Oracle backup channels throughput

[shahriar_sadm]

Hi all, 

I have an Oracle databse backup with Rman script and store on tape library, previously Rman starts 8 channels and maximum stream per drive sets to 4. every 4 channels use one tape drive and their KBps was 35k. for reduce backup time I set maximum stream per drive to 2 and I was hoping 2 channles per drive an 4 drives involved can increase drive throughput and KBps. but any change not happend on backup performance. Please help and correct me if my idea was wrong.

Thanks

[RiaanBadenhorst · Accepted solution]

Hi,

Performance is not just about the target, there are other components involved.

If you explain what you saw, before and after your change in more detail it would be great.

Your source is just probably not capable of sending more than it is now (performance wise)

[Marianne]

Modern drives can write more than 100MB /sec. If 4 channels/streams are not writing close to 100MB/s, then spreading jobs across more drives are not going to help. Please tell us more about the data path. Is Oracle server configured as (San) media server or backups going across the network to a media server? If network, then tell us about network connection between client and media server. Thanks!

[Michael_G_Ander]

Start with checking how fast you can read with rman, backup validate is the command if I remember correctly.

Think you best bet is to get the DBA involved in this, as Netbackup is just recieving the rman output through the api and there a lot things in the database that impacts the backup speed like pools inside SGA and of course the rman parameters.

If it is a incremental backup, running Block Change Tracking or not can also have a big impact.

Is other databases running the machine and/or disks that might impact backup of this database ? often seen multiple instances causing poor performance because there no room left for thing like backup.

[Nicolai]

Have you tuned Netbackup SIZE_DATA_BUFFERS/NUMBER_DATA_BUFFERS ?

These buffer has a big impact on tape performance, and could improve the general tape performance.

That said, there should be other limitation in the infrastructure as well.

NUMBER_DATA_BUFFERTS = 256

SIZE_DATA_BUFFEERS = 262144

http://www.mass.dk/netbackup-guides/netbackup-buffer-tuning-2/

Buffers are configured on media servers and must be configured using the same SIZE_DATA_BUFFEERS across the entire Netbackup domain.

[shahriar_sadm]

Hi Michael,

Yes, you are right. but this backup time increase from 6 weeks ago. and DBA said we have no change on database or server. I think maybe data growing cause this problem or any buttleneck in database server or network, after check this side I will involve DBA to this issue, 

Thanks

[shahriar_sadm]

Hi Nicolai

No, I never set this parameters, I have more than 60 media servers, is there any way to set those confing one time instead set to all media servers? 

Thnaks

[shahriar_sadm]

Hi Marianne

Before change we have 4 channels per drive with 32MB/s that write close 130MB/s and after I change multiplex to 2 we have 2 stream with 32MB/s per drive and drive write speed is close 64MB/s. 

I want to decrease this backup time now, It starts 8 channels and every channels write 1.1TB in about 10 hours and then for remaining 1TB start a single channel and backup duration up to 20 hours. 

Are you recommend increase channels number that backup can start more than 8 channels at time or can I manage channel max write size and up to 1.5TB?

This Oracle server setup as media server and directly connected to tape library. 

Thanks 

[Nicolai]

Not in Netbackup.

You need to look at automation tools then or maybe a scriped soloution.

[RiaanBadenhorst]

Create touch files as required on one server, back it up. Then perform a restore to all media servers using bprestore from CLI. If you back it up to disk the restore will be super quick.

[Nicolai]

LOL - that's creative, but will work :)