Performance gain after introducing collections

[Tobbe]

Hi.
I'm looking for any references when it comes to improved performance of EV backups after enabling collections when the storage is on regular NTFS shares.
I realize that there are quite a few variables that comes in to play regarding type of backup, backing to tape vs disk etc etc, so having general guidelines is not very realistic but if there are anybody that has any real life data I'm all the ears.

Thanks in advance.

/Tobbe

[Wayne_Humphrey · Accepted solution]

Tobbe,

This is my opinion.

I would not use collections as this introduces performance decrease to end users.  I always recommend using Partitions, Closed partitions are not written to so can safely be backed up as a once of, and then backing up new partitions as is.

Downsides of Collections:
End User Suffer with latency when viewing an item.
Index rebuilds take FOREVER, you better off compressing all the cabs and then indexing.

That's just my two cents, I will NEVER use cabs :)

--wayne

[MichelZ]

I agree with Wayne We do never recommend cabs. However, on the "backup perfomance" side, there is definitely an upside when you have to backup say 10'000 cabs instead of 1 Million DVS files ;) But as wayne said, try to keep your partitions in a reasonable size, this should keep your backup window compliant, as you don't have to backup closed partitions as often as open ones. (you still have to backup them from time to time!) Cheers

[Tobbe]

Thanks guys.

In this specific case, the issue is actually caused by the closed partitions.
They are backed up on a separate and obviously much less frequent schedule but the backup of the closed partitions is allocating backup resources during a long period, resulting in a disruption of normal backup windows. Primarily we are looking at flashbackups as an alternative solution but it would be interesting to see any real life figures from persons that has implemented collections.

Also, thanks for the hint about Index rebuild time.

/Tobbe

[Liam_Finn1]

Tobb,

The only time that I know of where Collections adds performance is when storing on a Centera device.

Since the data is going to be compressed into CAB files it means the contents of the file needs to be decompresed each time and it does slow things down for the end users.

If you are performing your backups on the local server and not accross a network then consider a higher capacity tape drive something like the new LTO4 drives

They will give you higher throughput to tape and also higher capacity on the tape so your backup window is decreased

The LTO4 can store 1.6TB and has a transfer rate of 120MB per second

We presently do full backups of 2.5TB for our indexes and SQL databases on our archive and we complete our backups every night