iSCSI synchronization (or replication) software over latent lines

[trent1980]

*I apologize ahead of time if this already up somewhere. I've seen a couple discussions but none of mention the open source software available ... which is what I'm looking for

We're looking to "affordably" (/free) get some layer of redundancy to our DR site. We don't necessarily need replication ... just something that can get the data over to our DR site (even if it's 24 hours behind). We've looked at products like DRBD and Pratima, but the latency is too high for us to entertain synchronous replication.

Has anyone had any luck with file or device level mirroring/replication between two iSCSI storage devices? If so, which open source packages have you tried (I don't have any experience with CVSup but it said it would do device level).

Our setup:

1. MAIN: HP DL380 (RAID 5 / 420gb) running CentOS 4.6 in Los Angeles (100mbps internet connection)

2. DR: HP DL380 (RAID 5 / 420gb) running CentOS 4.6 in Arizona (1.5mbps internet connection)

Any help is greatly appreciated.

Thanks --

[cmanucy]

Ever taken rsync out for a spin?

Can't get much more economical than that.

----

Carter Manucy

[trent1980]

yeah but only at the file level.

in our iSCSI environment, there is no partition ... just /dev/cciss/c1d1

can rsync handle that?

thanks for the reply

[cmanucy]

Ahh, I see... you're sharing out raw volumes via iSCSI over your Cent hosts. I keep the FS as EXT3, but present via NFS for some things.

Although I haven't tried it myself, vRanger replicator might do the trick for you.

----

Carter Manucy

[trent1980]

We'd like to go to an enterprise level replication ... eventually. Just

don't have the budget right now ...

Thanks again for the suggestions.

[vSeanClark2]

I can only imagine the reason for the CentOS boxes is to provide a SAN on the cheap. For that reason I would recommend replacing the CentOS out with DataCore SAN Melody. Sorry it's not an Open Source solution, but it is Open Hardware!!

1. Bring your Arizona SAN server to LA and load it with DataCore SAN Melody.

2. Migrate production SAN to Arizona SAN. Then reload the LA SAN with Data Core SAN Melody

3. Establish replication between your SANs (requires the HA software and licensing from DataCore)

4. Pause replication and move one SAN back to Arizona.

5. Start up Arizona SAN server and bask in the glory of asynchronous replication. (Just make sure your daily data change rate can be handled by that 1.5 Mbps line in Arizona).

Happy Replication!!

Sean Clark, VCP

[Andrew_Judge]

It's not feasible to replicate using DRBD or SanMelody HA over slow WAN links (or any HA synchronous block replication). Those are real-time block-level replication operations. It needs to write to the secondary (cache) before is sends an acknowledgement of a successful operation back to the host. It makes no sense architecturally. DataCore can do AIM (async IP mirroring) with snap which is one way and conducive to this environment to create crash consistent copies of block devices.

I do agree that DataCore is awesome. It is not open, but gets away from vendor lock on the SAN side and makes absolute sense.

As for open source, good luck inventing the wheel. I'm not opposed to the open stuff, but I'll leave the developement up to someone else...

Andrew Judge

MCSE: Security, RHCE, ACHDS, ACTC, CSSA, CCA, NCIE, A, Security, 3Com VoIP, VCP, VSP, DCIE

CEO, Grove Networks Inc.

Microsoft Gold Certified Partner

Certified Apple Consultant Network Member

3Com Voice Authorized Partner

Silver Citrix Partner

VMware Enterprise Partner

DataCore Authorized Business Partner

Phone: 305.448.6126

Fax: 305.437.7685

http://www.grovenetworks.com

[trent1980]

Thanks for the posts guys ...

Andrew -- I don't want to invent the wheel, I want to freely use someone else's until I can afford a real enterprise solution. I have a requirement of X uptime and Y budget. X = somewhere between 0 and 100%, and Y = 0$.

Field is open. I'm not asking for the impossible (like synchronous rep ... read my first post) ... just the options on what I can do to replicate one iSCSI box to another via a T1.

example1: Shut off iSCSI host 1, mirror each disk and ship images over the T1.

example2: Install DRBD, break DRBD, start DRBD on off hours ... break DRBD during business hours ... repeat daily.

example3: ________?

[Andrew_Judge]

Maybe I'm too in the box, but that won't work. It would take too long to sync over a WAN (if it ever would). For the Linux machines, tar or cpio into bzip then rync diffs or incrementals to another box over the internet. FTP would work too with cron and since the files are new, rsync really would not help.

Keep a VM on ice at the remote location, ready for a master and diff restore. Viola - maybe...

Other than that, the proper strategy is what I mentioned before. Slow links at the block-level are for snap and AIM. Sometimes it's good, but sometimes it is bad. Imagine a database defrag. Might be more compressed, but the blocks are moved everywhere, therefore more data moves.

Andy

[trent1980]

No doubt. I would be happy if I could just rsync everything over starting on Friday night ... we have til monday morning at 6am (60 hours).

The problem is that my iSCSI machines don't have partitions on the storage array. (just /dev/cciss/c1d1 ... NOT /dev/cciss/c1d1p1 with a file system).

Do you know of any tools for syncing a device? (There's got to be something LIKE drbd that can sync devices A-synchronously)