+10Team,
Iam troubleshooting a low throughput read issue while recovery . Need some clarification :
When recovery job is initiated in Commvault for a VMs(VSA backup) or databases( streaming backups) does it perform sequential read or random read .
For disk library , what type of disk policies is suitable for better read and write throughput :
| Disk Caching | Read Policy | Write Policy |
| Enabled | Read Ahead | Write Back |
Best answer by Onno van den Berg
I personally would set it to Write Through because once the first full have landed you will have benefit from deduplication resulting in much less writes. Doing so makes sure the data lands on the disks which is somewhat saver. Yes, it is somewhat slower but on backup I do not care so much because I prefer data integrity above write performance. In the end it all comes back to reads because once you have recover data then you want to have it back as fast as it can be!
We’re happy to provide additional assistance with troubleshooting the performance issues, but will need some log information to do so. Please let us know if we can be assistance.
Commvault’ s IO profile ranges from 40% read & 60% random during backup/restore operations.
As far as the recommended settings for the disk library, we would need to know the brand/model that you’re using to suggest recommendations. Also, things like raid configuration (if used), Commvault Version etc.
+22It will perform sequential read operations during recovery tasks. As
+10Thanks for the reply .
The underlying SAS disks are directly attached to windows media agent .
Virtual disk is created using 12 -- 8TB NL SAS disk 12 Gbps with RAID 6 Configuration.
Read Policy should be Read ahead or No Read Ahead ?
Write Policy should be Write Back or Write Through ?
+22I personally would set it to Write Through because once the first full have landed you will have benefit from deduplication resulting in much less writes. Doing so makes sure the data lands on the disks which is somewhat saver. Yes, it is somewhat slower but on backup I do not care so much because I prefer data integrity above write performance. In the end it all comes back to reads because once you have recover data then you want to have it back as fast as it can be!
+10Are you sure when recovery happens from disk library its sequential read from underlying disks and not random reads ?
Read Ahead is suitable only for sequential reads .
+22Yes, it is sequential but read ahead without having context with the data will not result in any benefit. This because in case deduplication is enabled hen it will write the unique data in a chunk file but there is a big chance that duplicate data is located in a different chunk file. so yes, the behavior is sequential but it doesn't mean Commvault will have to read all the data segments from the file.
Example: Commvault also offers read-ahead tweaks who offer similar functionality for a cloud library. Now we noticed during recovery operations that if you enable and tweak it aggressively that you will pull more more data from the cloud then required. We sometimes pulled over more than 150% of the amount of data for a particular restore.
Would be awesome if someone from engineering could bridge in here to verify if my dialog is correct and share additional information with us.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
