Experience using StoreOnce Catalyst

[JonArne]

Is there anyone with experience in using HPE StoreOnce catalyst in environments where 100+TB of data is backed up?

We're having some troubles where we find the synthetic full creation is to slow when the environment is at this size.

[Gostev]

Are you using 9.5? If yes, the cause for slowness is most likely outside of Catalyst (for example, in metadata processing). Thanks!

[JonArne]

Yes, we're using 9.5.

[tsightler]

Can you provide more detail? Synthetic processing with Catalyst certainly takes time, but that's true of of all synthetic processing really. Even with the ReFS fast clone, which is certainly the fastest synthetic processing in the product, I see it take 15-20 minutes for ~10TB on a 100TB ReFS volume formatted with 4K clusters, so 100TB could easily take 3-4 hours even there. With Catalyst the synthetic processing is quite CPU intensive on the StoreOnce itself (at least in my testing) so there are limits to the speed any given StoreOnce can perform synthetic operations based on it's hardware capabilities.

With any synthetic processing, it's best to spread out the schedule so that different jobs perform the synthetic operation on different days of the week to minimize the overall impact. Catalyst is not an exception to this recommendation. If you can share more details of the setup, and some more statistics on what performance you are seeing for synthetics, perhaps there might be more insights that can be provided.

[JonArne]

We're moving from another backup product to Veeam.
So far we've migrated approx 30% of the backups to veeam, but we see the time spent on creating synthetic fulls is to long to be able to process all backups during the week.

In our setup we have both Hyper-V and VMware and these are the servers in the setup:
13 proxies in vmware
1 repo server serving as mount server for the StoreOnce Catalyst store.
1 server running B&R and Enterprise Manager
6 Hyper-V servers
44 VMware ESXi servers

In Veeam we've created 32 jobs which backes up approx 40 TB of data. All jobs create synthetic fulls once a week, but the jobs are spread over all days during the week.
The largest job is approx 11 TB and that job used 8 hours to complete on it's last run.
In earlier runs, it has used 20+ hours to complete, but we changed the compression level from optimal to none a couple of weeks ago.

[kevin@wei.com]

Which StoreOnce unit are you using? Also, how many drives are in it?

The larger the unit and the more drives installed will greatly affect the performance. I will say that the 5100 I am using at a customer's site does take some time to perform a Synthetic Full. An example of a job would be a 3.2TB job with a 157GB incremental took just shy of 5 hrs to synthesize. Mind you, I have a bunch of other jobs running at the same times that will slow things down.

[JonArne]

This is a 5100 with 48 drives (4 encosures)