ms. The real troughput is 4-5 MBit/s per TCP Connection. That's o.k. for us with 250 Accounts per Server.

[ewilen]

That is very interesting--first time I've read of using DRBD over a WAN link for Zimbra. Are you using fully synchronous or asynchronous mirroring? Have you tested bringing up the the mirror?

However, this is not native replication, which is what Zimbra has been promising/planning for some time now, but hasn't yet materialized.

[FritzBrause]

Originally Posted by ewilen

That is very interesting--first time I've read of using DRBD over a WAN link for Zimbra. Are you using fully synchronous or asynchronous mirroring? Have you tested bringing up the the mirror?

We're using a DRBD over WAN setup for one of our customers between Berlin, Germany and Santa Clara, CA. AFAIK it's working in async mode without complaints for about 18 months or so. You may also be interested in LinBits professional WAN DRBD Proxy Solution: www.linbit.com: DRBD Proxy

Originally Posted by ewilen

However, this is not native replication, which is what Zimbra has been promising/planning for some time now, but hasn't yet materialized.

That's true. How about filing an RFE? I'll vote for it ;-)

[bofh]

and i dont think theres much more datatransfer with drdb than with a normal sync (which does not exist yet so no comparison just guessing :-)

but of course its not a backupo its a failover / and or / loadbalancing not a BACKUP :-)

(you never cant say this often enough, clusting is never a backup,.. never ever ... :-)

[FritzBrause]

Originally Posted by bofh

and i dont think theres much more datatransfer with drdb than with a normal sync (which does not exist yet so no comparison just guessing :-)

With 200ms latency you get 2-4 MBit/s average troughput for a single TCP- Connection. That's fine for us.

Originally Posted by bofh

but of course its not a backupo its a failover / and or / loadbalancing not a BACKUP :-)

In our case it's just a failover.

Originally Posted by bofh

(you never cant say this often enough, clusting is never a backup,.. never ever ... :-)

True, true, true :-)

[mncheng]

I'm thinking that to run the zimbra on two spearate hosts while the binary sitting on a distributed file system, such as OpenAFS or even NFS, any comment? Thanks.

[bofh]

how should that be possible?
you run an ldap database, an mysql database, an (if enabled) dspam database and so on. so those have to be replicated and the filesystem space too

i cant see any way to do this, cant see a good reason to make a "cluster" that way except poor knowlege (sorry im a bit harsch diodnt had my first coffee today)

its nonsense in every way naaa dont try that at home and do it only at work if you work for a bank or another kind of crime syndicate :-)