This content has been marked as final. Show 13 replies
Basic question: Did you gather statistics for the database after import on the new server?
Why did you increase the SGA from 2.6 to 9GB....?
Besides Sathish question
This can be potential issue...One of my client burnt their fingers after increasing the memory from 1.5 to 4.5 GB... THe system was too slow after that...
Yes,i did .I found a new server use less utilization resource than old server such as cpu usage 10%,disk 10 % .on a new server i added parameter db_keep_pool_size=3gb and the other oracle parameter i was not change.
Can we have the Server OS Version? both OLD and NEW
You might need to tune at OS level
Edited by: Maran Viswarayar on Mar 18, 2009 4:10 PM
os is solaris10 on both
I don't think that there would be a reasonable guess without looking into what's going on? Can you get a Statspack/AWR report of the slow system and post here?
My Main suspect is the Increase in memory
Are you running 32 bit or 64 bit database?
How many GHz does your CPUs in your X2100 and t5120 exactly have?
From what I have seen on the specification of this 2 server models it seem that a CPU core on the t5120 is slower then one on the x2100.
If you have only a few concurrent sessions requiring CPU resources it could be that your performance problem is related to the slower CPUs.
Why did you chose the t5120 to replace the X2100 ? I don’t think that the T2 processor is an optimal CPU for DB servers.
Old server : cpu=2,Mem 2gb,Disk storage 200 gbmirror,sunfire x2100,db oracle 10.2.0.1,sga 2.6 gbAre the numbers right? 2.6GB SGA with 2GB RAM seems like a recipe for a swap fiesta..
Anyways, just check your O/S setup on the new box. Check how filesystems are mounted, w/ or w/o forcedirectio etc. Trace. Rinse. Repeat.
And don't DISM ;)
To add to Maurice's comment, see Glenn Fawcett's blog
He has posted a number of test results on the new Sun CMT servers.
Essentially, these server deliver throughput gains only if you have a large number of process/batches/jobs/sessions to run concurrently -- they give good parallelism. If you have a small number of threads/users, then performance will be POORER, certainly.
In the metalink note 781763.1 you will find the explaination why your new server is slower.
Here a short extract of the note:
# A CMT pipeline runs at say 1.2GHz and has 4 threads sharing it
# Therefore each thread only gets 1/4 the cycles and runs 300MHz
# This makes it less performant than an old US II chip
So to me it seems that as long as your servers are not CPU bound you will get a much slower performance with servers based on T2 CPUs.
Maurice has identified the cause of the problem already.
The X2100 probably has AMD CPUs running at over 2 GHz each. Each individual core is single threaded and executes at this speed i.e. over 2 GHz.
The T5120 has the T2 processor with 8 cores each running at 1.2 GHz or 1.5 GHz (you don't tell us your CPU speed). The 8 way multi-threading per CPU reduces the effective speed of each core to under 200 MHz from the perspective of a running process.
So clearly the T5120 at 1.5 GHz has a slower CPU than the X2100, and each job will run slower. It might scale better to run more jobs concurrently than the X2100, but each individual job will run slower. There is no other way it can be. Laws of physics and all that stuff. Each of the 8 cores in the T2 processor is far simpler than the cores in the AMD CPUs in the X2100. So apart from the 1.5 CPU speed difference, the T2 processor is much slower than the AMD processor internally when doing certain operations. This means that the net performance difference is about 2x i.e. the X2100 will run individual jobs twice as fast as the T5120 does, because its CPU cores are twice as fast i.e. execute instructions at twice the rate.
Furthermore, as Maurice points out, each core supports 8 threads at a time. But each core can only ever execute instructions from one thread at a time - it just switches between the threads very quickly. Assuming a 1.5 GHz core and 8 active threads getting an equal share of CPU capacity, then each thread is executing at the same rate as if it were on a dedicated 187.5 MHz single core CPU (1500 / 8).
As far as I am aware the T2 processor and its predecessor T1 processor are the only processor that run jobs slower as you use up more of the threads in each core. As more threads are run, so each thread gets less of the share of the CPU core it runs on, as it is shared out between the other threads.
In simple terms the T2 processor is not suitable for running database server software such as Oracle. In spite of everything Sun will tell you and their published test results, the reality is that the T2 processor is very slow to begin with (1.5 GHz), and gets slower in real terms due to the way it switches between multiple threads (processes) so that each Oracle process will get the equivalent of something like a 200 MHz CPU. We left 200 MHz CPUs behind over 10 years ago.
[Database Performance Blog|http://databaseperformance.blogspot.com/]