Unsolved
6 Posts
0
4377
Data Domain Cleaning Throttle
Hi,
We have a setup consisting of our Networker Server (version 9.2.1.2), 2 storage nodes and 2 Data Domain DD4200.
We now have the following problems:
- Until two months ago our Data Domains where on version 5.6.1 of the DDOS and when the cleaning ran on thursday we noticed an impact on our backups to tape (Yes, we still have them ). They were much slower than other days. We set our throttle of the cleaning to 60% which had a little more impact, but the cleaning was done faster. However extising jobs still went as slow as hell, but new ones would go at their normal speeds.
- We upgrade at the end of May to the latest target code of DDOS (6.0.2 then) with the promise that the cleaning had some serious improvements. However we have more problems ever since.
When the cleaning is running we now observer following things:
- It seems that the clone to tape is going even slower than before.
- The regular backup of clients is struggling to complete.
- We have a few SQL server that put their backups on a CIFS share on the Data Domain. But when the cleaning is running they are struggling to create a connection to this share and the backups fail. If they are retried some succeed, other still fail and need a third rerun. On other days it runs just fine. The SQL backups that are being done via Networker (using DDboost) are running fine.
We already set the throttle back to 40% and the problems still stay. If we set it further back to 20% will we benefit from this? Will the backups and clones go a bit smoother and with less errors. We don't care about the duration of the clean. We only have a obligation to get our backups on tape out the door at a certain time and that our SQL backups are consistent.
oldhercules
116 Posts
0
July 17th, 2018 04:00
Hm, this is interesting. We are using networker/ddboost only, and we didn't noticed any backup performance drop even if the cleaning was running with 95% throttle (on a DD990). Btw. you are lucky that you don't need to worry about the cleaning duration
I think some processes are heavily pushed by the cleaning - what I've seen is autosupport generate, it took much more time than it is usual when the cleaning was running. It seems CIFS server has the same challenge.
I think this problem worth an SR, the DD support guys are good, they can go deep in dd perf stats and maybe they'll able to find a resolution for you.
rugby01
85 Posts
0
July 17th, 2018 05:00
Cleaning is faster in 6.0.2, but even better on the 6.1.2 code. There were a lot of changes to the 6.0 code path and it might be Networking or routing issues. I know one customer that lost some routes when they upgraded because they were manually added. Make sure your ifgroups are still correct with all NIC's added, I've seen issue were upgrade and nic changes didn't make it past the upgrade. so please validate the entire boostfs stack end to end.
for Item 3 - If you upgrade to 6.1.2 you can move to BoostFS for windows. You will get faster backups and restores and reduce your client CPU usage.