ADSM-L

ADSM on SP - restores at 9MB a MINUTE!

1997-08-05 15:14:00
Subject: ADSM on SP - restores at 9MB a MINUTE!
From: "Coleman, Barbara - WWH" <colemba AT WITCO DOT COM>
Date: Tue, 5 Aug 1997 15:14:00 EDT
PROBLEM:
The backup time of our files is acceptable5-8MB second.  The restore time is
not, at 9MB a minute.
My expectation of restore time is about double the backup time - not 50
times more!

OUR CONFIGURATION:
     Currently we  back up our files to a 3590 at a rate of 5-8MB/second
depending on file size
     We have an SP system and are backing up over the SP switch(TB3).
     Our tape unit is connected to node "A", and I am restoring files to
node "B".
     The ADSM server is a wide node with 256MB memory.

A 660MB took 1hour 14 minutes to restore to our 7135 running raid 5. (Yes,
we are attempting to get rid of the 7135). In this case the file was being
restored to a wide node with nothing else running on it.

The same 660MB file took 54 minutes to restore to our EMC disk which is
mirrored. In this case the file was being restored to a 4way SMP with
minimal usage.

The backup of this 660MB file took about 2 minutes.

The parameters on the ADSM server are:
     /usr/sbin/no -o thewall=65536
     /usr/sbin/no -o tcp_sendspace=65536
     /usr/sbin/no -o tcp_recvspace=65536
     /usr/sbin/no -o udp_sendspace=32768
     /usr/sbin/no -o udp_recvspace=65536
     /usr/sbin/no -o ipforwarding=1
     /usr/sbin/no -o sb_max=1310720
     /usr/sbin/no -o rfc1323=1

The parameters on the ADSM client are:
     /usr/sbin/no -o sb_max=1310720
          /usr/sbin/no -o rfc1323=1
          /usr/sbin/no -o tcp_sendspace=65536
          /usr/sbin/no -o tcp_recvspace=65536
          /usr/sbin/no -o udp_sendspace=32768
          /usr/sbin/no -o udp_recvspace=65536

The dsm.sys file on the client looks like:
     COMMmethod         TCPip
     TCPPort             1500
     TCPServeraddress   170.178.12.173
     TCPBuffsize         32
     TCPWindowsize      2048
     TCPNodelay          yes

The dsmserv.opt file on the ADSM server looks like:
      TCPWindowsize 2048
      BUFPoolsize 256

Any suggestions to speed the restore up would be greatly appreciated.  We
will be running a disaster recover test this week and according to the
results I am getting, it will take 60 hours to reload our database files.

Barbara Coleman
E-MAIL:   colemba AT witco DOT com
Phone: (203)552-2112
<Prev in Thread] Current Thread [Next in Thread>