NetBackup Multiplexing for Oracle RMAN Backups

Solution 1:

The first thing to check is how much network (TCP) throughput your server can handle. Use netcat, etc. If it is less than around 30 MB/s, multiplexing from network is of no use to you, and my further advice can be ignored. Work on tuning your network throughput instead. Now, to the point.

The LTO3 drive, just like any other linear tape drive, works well only when it gets a stream of data with a certain constant throughput.

The tape is passing under the head at a high speed, and you don't want to stop it. At each stop the drive has to perform lengthy procedure: decelerate to full stop, accelerate back, pass the end-of-data point, decelerate again, accelerate forth to reach the end-of-data point. When data is not feeded by NetBackup fast enough, the buffer underruns frequently and so the drive has to stop/rewind/start frequently. The performance is hurt dramatically. This is called "start-stop" operation or "shoe shining".

Drive adjusts the speed of the tape somewhat, but not very much, it can drop to about 50% of maximum speed.

The whole point of Netbackup multiplexing is to provide a better streaming throughput and avoid start-stop operation. Check the throughput of your RMAN backup, if it is 30 MB/s or less you have a classic start-stop operation.

Now, let me make one thing clear. If you do not have start-stop I would not recommend multiplexing RMAN backups at all. RMAN is complicated enough without multiplexing. I don't want to mess with RMAN, I want my restore to be as fast, easy and seamless as possible.

However, if you find your backup throughput unacceptably low, I would suggest implementing around three multiplexing streams for starters. Increase the number each night until you will not gain any more throughput. And make sure each stream is coming from the different disk spindle(s). Not from different partitions/tablespaces/filesystems/databases/servers/LUNs/other-virtualization-layers. These matter little, if any. Physical disk spindles. If you feed many streams from the same spindles you will just cause thrashing and overall performance will drop even more.

Note: NetBackup theoretically can also de-multiplex a restore. If I remember correctly, it pauses a little before a restore, to give a chance for more restore attempts to launch. In this case they will run jointly, just like multiplexed backups. But please verify this with a manual, I am only 90% sure on this one.

Solution 2:

That depends entirely on if your oracle server can move data fast enough to keep the LTO3 drives streaming. I do not multiplex Oracle data because the large files get served up fast enough to keep the drives running at an acceptible speed.

Before we replaced the Oracle servers however, and they only backed up at about half their current speed I did in fact multiplex them.

It is important to note that restores are slightly slower when multiplexed with NetBackup, but not much slower. I know for certian you can de-multiplex on a restore. We do it all the time both to do restore tests, and on rare occasion to actually replace lost data.

I strongly recommend testing both ways and seeing if you can keep your LTO3 drives moving fast enough without multiplexing.