slow rman backup

[Kisad]

Hi Friends,

MY DBA is complaining that the rman  backups are taking too long to complete. We have the NIC on the database servers set at 1000MB full. What else should i be checking? Will adding SIZE_DATA_BUFFERS and NUMBER_DATA_BUFFERS on the database server help increase the speed.

Please help

Thanks

kisad

Logs from the DBA Email sent to me.

------------------------------------------------

00:02:03 INF - Name socket = media server.IPC:/tmp/vnet-18348292569322883040000000000-dwaG1J;11df17b224a8f41c169d7e0d6e902cc1;4;300

00:02:03 INF - Job id = 5220334

00:02:03 INF - Backup id = client2.bu_1292569122

00:02:03 INF - Backup time = 1292569122

00:02:03 INF - Policy name = client_RMAN_VTL

00:02:04 INF - Snapshot = 0

00:02:04 INF - Frozen image = 0

00:02:04 INF - Backup copy = 0

00:02:04 INF - Master server = master

00:02:04 INF - Media server = media server

00:02:04 INF - Multiplexing = 32

00:02:04 INF - New data socket = media server.IPC:/tmp/vnet-18346292569322699678000000000-ltaq1J;be6dddbbaf22b6ae3f4ae6a2ba46b490;4;300

00:02:04 INF - Encrypt = 0

00:02:04 INF - Use shared memory = 0

00:02:05 INF - Compression = 0

00:02:05 INF - Encrypt = 0

00:02:05 INF - Keep logs = 30

00:02:05 INF - Client read timeout = 1800

00:02:05 INF - Media mount timeout = 0

04:38:45 INF - Waiting for mount of media id M50531 on server media server for writing

04:39:24 INF - Mounted media id M50531

[oramtp@client2:database2] /usr/openv/netbackup/logs/user_ops/dbext/logs

[Marianne]

What kind of transfer rate are you seeing for RMAN backups? And for filesystem backups on the same client?

SIZE & NUMBER_DATA_BUFFERS is only used on a media server. NET_BUFFER_SZ can be configured on clients.

Before you adjust buffer settings, you need to determine where the bottleneck is:

Oracle, Filesystem, network, media server, backup devices...

Do you have bptm log on the media server?

Check bptm log to see what buffer settings are used. Then, look at the end of the RMAN backup for 'Waited for full|empty buffers'. If you see 'waited for full', you need to find the problem on the client.

Test with filesystem data first - use something like tar to backup +- 2Gb to /dev/null combined with 'time' command, e.g.:

time tar cf /dev/null /folder

Divide /folder size ('du -sk /folder') by time taken to run backup.

If read speed from disk is satisfactory, test network transfer rate - ftp +- 2Gb of data from the client to the media server.

Let us know what the results of these tests are.

Please also share RMAN settings, such as 'filesperset' and number of channels defined in script.

[AAlmroth · Accepted solution]

Immediately from your log, it would seem that it takes until 04:38 before a tape is mounted. This is normally a good indicator that your have run out of available tape drives. The Oracle job is sitting and waiting for over four hours before it can actually start.

The performance once started is probably at acceptable speed though. You would have to look into finding a backup windows where not all of your tape drives are occupied with other backup/restore/duplication operations.

/A