Yes, you are blowing out your shared memory.
Each job will attempt to configure 256K*256 (65MB) of shared memory.
If this is what you intend to use, then you will need to modify your
/etc/system file to ensure you have ample shared memory configured.
Typically, I don't see too many people use 256 Data Buffers, more along
the lines of 16, 32 and 64. With the LTO Gen2 drives you should be good
to go with the SIZE of 256K, but I would suggest you reduce the NUMBER
of data buffers down.
There's a tuning paper I wrote a while back during my life as a
consultant with another company that may still be around in the archives
of this list or the Book that Dave Little and I wrote on NetBackup
entitled Implementing Backup and Recovery that covers this subject as
well.
David A. Chapa * ADIC * 720.249.5836 * 720.272.8452 cell *
david.chapa AT adic DOT com
HIGH PERFORMANCE, SCALABLE, RELIABLE - http://www.adic.com/gopathlightvx
-----Original Message-----
From: veritas-bu-admin AT mailman.eng.auburn DOT edu
[mailto:veritas-bu-admin AT mailman.eng.auburn DOT edu] On Behalf Of Rongsheng
Fang
Sent: Sunday, July 18, 2004 7:19 PM
To: veritas-bu AT mailman.eng.auburn DOT edu
Subject: [Veritas-bu] bptm log: buffer settings
Hi,
We are running a Sun StorEdge L700 tape library which is connected to a
Sun Fire V880 (master+media server) directly via fibre. There are 6 IBM
Ultrium (Gen 2) tape drives installed in the L700. Solaris 9 is running
on V880 along with Netbackup DataCenter 4.5MP6. I have set the following
settings based on the performance guide:
On Master/Media:
NET_BUFFER_SZ: 256K
SIZE_DATA_BUFFERS: 256K
NUMBER_DATA_BUFFERS: 256
tcp_recv_hiwat: 256K
tcp_xmit_hiwat: 256K
On clients:
NET_BUFFER_SZ: 256K
But from the bptm log below, it looks like the network buffer was still
set to 64240 bytes, not 256k:
...
18:08:49.727 [9884] <2> io_set_recvbuf: setting receive network buffer
to 262144 bytes
18:08:49.727 [9884] <2> io_set_recvbuf: receive network buffer is 64240
bytes
18:08:50.417 [9713] <2> tapelib: SIGUSR1 from bpbrm while waiting for
tape mount
...
18:08:50.418 [9713] <2> io_init: using 262144 data buffer size
18:08:50.418 [9713] <2> io_init: CINDEX 6, sched Kbytes for monitoring =
20000
18:08:50.418 [9713] <2> io_init: using 256 data buffers
18:08:50.418 [9713] <2> io_init: child delay = 20, parent delay = 30
(milliseconds)
18:08:50.419 [9713] <2> getsockconnected: host=backupsrv service=bpdbm
address=10.15.0.51 protocol=tcp non-reserved port=13 721
18:08:50.419 [9713] <2> bind_on_port_addr: bound to port 59831
18:08:50.420 [9713] <2> logconnections: BPDBM CONNECT FROM
10.15.0.20.59831 TO 10.15.0.20.13721
18:08:50.422 [9713] <2> check_authentication: no authentication required
18:08:50.491 [9713] <2> mpx_setup_shm: buf control for CINDEX 6 is
0xfed89018
...
Did I miss anything in configuration (both Solaris kernel side and NB
side)? Or there is a OS limit on the size of the receive network buffer?
What needs to be done to ensure NetBackup can set and use 256K receive
network buffer?
Thanks for your help,
Rongsheng
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
|
