Copy Job Speed using WAN Accel

[MCU_Networking]

Looking for any guidance on whether we can get any faster on transfer time of our Copy Jobs

We have 2 servers that are Repository servers, repo1 and repo2.
We have 4 servers that are wan accelerator servers, repo1, repo2, accel1, accel2, all set to Low Bandwidth mode
We have Site1 which houses repo1 and accel1
We have Site2 which houses repo2 and accel2
Our Network traffic rules in Veeam are set to 50Mbps.
Our Firewall at both locations has bandwidth limiting./shaping set to 50Mbps
We have a connection/MPLS between the two sites than is 100Mbps and low latency, 20ms and under, usually under 10ms

We currently have 2 copy jobs
Copy Site1 to Site2

• Source Accelerator: repo1

• Target Accelerator: accel2

• Virtual Machine Count in job: 35

Copy Site2 to Site1

• Source Accelerator: repo2

• Target Accelerator: accel1

• Virtual Machine Count in job: 13

Current methodology:

• I chose the Repository servers as the sources because they are stand-alone servers with more resources (20cpu/64GB) and I read(bp.veeam.com) that Source WAN Accelerators take up more resources

• Target WAN Accelators are virtual machines in the same site, which have 1GB connections to the Repository servers, they have 4CPU/GB, but I can give more

• The WAN Accelators are set to Low Bandwidth mode because it looks like, according to the settings, High Bandwidth mode would be for links greater than 100Mb/s.

Results:

• Both Copy Backup Jobs say the Bottleneck = Target WAN

• Site1 to Site2 takes 4hrs with a Processing Rate of 82MB/s, Processed 587GB, Read 295GB, Transferred 32GB

• Site2 to Site1 takes 3hrs with a Processing Rate of 36MB/s, Processed 148GB, Read 75GB, Transferred 9GB

Main questions:

• Anyone have a similar setup but but a change that increase the speed noticably?

• I am thinking about making the Repository servers the target WAN accelerators instead, since the Targets are the bottleneck. Maybe this would increase the speed. Thoughts on whether this might make a difference?

• If I have 100Mb/s available, how much of a difference do you think it will make if I set the Firewall and Network Traffic Rules(Veeam) to 100Mb/s compared to 50Mb/s? I was trying not to flood the pipe, but these copy jobs are off hours, so it isn't too big of a deal.

[HannesK]

Hello,

I am thinking about making the Repository servers the target WAN accelerators instead

yes, that makes sense. No need for additional VMs (and the VMs are too small, because there is no memory left for the operating system. 8GB is system requirement for the WAN accelerator)

I heard that firewall-based shaping can have impact on Veeam performance. So switching it off can be worth testing.

Predictions on speed change are hard. Testing "high bandwidth mode" also makes sense. Did you already test different streams settings?

Best regards,
Hannes

[MCU_Networking]

Switched it to where the Source WAN Accelerators were the separate virtual machines (in same site as source data) and then Target WAN Accelerators were the destination Repositories, it however did not seem to make a difference in the first couple of runs. *Clarification* Based on your comment Hannesk you might have thought I meant taking the additional VM WAN Accelerators out of play, but that is not the case. I have one Job going each direction, so the 2 Repository servers cannot handle both jobs at once. That setup fought for resources and the Jobs took about 12(Site1 to Site2) and 6(Site2 to Site1) hours.

I would hope the firewall setting wouldn't affect it since it is limited to the same Mbps in Veeam Network Traffic Throttling.

I upped the Network Traffic Throttling Rules in Veeam to 100Mbps and changed the firewall to 100Mbps too and it did not seem to make a difference.

I tried High Bandwidth in a couple of smaller tests and it seemed to perform worse than low bandwidth mode. I know I tested it in 50Mbps setup, I think I also test it in a 100Mbps setup.

[MCU_Networking]

Switched it to where the Source WAN Accelerators were the separate virtual machines (in same site as source data) and then Target WAN Accelerators were the destination Repositories, it however did not seem to make a difference in the first couple of runs.

I also want to add that it still says the Bottleneck is the Target WAN.

[HannesK]

there is always a bottleneck

Testing different settings is the only practical way from my point of view and there is no way to go beyond physical limits (just guessing)

[MCU_Networking]

there is always a bottleneck

I get that, but I can't find Veeam documentation on what Target WAN bottleneck means. Can you point me to any documentation that goes over the different Bottlenecks and what they mean?

[HannesK]

https://helpcenter.veeam.com/docs/backu ... ck&ver=110 documents that "target WAN" means the "WAN accelerator deployed on the target side."

[dbewernick]

I get that, but I can't find Veeam documentation on what Target WAN bottleneck means. Can you point me to any documentation that goes over the different Bottlenecks and what they mean?

Have a detailed look at the usual suspects like disc latency, CPU and memory usage on the WAN accelerator machines. I would recommend to have a very detailed read again through https://bp.veeam.com/vbr/2_Design_Struc ... rator.html to make sure sticking and checking to the BP recommendations.

Also as Hannes mentioned, the shaping of traffic through a firewall can cause trouble. Have a check if you disable it completely and not setting it to higher values.