Load sharing of c++ application on multiple cpu vs multicore machines

Question

Hi all,
We have a multithreaded C++ application(13 threads) developed for
sunsparc, and its been running on sun blade server. Now the load was increased so we had to deploy it on some high end
servers, we have choosen V.490 and T1000.
When application was run on a V.490 machine, it always shows 100%
cpu utilization on one of its cpu. Where as the same application
deployed on T1000 shows less cpu usage(it uses all cores?). The performance was better in T1000.
Does it mean that v490 machine will always use a single cpu to execute
the particular process?. Will it not run the same process across
different CPUs?. Cann't it distribute all the 13 threads on available CPUs?. Where as the load on T1000 is relatively less than v490,
so i assume this T1000 distributes the load on available cores.
or
Does oyr C++ application needs to be compiled with some specific compiler
option or supported sun studio to give a better performance on a multiple cpus and multi core machines?.
Thanks.

Answer 1

saravananm wrote:
Does our C++ application needs to be compiled with some specific compiler
option or supported sun studio to give a better performance on a multiple cpus and multi core machines?.
You can try to build your application with "-fast" option on each machine.
Note, in this case the application is tuned to run on this H/W only, and
may not show good performance results on another type of hardware.
So, you have to build it twice:
1. for T1000
2. for v490
If you want to build it at high optimization level, which is good for all
types of hardware, use "-xO4" compiler option.
Thanks,
Nik