I used profile provider to look what ustack() was most common for
ora_ckpt proccess (tid==1) when a tablespace is being put into a
hotbackup mode and Edward was right - it's not related to semop() or any
interposing libraries in S9 zone - it were almost Oracle ustack()s only.
So I cloned the database and started exactly the same binaries and data
in a global zone.
Now putting a tablespace in a backup mode takes well below a second
while it takes about 4s for the instance running in a S9 zone... but
there is one difference - much smaller buffer cache for oracle SGA
running in the global zone (not enough memory to start both of them with
the same memory allocations). Forutnatelly I could stop the other
instance and start the one in the global zone with the same memory
settings so now all setting are the same and now it takes about 4s for
the one in a global zone as well.
So it seems there is no issue with Oracle 8.1.7 interacting with S9 zone
- rather it slows down considerably when large SGA/buffer cache is
configured and a tablespace is put into a backup mode (and possibly
other operations as I got dba's complaining of slow performance) - I
will experiment more and will probably tune the buffer down and rely on
ZFS caching to make use of a memory.
Thank you guys for looking into it.
zones-discuss mailing list