LACP/bonding between Backup Repo and DataDomain appliance

[casaubon]

Hi,

I have a physical VBR6 server with Backup Proxy and Backup Repository role enabled. It works with the Direct SAN access transport mode. The Backup Repository destination is a drive where is mounted a CIFS share provided by a DataDomain Appliance (Deduplication setting : local, compression setting : none).

The DataDomain Appliance is connected to a CISCO switch with a 10 Gbps link. The VBR6 server is connected to the same CISCO switch with 4 Gbps links: I want to agregate this 4 links to have a 4 Gbps bandwith and overall, i want VBR6 using this 4 Gbps bandwith (and not only 1 Gbps).

My questions are :
1) what type of agregation should i use in order to let VBR6 use all the bandwith ? LACP, EtherChannel ?
2) How works the Backup Repository ? can it use multiple TCP sessions or something else in order to spread the data accross all the 4 links ?

Thank you for your help !!

[Gostev]

Hi

1. Not sure about this one. Need help from somebody with good knowledge of networking please!
2. No, this is not possible today (at least easily, without some network-level tricks). Veeam will connect to the specified/resolved IP address (single NIC).

However, on the other hand I do believe that Veeam will be able to saturate the IOPS capacity of your device through the single 1GBps link. Deduplicating storage devices are not that fast anyway.

Thanks!
Anton

[dellock6]

About link aggregation, you cannot in any way go beyond the physical link on a single ip-ip connection, if you have 1 Gbit beetween two systems, a connection from 1 IP address to the other 1 IP address will always use 1 link.
You can use whatever link aggregation model you want, but you will take advantage of it only by using multiple TCP sessions to/from a device. this is way link aggregation works well on hypervisors or switches, where you are going to have multiple mac addresses running at the same time. The only real solution is to install a 10 Gb nic in the VBR6 server.

Anton, about the link usage, with 1 gbits link it's only needed to have 125MBs traffic to saturate it, and I've seen several scenario where the backup can go over that number, even with medium storages.

Luca.

[Gostev]

Luce sure, but not when we are talking inline deduplicating storage... these are generally much slower than the raw storage. Of course, depending on the model, the ingest rate may be faster than 1 Gbps wire speed. In that case, you can just create multiple backup repositories, each pointing to different NIC on the storage appliance, and use concurrent jobs.

[dellock6]

Right! In fact this is another method to workaround the 1-to-1 limit on etherchannel.
About the speed, I tested post-deduplication (ExaGrid) and is much faster than 1 gbps, I never tested a datadomain so I do not now how much fast they can go, but I would suppose that having 10G network card they can go over 125 MBps.

Or that 10G connection is only a marketing lure

[Gostev]

Sure ExaGrid will be very fast - as it does not do inline deduplication, but rather post-process deduplication. This makes it very different from any other major deduplicating storage. Originally, all data goes to the so called "landing zone", which is basically normal raw storage (very fast). This landing zone is also why ExaGrid makes it one of the best target for Veeam backups (think vPower based functionality). And I know they are going to have one more Veeam-specific joker in their sleeve very soon! Love those guys, great partners!

[dellock6]

Ahah, so I need to talk with them to find out what this joker is
I love their technology too..

[Gostev]

Yep, sorry cannot tell - I know this under NDA.

[Daveyd]

Yep, sorry cannot tell - I know this under NDA.

Exagrid customer here as well. Can't wait to hear what it is