https://support.emc.com/docu50233_OneFS-7.1-Backup-and-Recovery-Guide.pdf

max. supported: 64 concurrent streams per BA

recommended: 4 concurrent streams per BA

Over here we have made good experience with 4 streams (full backups, NDMP level 0)

or upto 8 streams (diff/cumul or NDMP levels >=1) on the OLD BA model and OneFS 6.5.

Hi Peter,

thanks a lot ...it helps me to design the backup concept.

Could you say something about the Isilon performance impact if four NDMP Streams are running?

This cannot be easily answered; it depends, as usual...

The NDMP streams will take some IOPS away from the drives,

and if your cluster is busy, these can just break the camel's neck,

performance-wise.

We run the backups during off hours including

weekends, and the impact has never been critical for us.

Our cluster (3 x X200, 4 x 108NL/NL400) is usually

"moderately" loaded, but very unevenly,

mostly 10-30% of ideal "peak" performance.

When a MultiScan or AutoBalance job is running, we see

some effect on the backup though; the NDMP throughput

rates are less stable, and drop by perhaps ~20% on average.

Hi Peter,

thanks a lot for your answer...it helps me to become more feeling for designing the backup strategy for my customer.

Regards from Germany (we are now in the final quarter )

Peter

Cheers from a German in Shanghai

Peter

The recommendation of "three storage nodes per BA" is actually the same for all models from S nodes to HD nodes in the most recent version of the document. This only make sense if one acknowledges where this is coming from.

Doing the simple math it says one storage nodes contributes 160 MByte/s to the backup throughput.

That's a small fraction of what S210 and X410 models can deliver, but you would be on the safe side when doing backups together with normal traffic and restripes ongoing. If your nodes are otherwise idle during the backup window, they can drive much higher backup throughput -- /theoretically/ keeping one or even two BAs busy per single storage node, perfect for very short backup windows.

On the other hand you might NOT want backups to draw more than 160 MByte/s per node from the HD400 model, considering that these nodes need considerable amount of "headroom" for restripe jobs. Window-wise you can't backup say 300 TB worth of user data (per node) at 160 MByte/s (per node) over night, so assume the backups will run most of the time simultaneously with front-end traffic and restripes.

This shows how "three nodes per BA" is "more like a guideline", and is subject to a more thorough assessment in each case.

Hope it helps -- feel free to share more details on your particular configuration if you like.

-- Peter