Anyone used WAN acceleration tools to replicate over WAN?

[typoman]

Anyone have experience using Veeam Replication with WAN replication tools? I've read about HyperIP, SilverPeak, Riverbed, Cisco WAAS, and a few others to see how well they might improve Veeam Replication over the WAN. I currently replicate a few VM's across about a 10 MBit link, but I want it to be faster. Anyone have real experience with some metrics they could post?

[tsightler]

We have Cisco WAAS but I haven't really tested it much yet with Veeam. Interestingly enough though, just I started testing replication of a couple of servers at one of our remote sites with Veeam. That site has WAAS and a 10Mb link, so I may have some stats very soon.

[Gostev]

Wow, I wonder what happened today? This was never discussed before, and now 2 topics in one day
Someone already created a thread about Riverbed a few hours ago: VEEAM with Riverbed

[tsightler]

OK, here's some initial numbers to get an idea with. We decided to replicate a production VM for the testing. This VM is an application server that really doesn't get a lot of changes. For a typical 24 hour period it would be about 2GB of VBR to backup and it's a 70GB total space (2 drives, a 20GB system drive and a 50GB drive for the application/logs) but a lot of this space is currently unused.

WAN Info: 10Mb MetroE fiber, MPLS, ~40ms latency, ~700 miles between sites

Without WAAS:
Initial Replica -- 12.8GB transferred via WAN -- Total time: 3hr 12min 17sec
Second Replica -- 948MB transferred via WAN -- Total time: 23min 45sec

With WAAS:

Initial Replica -- 4.8GB transferred via WAN -- Total time: 2 hours 22 minutes
Second Replica -- 149MB transferred via WAN -- Total time: 11min 21sec

For the WAAS here was the compression rates:
Initial Replica -- Veeam transferred 12.8GB, WAAS optimized to 4.8GB -- Total compression 62%
Second Replica -- Veeam transferred 901MB, WAAS optimized to 149MB -- Total compression 83%

So, at least initially, it made some difference in speed, and a significant difference in the amount of traffic which actually crosses the WAN. These are only very initial numbers, but I'm pretty happy with them.

[Gostev]

Very impressive results! Clearly makes huge difference...

[tsightler]

More data:

Third Replica -- VBR size 1.2GB, Veeam transferred 2GB, WAAS optimized to 168MB -- Total compression 91%

The third replica was after two day's of changes. It took around 22 minutes for the entire job, which is reasonably impressive since, assuming there was no acceleration, it would have taken about 29 minutes to transfer the data fully saturating the 10Mb link. Instead it took 22 minutes (probably 2-3 minutes were for snapshot creation and removal) and used only 10-20% of the links bandwidth.

[jgravity]

Has anyone completed a live test of this process. I mean an actual Restoration.

[Gostev]

As far as I know, most of our customer do periodic backup and replica testing. Generally it is a common best practice to test your disaster recovery plan (nothing to deal with virtualization really, backup testing is as old as backup itself).

Testing replica with Veeam is really easy:
- Disable the Veeam replication job temporarily
- Use VMware Infrastructure Client to create snapshot on replica (so that VMDKs do not change during the testing, and you can continue replicating)
- Use VMware Infrastructure Client to edit replica VM configuration and change VLAN from production to test network
- Power on replica VM with VMware Infrastructure Client, and see if it can boot up properly
- Shutdown replica VM
- Use VMware Infrastructure Client to discard replica snapshot (to revert replica to original state, as before the testing)
- Enable the Veeam replication job back

UPDATE: Replica testing procedure with screenshots
http://www.virtualizationteam.com/veeam ... edure.html

[tsightler]

Has anyone completed a live test of this process. I mean an actual Restoration.

Can you clarify what you mean by "restoration"? This thread is about replication. You can restore files from a replica, and yes, we've done that with WAAS, it works fine and has a minimal impact on restore performance via the WAN (I'd say 30-50%). Other than that, there's really no such thing as "restoring" a replica. You can fail over to a replica, and yes, we've done that as a test, but that's not really a restoration. If I have a running failover replica, I can certainly replicate it back to the original site, so if that's what you mean by "restoration" then yes, we've done that too. WAAS does a great job in this case speeding the reverse replica since a lot of the data is already in the WAAS cache.

[Bunce]

Some stats using Riverbed:
http://www.veeam.com/forums/viewtopic.p ... 607#p15480

Cheers,
A

[typoman]

Anyone have experience using Veeam Replication with WAN replication tools? I've read about HyperIP, SilverPeak, Riverbed, Cisco WAAS, and a few others to see how well they might improve Veeam Replication over the WAN. I currently replicate a few VM's across about a 10 MBit link, but I want it to be faster. Anyone have real experience with some metrics they could post?

A year and half after I created this post I have some data to share. One of my smallest clients wanted to do Veeam replication of a few VM's to a DR site. I setup Veeam at the DR site to pull the VM's from production. I just got it going last night and its still doing the footprint of some big VM's, so I don't have logs and stuff to share yet, but I wanted to share what I'm seeing already.

The production site is in a bad geographical area for telecom and has a cable modem with effective 1.7Mbit upload. The ping times are horrendous and the circuit gets hammered by users. I can replicate only at night. I throttle the replication in Silver Peak down to a trickle during the day.
The DR site has a 5Mbit down / 5Mbit up internet connection.
Over time, production will migrate to the DR site. Establishing reliable VM replication is going to be key to that migration.

I settled on Silver Peak as the WAN acceleration tool. They have had a virtual appliance option for some time, but only recently was there a big price drop in their virtual appliance vs. their physical appliance. I also really like how their smallest appliance will accelerate up to 4Mbits of bandwidth. This covers my SMB clients with crappy dsl, a full T-1, or 3 bonded T-1's. It makes it easy for me to sell and easy for my clients to add a little bandwidth without having to upgrade their WAN acceleration appliance.

I will add more detailed logs later, but this sums it up almost exactly:
Before Silver Peak: 1 GB per hour
After Silver Peak: 2 GB per hour

I got this rate by closely watching the datastore browser in vSphere client for several hours and watching the vmdk files grow.

I just realized that I still have compression and deduplication enabled in the Veeam job. I will play with these settings and report back. Silver Peak generally recommends that you turn off compression/dedupe in the source product if you can. Silver Peak self-reports its stats on how much it is helping. It is reporting about 50% reduction in data sent over the WAN.

[Vitaliy S.]

Thanks for sharing your experience with Silver Peak, but...

I just realized that I still have compression and deduplication enabled in the Veeam job.

Replication doesn't use deduplication and compression, because an original, unchanged .vmdk should be presented to a target host. Compression and deduplication are only applied to rollbacks (VRB files).

[rniggemann]

Replication doesn't use deduplication and compression, because an original, unchanged .vmdk should be presented to a target host. Compression and deduplication are only applied to rollbacks (VRB files).

Then it is very deceptive of the product to offer the option when configuring a replication job.

Is there any plan to enhance Veeam Backup and Replication so that one could have a replication appliance at both source environment and destination environment that would allow at least compression?

And I'm sure someone's mentioned it before, but it would be tremendously more efficient if one could take a backup and then replicate that to the destination, rather than effectively having two backups.

[Vitaliy S.]

Is there any plan to enhance Veeam Backup and Replication so that one could have a replication appliance at both source environment and destination environment that would allow at least compression?

There are lot of changes coming to our replication engine in the next release, so please stay tuned to our forums to know all the information first.

And I'm sure someone's mentioned it before, but it would be tremendously more efficient if one could take a backup and then replicate that to the destination, rather than effectively having two backups.

Many of our customers have been doing exactly this with RSYNC.

[Gostev]

Is there any plan to enhance Veeam Backup and Replication so that one could have a replication appliance at both source environment and destination environment that would allow at least compression?

Well, if you are talking about network traffic compression, then it is in fact present. Despite the actual replica is uncompressed, its data does flow as compressed. However, this is only available for "fat" ESX targets today.