Hello all,
I am hoping someone here can provide some insight into weird
Oracle behavior on Solaris 10 in non-global zones. To start with, here
is my configuration:
E2900
-> 12 CPU (shows up as 24 due to the dual cores)
-> 64 GB memory
-> Disk is a mix of an SE-6120 partner pair and 6920 SAN
-> Solaris 10 6/06 s10s_u2wos_09a
-> Oracle 10g1
-> vmstat shows about 36452448 memory free and 88386272 swap free
-> 8 non-global zones
-> 5 non-default resource pools
-> the zones are a mix of oracle DB and oracle App (each zone has
either DB or App running in it)
-> at least one instance or oracle in each non-global zone (most have
more than one)
When our DBA tries to start an Oracle DB (call it instance 1),
it tries to allocate a 6GB shared memory segment... this fails. It
then manages to allocate a shared memory segment of about 1.8 GB and
then allocate more shared memory segments until it has the 6 GB it
wants. There is a second Oracle DB instance in that same non-global
zone (call it instance 2). If the DBA starts instance 2 first, then
instance 1, they both start, but complain about shared memory
allocations. If he starts instance 1 then instance 2, instance 2 will
fail to start due to shared memory allocation failures.
We do not have a project setup for oracle (we didn't see the
need, as everything on the box, in every non-global zone, is oracle;
and with 64 GB of memory, the default max shared memory size should be
16 GB... more than we need).
I am an Admin, not a DBA, but we suspect this may be a systems
problem and not a database problem (we are trying to figure out how to
easily tell if the shared memory request is failing or if Oracle isn't
making the request correctly...I need to poke at dtrace for this).
Is there anyone here who has an idea of what we are running into ?
Thanks, in advance.
--
Paul Kraus
