Forum Discussion

Graeme_Welsh's avatar
20 years ago

multistream or not to multistream

have a large # of application servers, with 2 x 36Gb disks, RAID1 configured as 1 logical drive with C:\ and D:\ partitions, would enabling 'allow multiple datastreams' (policy using 'ALL_LOCAL_DRIVES' directive) have an adverse effect on backup times? or is it best to back a C:\ and D:\ in one stream?

likewise, where we have Virtual Machines, with C:\ and D:| (actually physically located on different LUNS on the SAN), is it best to disable multistreaming?

using disk staging... and netbackup 6.0 enterprise on w2k3

cheers
graeme

7 Replies

  • Lot of factors here.
    What speed is the network between the clients and media server?
    What is the throughput of your disks on the media servers?
    What is the throughput of your disks on the clients?
    How much data is on each partition?

    With multistreaming off, it backs up 1 item at a time, so for All Local Drives, it will backup System State, then C:, then D: with 1 bpbkar process.

    With multistreaming on, it will create 3 bpbkar processes and each one will backup one of the items.

    As an example:
    If you get 10MB/sec throughput on C and D on your clients and you have a GIG nic, you could increase performance by turning multistreaming on as long as your media server has 20MB/sec performance as the backup of C: and D: will happen at the same time. With multistreaming OFF. C: will backup at it's normal 10MB/sec and then D: will backup after C: is finished.
  • why not Bob? not even for the VMs? the C: and D: are located on different physical disks, albeit chunks of LUNS presented to our ESX servers...
    i can see the point with regards to app servers with 2 x disks as mirrors, with 2 x logical drives, however, id like to simplify policy creation, and attempt, where possible, a 1 size fits all idea, would performance be so badly affected?
  • Multistreaming would add 3 jobs to the jobs db for each server. This adds complexity to monitoring/redoing failed streams as opposed to a failed clients. The images are relatively small. I would simply set up 1 policy using ALL_LOCAL_DRIVES for all of the servers.
  • Bob is correct here. The only servers I use multistreaming on are ones with multiple VERY large drives. Usually C: is only 8-10GB for us and only takes 10minutes or so. We have some clients with 3-4 400GB drives and those we DO use multistreaming which saves TONS of time.
  • If those 400GB drives were actually 400GB logical partitions on a RAID 5 set, would multi-streaming make sense then? (I'm thinking no, but am not sure.)
  • It all depends on throughput.
    As an example, I have tested logical partitions of a RAID5 set with our LTO1 drives and can saturate the drives with 1 drive going to each stream. If we had LTO2 or 3 it would probably be a different story.