Deduplication for Backup and Recovery
by Hu Yoshida on Dec 29, 2006
I recently saw an article by Beth Pariseau about reasons why people are not buying deduplication products. The main reason given was the slow speed of backup compared to non deduplication VTL solutions. Compared to tape libraries and VTLs which top out around 500MB/s, the Diligent ProtecTier solution is cited at 220 MB/s. While that is better than the 100MB/s that is cited for Data Domain, it is judged to be too slow.
I have another take on this. The speed of backup was a concern when backup was done during backup windows. Today most users will create a snapshot or clone copy and do their backup in the background without the need for backup windows. The main concern today is the recovery speed of backups. For recovery, we achieve 400 to 500 MB/s using one ProtecTier in front of an AMS 1000 modular array. So while the backup time does take longer today, it can be done using a shadow image copy, without impact to the application, and the recovery times are still comparable to other VTLs. The major advantage of Diligent’s deduplication is to reduce the amount of data backed up by a factor of 25 to 1, and that is money.
In the meantime Diligent recently released a Feature Pack for ProtecTier that, among other enhancements, introduced performance enhancements which has shown sustained backup throughput of 260MB/s with 17% new data on an AMS 1000 disk system. This is close to 1 TB per hour.
The referenced article also cites unnamed Diligent reps as saying that adding clustered servers to the ProtecTier VTL to boost performance is on the products roadmap.
Comments (4 )
500GB/s? you sure it is not a typo?
Thanks Ming Zhang, you are correct that was a Typo which I corrected.
I know some VTL de-duplication(Sepaton,etc) can be done AFTER the backup done, do you think it is a better way?
I believe the backup numbers achieved at Goldman’s were significantly higher than the 200 MB/s referenced. Is this true? Also- Diligent said clustered heads were roadmapped for Year End 2006….why has this now pushed to Q4 2007? eal enterprise customers highlight the SPOF with a single head.