Restore speed on imported backups via DataDomain

[tpt]

Hi. We backup to a DataDomain DD510 using firmware 4.7. This replicates to an offsite partner, and we import the backups from there onto an offsite Veeam server, which restores onto remote ESX. We are having performance issues as follows:

1. When the offsite Veeam box creates a backup on the offsite DataDomain box itself (to a non replicated folder), restore performance is approx 37mbs.

2. When the offsite Veeam box imports a backup from the replicated folder and restores that, performance is much slower - we have never seen the speed go above 17mbs.

Is this expected? Does the offsite Veeam server perform worse for imported backups as it has to rebuild some kind of database information whilst the restore is in progress. Or do you think this is a DataDomain issue - that restores from a replicated folder will run slower than those from a non replicated folder? No replication was in progress during the restore testing. We just can;t work out why one is so much faster than the other (same VM both times).

[Vitaliy S.]

Hi Darren,

Do you restore to the same ESX host in both cases? I doubt that there is anything to do with the Veeam backup server, because imported backups should have the same restore speed in any case. Just in testing purposes could you please try to use any regular local hard drive in the DR site to backup/restore a VM? Do you see the same difference?

Thanks.

[tsightler]

In my experience restore speeds are limited primarily by the performance of the VMFS volumes (we've found we can restore tremendously faster to NFS datastores, 3-4x faster). Are you VMFS volume hosted on the same type and performance of disks? Are you using ESX or ESXi. Since your post says ESX, have you entered SSH credentials on both sides?

[tpt]

Hi, onsite I am using ESX but offsite its ESXi. Wondering now if ESXi (4.1) will offer the same restore performace? If anything, the offiste restore is using a SAN that is better than the one onsite, which for the same restore performs much faster. Same ESXi host used for restore each time...

[Vitaliy S.]

Wondering now if ESXi (4.1) will offer the same restore performace?

Well, please note that it is expected to have better performance while uploading data to ESX host due to service console unavailability on ESXi host.

By the way, if you're trying to predict how much time would it take you to perform a restore in case of a disaster, then could you please tell us if you've considered using Instant Recovery feature + Storage vMotion/Replication jobs instead?

[tsightler]

Hi, onsite I am using ESX but offsite its ESXi. Wondering now if ESXi (4.1) will offer the same restore performace? If anything, the offiste restore is using a SAN that is better than the one onsite, which for the same restore performs much faster. Same ESXi host used for restore each time...

My experience is that restores via ESXi are much slower than restores via ESX, on the order of 2-4x, based on the data. Basically, it's not really ESXi, but the fact that ESXi cannot leverage the Veeam agent. If you configure ESX to force agentless mode, I'd suspect that performance would be similar.