Re: Shared pool and library cache latch contention

Hi,

> Hi Folks,
> We recently upgraded an 11i (11.5.6) environment from 8.1.7.2 (32-bit)
> to 8.1.7.4 (64-bit) on Solaris 2.8 running on Sun E15K. The server has
> 48 CPUs and 96GB RAM. The shared pool size was 500MB and I carried
> over the same size after the 64-bit upgrade. Soon after we upgraded,
> we started seeing "latch free" waits among the top two wait events,
> sometimes even staying at the very top. I looked at v$latch and
> observed that "shared pool" and "library cache" latches had highest
> number of sleeps. At one point when the system slowed down
> considerably, I looked at the x$ksmsp and noticed that one shared pool
> bucket had over 2400 free chunks. At one point we also encountered
> ORA-4031 error and I figured that the shared pool was 'under sized'.

Aso check activities in shared pool reserved area (query x$ksmspr instead x$ksmsp).
Personally, I think that on 64bit the 11i normal recommendation for _shared_pool_reserved_min_alloc = 4100 is too small, because the same code can take more memory compared to 32bit when loaded into shared pool. But this shouldn't be the main problem..

Also, when you increased shared pool to 800M, did you inrease shared_pool_reserved_sze to 10% of it as well?

Tanel. Received on Thu Mar 13 2003 - 09:16:15 CST