Hi
Problem:
The throughput of the duplication phase is very slow and only one job per tape drive. The total throughput to the tape library is about 120 MB/s avaraging about 30-40MB/s per tape drive.
____________________________________________________________________________
The SLP setup is:
Backup :: Local Data Centre [Source---10GbE--->Media Server, DD OST plug-in]---- Stretch LAN (2x10GbE) -----> Remote Data Centre [DataDomain]
Duplication :: Remote Data Center [DataDomain ---10GbE---> Media Server ---8 GbFCP---> Tape Libvrary , 4x LTO-5 all 8 GbFCP]
The idea is to use the DataDomain OST to reduce the backup traffic across data centre link. The DD plug-in should eliminate a substantial amount of IP because we are getting an avarage of about 93% data reduction from deduplication. I would estimate the traffic reduction across the link to be about 15% of the original data size. Once the data is staged on the remote Data Domain its copied from the to a tape library. All the processing for duplication job stays within the remote data centre.
The remote data centre media server and Data Domain have no resource problems (the max utilisation of any resource is < 12%) and the slowest data transfer interface is 8Gb/s (FCP). The remote DC media has been clocked at running sustained backups almost 8GbE to the Data Domain. Tapes are multplexed to 12 streams as the bulk of the data is squillions of smallish VM machines that don't have to be recovered in a hurry becasue its not production.
The deduplcation phase batches up the backup jobs as expected but it appears the number of primary backups per SLPbatch is huge. This results in very long running single stream jobs and a lot of duplication queing.