Is my Oracle Server issuing more IO than it can handle
- From: Oracle Dba Wannabe <oracledbawannabe@xxxxxxxxx>
- To: Oracle-L@xxxxxxxxxxxxx
- Date: Wed, 8 Dec 2010 10:26:08 -0800 (PST)
Hi All,
So I was wrong about a different storage being used - well basically the
volumes
on the poor performing server have been mounted onto another server which uses
a
different san. The volumes are veritas file systems for the HP-UX platform. So
I'd assume the following would not really come into play:
1. size and number of redo log files - since these are the same volumes on the
server that perform poorly and the one that performs with acceptable performance
2. the possibility of a faulty disk - again its the same volumes so issues with
the disk would apply to both servers - since again same volumes
One difference I do see, however, is that the server that performs poorly has
32gb (and 5 procs) of memory whereas the one that performs well has 64gb (16
procs). Now both databases have an 8gb buffer cache (the good performing
server
however also hosts another oracle database). Now, that said, HP-UX has two
kernel parameters that control the file system buffer cache - which by default
(from what I've read) is pretty generous - is there a possibility that the
extra
memory on the server that performs well is being utilized by the file system
buffer cache and therefore allowing writes or the acknowledgment of those
writes
to perform quicker (no direct io being used) - compared to the server with less
memory?
Here is a side by side comparison of the load profile:
TPS Redo PhyW PhyR LogicalR BlockChanges
Poor Server 54/s 4.2mb/s 428/s 1983/s 13,000/s 9155/s
Good Server 225/s 7.4mb/s 564/s 243/s 110,000/s 7500/s
Here's a comparison of top 5 events for both servers:
Poor Server:
Top 5 Timed Events
Event
Waits
Time(s)
Avg Wait(ms)
% Total Call Time
Wait Class
free buffer waits 12,471,382 133,152 11 52.7 Configuration
log file switch (checkpoint incomplete) 50,033 38,433 768 15.2 Configuration
log file sync 198,985 30,749 155 12.2 Commit
buffer busy waits 46,927 17,681 377 7.0 Concurrency
db file parallel write 373 13,772 36,921 5.5 System I/O
Good Server:
Top 5 Timed Events
Event
Waits
Time(s)
Avg Wait(ms)
% Total Call Time
Wait Class
log file sync 821,418 35,503 43 83.2 Commit
CPU time 7,158 16.8
db file parallel write 128,596 2,851 22 6.7 System I/O
log file parallel write 235,049 2,845 12 6.7 System I/O
db file sequential read 179,777 727 4 1.7 User I/O
Other related posts:
