Hi Andrew,
1- The old box+ASE is not available but to simplify my question, I purposely did not mention that the new box+ASE is a replica (REP15.7.1) receiving transactions from a primary db which is still on a non-migrated M4000 which is the same as the old box. The comparison is not perfect but the query runs around 7 mins on the primary with more activity,
2. I confirmed that ASE config file are exactly the same. The system team unmounted the disks where ASE software + devices were hosted then re-mounted on the new server. so the config files are the same.
3. I collected the MDAs + plans comparing the slow query and the fast query using #temp which I sent to the SAP support. Both query plans were showing a large number of IOs on a worktable:
Table: Worktable1 scan count 1, logical reads: (regular=132748 apf=0
total=132748), physical reads: (regular=16931 apf=0 total=16931), apf IOs used=0.
-> I was told that LIO for Worktable1 will be stored in procedure cache and that #temp will be stored in my TempDBcache (3GB out of 8GB of tempdb is bound to a dedicated cache).
I should make the same collection for the query with "set statement_cache off".
4- We kept Solaris 10 1/13 although I would have preferred to move to Solaris 11.
Today system team added memory to the hypervisor and added some values to /etc/system.
OS support said that "The profiling lockstat did not show page_freelist_coalesce " , saw a "__aio_mutex libaio.so.1`_aio_lock+0x2c" lock and recommends that "limit the number of threads doing asynchronous I/O in the application or to have the application wake up only one thread at a time"....not sure if the OS monitoring is relevant in my case.
I will keep you posted. Thanks for your time!
Vincent