WAN Acceleration and what ifs

[dgapinski]

Hello,

I am looking to build WAN accelerators and establish off-site replication. As I build our servers, I have to ask, what if the SSD tanks? Does that screw the traffic or does Veeam just abide with a secondary source? Will Veeam send an alert for that?

Also, do you have any guidelines for figuring out at what point you'd want to separate out duties for replication/proxy/WAN acceleration?

Thanks!

[veremin]

As I build our servers, I have to ask, what if the SSD tanks? Does that screw the traffic or does Veeam just abide with a secondary source? Will Veeam send an alert for that?

You mean you're about to stuff WAN Accelerator with SSDs and want to understand what happens if those go down for some reason?

Also, do you have any guidelines for figuring out at what point you'd want to separate out duties for replication/proxy/WAN acceleration?

You can deploy those roles on one machine, as long as it meets combined system requirements.

Thanks.

[dgapinski]

Thanks - Yes that was the idea. The question is less important as I decided to build a mirror SSD in our server for this purpose, but I am still curious as to how it would behave.

For the second question, I'm more curious about finding out when one would be compelled to separate out roles - if you had a rule of thumb to offer, based on anything really.

[foggy]

For the second question, I'm more curious about finding out when one would be compelled to separate out roles - if you had a rule of thumb to offer, based on anything really.

In terms of system requirements, if you have a server that is capable of carrying several roles, let it carry them to prevent unnecessary traffic between servers performing separate roles.

Separating a proxy server role to a VM would allow to utilize hotadd transport mode.

[veremin]

If disk hosting Accelerator's cache becomes unavailable, the corresponding WAN Accelerator will be unresponsive, as well. Thus, a backup copy job using it will fail.

[dgapinski]

Thanks, that helps. I assume that a replication job would fail as well? That is will help us plan so that it doesn't by way of making redundant SSDs available.

That said, is WAN acceleration using an SSD's native compression scheme to squeeze the data down before sending it over the wire?

[poulpreben]

Hi,

There is generally speaking no usage of the cache on the source side of any WAN accelerator configuration. The only files present here are digest files describing the contents of any VM disk (being VMDK or VHD(X)). While the Backup Copy or Replication jobs are running, there will be created temporary payloads, which are also stored in the VeeamWAN folder, until they have been transferred.

On the target side, however, is where the cache resides. You mention that you are going for SSD drives -- what bandwidth are you accelerating? I am asking because our stress testing shows that you can pretty easily get 100 MB/s processing rates with even regular SAS10k drives. If you are planning on accelerating multiple sites in a many-to-one configuration, I would still recommend using SSDs though.

You are asking what will happen if the VeeamWAN folder is lost. Obviously, you would need to reconfigure the WAN accelerator to use another drive, as the inaccessible drive will cause failures. Assuming you have another drive ready, the following will happen:

• If possible, populate the empty cache manually from the backup repository. This will improve the performance of the next copy
• If populated, during the first run, a copy of hash table describing all blocks in the new target cache, will be copied to the source WAN accelerator
• If not populated, the target will be populated slowly - either by secondary lookup in the target repository, or by transferring the blocks over WAN once again

Thanks,
Preben

[dgapinski]

Thanks Preben, I appreciate the explanation - super helpful! I will keep SSDs in our hardware quote then and try replication both ways - it might be that for a WAN link of 40Mbps that platter drives are entirely sufficient, and we could use the SSDs for other purposes. Or not and sleep that much easier.

[poulpreben]

Thanks Preben, I appreciate the explanation - super helpful! I will keep SSDs in our hardware quote then and try replication both ways - it might be that for a WAN link of 40Mbps that platter drives are entirely sufficient, and we could use the SSDs for other purposes. Or not and sleep that much easier.

SSDs are cheap these days, so at least they will give you some additional peace of mind. Just make sure you follow the system requirements for WAN accelerators as well -- they will use the 8 GB RAM required in most cases.

You mention "replication both ways"; remember the WAN accelerator works in "many-to-one" scenarios -- not "many-to-many". This means replication from Site A -> Site B and Site B -> Site A simultaneously is not possible, unless you deploy two pairs of WAN accelerators. One of the jobs will be paused in the "Waiting for resources" state, until the other finishes. Since you are considering purchasing special storage for these components, I thought I wanted to make sure this is clear.

[dgapinski]

No - that wasn't clear. My thought was to try replication with or without SSD storage. No biggie - thanks again.