Forum Discussion

philalbert's avatar
philalbert
Level 4
9 years ago

Best practices for Multiplexing/Streams per drive

Hi,

We are trying to make our backups/restores more efficient and we are looking at Multiplexing and Streams per drive. What are the best practices in that regard? Here is our environment, fell free to ask questions if my description is not sufficient.

Our Netbackup environment is running on Windows 2008 R2 SP1.

-1 Master and 7 media servers

-16 LOT5 drives

-Most of our backups are using the MS-Windows policy type.

-We are also using Flashbackup for some servers where there’s a lot of files or a lot of data.

-We have around 800 clients.

  • We have found that MPX of 4 gives good backup performance and good restore performance.

    My mentor taught me to implement multiplexing in baby steps:
    Start with a value of 2.
    Record backup performance over the next week (full and incr backups)
    Do some restore tests and record performance.

    Increase MPX to 3.
    Perform same tests.

    Increase MPX to 4.
    ......

    Find an MPX value that works best for you.
    I have seen customers using MPX of 8 and they were happy with backup and restore performance.

  • Tape Multiplexing

    There is no real fixed rule on this.......... The more you multiplex then longer it takes to restore because the recovery process has to get each data fragment from a tape where they are no longer in a nice sequence suitable for streaming.   I'm running my libraries on a multplex of 24, which is unnaturally high. 

    Why :

    • Backups are not production and taking far too long.  More streams = quicker backup time.   Restore times for non-production are flexible.
    • The tape libraries needs more tape drives to increase throughput.  ATLs are not like disk-based backup targets which have a fixed max throughput capability.
    • Data delivery throughput to tape drives could be better.   The more streams you have to better the throughput. 
    • About 95% of the servers are small.  Increase the recovery time for a small server by 50% - is anybody going to notice the extra 30 mins? For the large and more critical systems I have reduced multplexing.

    Other Stuff

    Other things that make ATLs ineffiecient.

    • Having too many data retention times. NBU will only write to a tape where the data has the same retention time.  Keep the number rention times to the smallest number possible.  In NBU you can mixed retetion times on a tape but it is not recommmend.
    • Set the number of partially full tapes to = (number of tape drives +1 )  this applies to each data retention time. 

     

     

6 Replies

  • We have found that MPX of 4 gives good backup performance and good restore performance.

    My mentor taught me to implement multiplexing in baby steps:
    Start with a value of 2.
    Record backup performance over the next week (full and incr backups)
    Do some restore tests and record performance.

    Increase MPX to 3.
    Perform same tests.

    Increase MPX to 4.
    ......

    Find an MPX value that works best for you.
    I have seen customers using MPX of 8 and they were happy with backup and restore performance.

  • Exactly what Marianne said ...

    High MPX values do two things ....

    1.  Use more memory

    2.  Take longer to restore

    I had a case a while back where MPX of 15 or something mad was used.  Customer was complaining that the restore had taken 30 hours, was still running and was urgent.

    Somehow, and I've still not quite worked it out, but this was apparently my fault ... and what was I going to do about it.

    The answer was pretty much, 'nothing' because it's the high MPX value casuing the issue.

    Long story short, we rebacked up the data using MPX of 4 and reran the restore which then completed in just a few hours - can't remember exactly how long, but it was an acceptable time.

    The other question is, why are you using MPX.  If clients are able to individually run at a decent speed, then mpx'ing them together has no benefit.

  • Hi Marianne,

    I'll modify some storage units to use 4 streams and test the backup speed and restore like you suggested.

    mph999 to answer your question, why are we using multiplexing, we mostly use them on large backups, servers that have multiple large volumes to backup, 2-3TB each, and large SQL backup. If we don't use multiplexing we are not able to complete the backup in the allowed window.

  • Please remember to also increase policy schedules to 4 as well.

    The lower value of the 2 (STU and Schedule MPX) is applied when the job goes active.

    We have seen excellent results with MPX and multi-streaming on a customer's file server - full backup took entire weekend (starting Friday evening) up to Sunday evening.
    After implementing MPX and multi-streaming (and increased Max Jobs per Client), the Full backups completed shortly after midnight on Friday.

  • Thanks for the reminder Marianne.

    We are already using multiplexing, but from what I understand, we are using way to much. (Some policies are set at 10-16)

    What would be the ideal value for a policy that backs up standard Windows servers. Should we use all 16 drives and set the Multiplexing at 4? Or is using all the 16 drives for 1 policy not recommended but the other policies might not be able to get a drive?

  • Tape Multiplexing

    There is no real fixed rule on this.......... The more you multiplex then longer it takes to restore because the recovery process has to get each data fragment from a tape where they are no longer in a nice sequence suitable for streaming.   I'm running my libraries on a multplex of 24, which is unnaturally high. 

    Why :

    • Backups are not production and taking far too long.  More streams = quicker backup time.   Restore times for non-production are flexible.
    • The tape libraries needs more tape drives to increase throughput.  ATLs are not like disk-based backup targets which have a fixed max throughput capability.
    • Data delivery throughput to tape drives could be better.   The more streams you have to better the throughput. 
    • About 95% of the servers are small.  Increase the recovery time for a small server by 50% - is anybody going to notice the extra 30 mins? For the large and more critical systems I have reduced multplexing.

    Other Stuff

    Other things that make ATLs ineffiecient.

    • Having too many data retention times. NBU will only write to a tape where the data has the same retention time.  Keep the number rention times to the smallest number possible.  In NBU you can mixed retetion times on a tape but it is not recommmend.
    • Set the number of partially full tapes to = (number of tape drives +1 )  this applies to each data retention time.