RMAN output buffer size on Sun Solaris with BLKSIZE parm.

From: Edward Maynard <ed.maynard_at_notes.fairchildsemi.com>
Date: Wed, 12 Dec 2001 18:50:44 +0000 (UTC)
Message-ID: <30633446d956ac9f6794878e4d2b6c5b.50760@mygate.mailgate.org>

Hi,

Here is the message that I have sent in to Oracle:

SBT_TAPE channel, output buffer size for Sun Solaris

I'm trying to tune my RMAN backups. I'm on a SUN E4500 and I'm using RMAN with Legato Storage Manager. The tape library uses DLT7000 tape drives. The network connect is a 100 megabit connection and I'm only able to get 3 tape drives max, usually only 2, from my particular pool during any backup/restore.

Because of the network, the max throughput that I'm seeing is 10m/sec total. I have target sessions and parallelism set correctly and I'm seeing the results of both.

We are going to be getting a 1gigabit pipe from the DB server to the backup server. This will greatly increase the throughput. I'm also going to try and get at least 3 channels and maybe even 4 once we get the pipe.

I've been experimenting with the BLKSIZE channel parm. I'm able to set it to 2m but no higher. If I try to set it any higher, say 4m, I get this:

RMAN> run

2> { 
3> allocate channel a1 type 'SBT_TAPE' 
4> parms 'BLKSIZE=4194304 ENV=(NSR_SERVER=fsce15-c, 
5> NSR_DATA_VOLUME_POOL=OraData Dev)'; 
6> allocate channel a2 type 'SBT_TAPE' 
7> parms 'BLKSIZE=4194304 ENV=(NSR_SERVER=fsce15-c, 
8> NSR_DATA_VOLUME_POOL=OraData Dev)'; 
9> backup full 
10> (database); 
11> release channel a1; 

12> release channel a2;
13> }

RMAN-03022: compiling command: allocate 
RMAN-03023: executing command: allocate 
RMAN-08030: allocated channel: a1 
RMAN-08500: channel a1: sid=15 devtype=SBT_TAPE 
RMAN-08526: channel a1: BMO v3.0 

RMAN-03022: compiling command: allocate 
RMAN-03023: executing command: allocate 
RMAN-00571: =========================================================== 
RMAN-00569: =============== ERROR MESSAGE STACK FOLLOWS =============== 
RMAN-00571: =========================================================== 
RMAN-03007: retryable error occurred during execution of command: allocate 

RMAN-07004: unhandled exception during command execution on channel a2 RMAN-10035: exception raised in RPC: ORA-19554: error allocating device, device type: SBT_TAPE, device name:
ORA-19557: device error, device type: SBT_TAPE, device name: ORA-21117: Message 21117 not found; product=RDBMS; facility=ORA Additional information: 13
Additional information: 17943024
RMAN-10031: ORA-19624 occurred during call to DBMS_BACKUP_RESTORE.DEVICEALLOCATE Is this because I'm hitting a wall/max size for the output buffer size? In note 73354.1, it says that the output buffer size is o/s dependent and on Solaris, the default is 64k. I'm using BLKSIZE to override this in order to increase the output to the backup server/tape. I'm assuming that increasing this increases the amount of data that I'm pushing to the tape. Can someone clarify this for me and also tell me where I can find the max value for output buffer size for a particular platform/machine? Is there any reason why I can't increase this value?

I've done calculations and I'm finding that the size of my input buffer per channel is 33M and my default output buffer size for Solaris will only give me 256k! I want to be able to get this closer to 33M. Wouldn't this be a correct assumption/goal?

Thanks in advance, Ed.

-- 
Posted from twmaine-208-3-251-17.twmaine.com [208.3.251.17] 
via Mailgate.ORG Server - http://www.Mailgate.ORG

Received on Wed Dec 12 2001 - 12:50:44 CST