| latch: session allocation and CPU [message #290775] |
Mon, 31 December 2007 13:53  |
|
|
I have serious problem on DBMS.
Sometimes it has lots of 'latch: session allocation' waits and DBMS performance getting lower.
At that time total connections are almost 4000 and active connections are between 300 and 400.
Yes maybe the reason should be there are a lot of session on DBMS.
However, that amounts of connections are not first time event. but lots of 'latch: session allocation' wats are started on this december.
I doubt on high CPU utilization but as you people know that latch is kind of memory lock. but on DBMS, 'latch: session allocation' is problem. There was some latchs but it is release quickly and it's fewer.
On H/W machine, CPU utilization is abnormaly high. When issue occured average CPU utilization is more than 90% and memory utilization is about 60%.
That's why I think there could be some kind of performance relationship between CPU and 'latch: session allocation.
and I have no idea about that.
If you people have any information between CPU and 'latch: session allocation, please let me know that. I appreciate.
P.S. The RDBMS version is 10.2.0.2 and SGA MAX and target is 20Gb.
[Updated on: Mon, 31 December 2007 14:06] Report message to a moderator |
|
|
|
|
|
| Re: latch: session allocation and CPU [message #290886 is a reply to message #290778] |
Wed, 02 January 2008 00:49  |
|
|
I'm not sure your recommendations are suitable for this issue.
It would be grateful, if you explain how new ASH help me to investigate this issue.
Also, SGA target is recommendation from local oracle. Our DBMS system had low buffer hit ratio and we have enough server memory, so local oracle recommand that increasing SGA target upto 20Gb. Before we had 8Gb for SGA target.
[Updated on: Wed, 02 January 2008 00:50] Report message to a moderator |
|
|
|
|
|
|
|
| Re: latch: session allocation and CPU [message #290937 is a reply to message #290897] |
Wed, 02 January 2008 04:12 |
|
|
Hmmm....
I don't understand why the focus of this issue is moving into SGA target. If the large SGA target could be the reason of lot's of 'latch: session allocation' issue, please let me know.
I want to show you the local oracle reports about our DBMS but I don't have any right to do. Mainly performance issue on SGA target is Physical Reads. Local oracle consultant thought Physical Reads on our DBMS can be reduced to increase SGA size. Also the DBMS BCHR is lower and our server has lots of memories so even if we increase SGA size 8Gb to 20Gb, memory utilization will be increased just 10% more. So we decided increasing SGA target to 20Gb.
As the result, Physical Reads reduced 75% and BCHR increased to 98%.
Because of large SGA, our DBMS performance is better. I don't think on-line transaction performance is better but we have heavy batch programs then that programs performance is get better.
|
|
|
|
| Re: latch: session allocation and CPU [message #291082 is a reply to message #290937] |
Wed, 02 January 2008 23:37 |
trantuananh24hg
Messages: 744
Registered: January 2007
Location: Ha Noi, Viet Nam
|
Senior Member |
|
|
hmm! I have not ever seen the server which had so big physical memory like yours. The SGA_TARGET is to be set by 20GB, it means that your machine's physcial memory about 40GB or 50GB.
You can post your report of SGA by connecting as sysdba.
Collect by Statspack, ASH or AWR to generate exactly about latch, and post its portion here.
|
|
|
|
] 
Blog
Search
Help
Members
Register
Login
Home
