ADSM Performance across HPS
1995-10-20 18:38:35
We are running ADSM V2.1 on an IBM SP2 across the high speed switch to a
pair of 3490 tape drives attached to one of the nodes. We are primarily
concerned with backing up Oracle Databases, so some of the files are rather
large. We have changed the network options according to recommendations
found on the SP home page. They now look like this:
dog_ticks = 60
lowclust = 200
lowmbuf = 500
thewall = 6000
mb_cl_hiwat = 1200
compat_43 = 1
sb_max = 1310720
detach_route = 1
subnetsarelocal = 1
maxttl = 255
ipfragttl = 60
ipsendredirects = 1
ipforwarding = 1
udp_ttl = 30
tcp_ttl = 60
arpt_killc = 20
tcp_sendspace = 65536
tcp_recvspace = 65536
udp_sendspace = 65536
udp_recvspace = 65536
loop_check_sum = 1
rfc1122addrchk = 0
nonlocsrcroute = 1
tcp_keepintvl = 150
tcp_keepidle = 14400
tcp_keepinit = 150
icmpaddressmask = 0
rfc1323 = 1
ipqmaxlen = 50
tcp_mssdflt = 512
directed_broadcast = 0
The dsm.opt file is:
nodename switch05
subdir yes
LANGUAGE AMENG
NUMBERFormat 1
REPLACE ON
TAPEPROMPT NO
TIMEFORMAT 1
The dsm.sys file is:
SErvername switch01
COMMmethod TCPip
TCPPort 1500
TCPServeraddress 198.113.200.129
TCPBuffsize 24
TCPWindowsize 256
TCPNODELAY YES
TXNBYTE 25600
COMPression OFF
The dsmserv.opt file is:
COMMmethod TCPIP
TCPPort 1500
MAXSessions 20
LANGuage AMENG
DATEformat 1
TIMEformat 1
NUMberformat 1
MIRRORREAD LOG NORMAL
MIRRORREAD DB NORMAL
MIRRORWRITE LOG PARALLEL
txngroupmax 256
TCPWindowsize 256
TCPBuffsize 32
txnbyte 25600
tcpnodelay Y
devconfig devconfig.back
volumehistory volumehistory.out
A recent backup of one of the Oracle databases yielded:
Files Backed Up: 13,815
Bytes Transferred: 2,552,244 KB
Transfer rate 553.635 KB/sec
This transfer rate is much lower than it should be across a high speed
switch. Are there any other tuning recommendations available? Suggestions
or information would be most welcome.
Russ Brooks
|
<Prev in Thread] |
Current Thread |
[Next in Thread> |
- ADSM Performance across HPS,
Russ Brooks <=
|
|
|