System Physical Memory: 2G
Storage Subsystem: Raid 5 without write cache (this is a problem)
Avg CPU Usage: 16%
OS Free Memory: 1G
Avg Concurrent Session: 40
Application System: Web Application(Running on another server, JSP, Connection Pool)
Server Mode: Dedicated Mode
What's your opition? Is this instance running efficiently?
In my opinion, I think that waitevent of "log file sync" donot mean a problem.
Last edited by Calvin_Qiu; 04-04-2003 at 05:31 AM.
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
After 45 seconds of detailed analysis (what do you expect for free?) my guess is that because you have everything on the RAID5, disk contention is slowing down redolog writing. I used a config like this once and found that somehow (not sure exactly why) paging happens at the same time as log writing. Solution - find a couple of disks and put the redogs on a RAID1 pair, by themselves (or perhaps with the control files). I don't have write caching either - it does not appear to be a problem for me - in fact I prefer not to trust a $1 battery on the RAID controller . . .
"The power of instruction is seldom of much efficacy except in those happy dispositions where it is almost superfluous" - Gibbon, quoted by R.P.Feynman
I think that amount of "log file sync" do not really mean disk contention. Every log file sync operation will increase the waitevent number of "log file sync"?
12ms(avg wait time, I guess that I/O speed is reasonable) * 3.7 (redo synch writes) = 44.4ms (minor part of I/O capacity)
avg (Physical Read Time + Physical Write Time) per second close to 5ms
Windows Performance Counter of "Current Length of Disk Queue" keeps on 0.
Write cache will speed write operation dramatically on a Raid 5 system. How about a redundant battery which cost another $1?
Is it adequate of only one member per log file group in a Raid 1, 1+0, or 5 storage subsystem, which is redundant?
Last edited by Calvin_Qiu; 04-04-2003 at 10:40 AM.
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
Originally posted by Calvin_Qiu Is it adequate of only one member per log file group in a Raid 1, 1+0, 5 storage subsystem?
Surely it will only go as fast as the slowest?
"The power of instruction is seldom of much efficacy except in those happy dispositions where it is almost superfluous" - Gibbon, quoted by R.P.Feynman
Originally posted by marist89 95% of your waits are I/O related. I would look at the I/O subsystem...
Don't you think the waitevents of "log file sync" is reasonable as my thought in my previous post.
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
Originally posted by DaPi Surely it will only go as fast as the slowest?
If multiplex online redo logfile, it cost n*time to do log file sync operation if they locate in the same logical drive?
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
CURSOR_SHARING = FORCE
This is a dangerous setting.
Do you know the side effect of this?
Tamil
Yeah, It's a bbs system, all tables are not gonna be skew. On spit of this, It's a web application programming in JSP, most of queries coming with literal value.
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
Oracle Certified Master - September, 2003, the Second OCM in China
*** LOOKING for PART TIME JOB***
Data Warehouse & Business Intelligence Expert
MCSE, CCNA, SCJP, SCSA from 1998
Reply With Quote
