Data Domain and Read Stream Count

[CurtLeCaptain]

Just for a better understanding/knowing if there is a way to do this, we've noticed on our DD2500 that performance was... mediocre at best when running an entire VM restore. When we spoke with EMC, they noticed that per restore job, Veeam consumes only 8 read streams within the Data Domain. They're saying we can use up to 50 total - is it possible to modify the number of read streams used when we want to restore, say via some registry setting or whatnot?

[Gostev]

As of today, Veeam should be using 1 read stream per restored VM. Increasing the number of read streams is not even applicable, as with image-level backups, the restore is done sequentially (block by block of the restored disk). Perhaps they are confusing this with file-level backup and restore, where you can restore multiple files at once. We are however looking to add parallel disk restore in the future versions. Thanks!

[Delo123]

+500 for parallel disk restore

[CurtLeCaptain]

So what we're seeing is that per-restore, we see 8 connections to the DDBOOST repository. Say that job runs at 10-15MB/second, we start another restore job and it consumes 8 more connections. That job also can run at 10-15MB/second without having an effect on restore #1. So we're trying to figure out, can we increase that count to say, 16 connections per restore, that way we can get 20-30MB/second/job, increasing our restore rate. If you'd want numbers to prove, I have statistics that show this happening.

[rreed]

Don't forget that reading data back out of a deduplicating device (rehydrating, decompressing, etc.) is going to be very slow compared to how fast they can ingest. This is the hangover of buying into dedupe devices we're all experiencing. Your restore speeds are matching what we're seeing out of our 1Gbps connected DD620's and DD2200's, and don't get amazingly faster restore speeds out of our 10Gbps connected Dell DR4100's, though v9 is noticibly faster at restores than v8 was.