re System Stats in RAC across multiple nodes -- was RE: Varying plans on different nodes

  • From: Hemant K Chitale <hkchital@xxxxxxxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Sat, 09 Jun 2007 12:19:34 +0800

I had been following the thread "Varying plans on different nodes"
(but may not have read all the emails in that thread)

The question I quote below seems to have come up a few times :

Could you please explain how to get two sets of system stats in one
database at a given time? AFAIK this is not possible because system
stats are stored in a single table (sys.aux_stats$) and in it there is
no mention of instance ID.

I would think that an RAC environment  _CAN_ comprise of two
dissimilar servers  in terms of Hardware Resources -- CPU and Memory.
In fact one of the premises of RAC is that you could run one service -- say "HR" --
on one node and another service -- say "Financials" -- on the other node,
both sharing the same database. The idea  being that the two services
don't necessarily need the same Hardware Resources and do not have
much Block Pinging but must share the same database.  If so, you
could size a smaller server for the HR service and a larger one for the Financials service.
Wouldn't CPU (and even, quite possible, I/O response time as seen by the
DB Instance -- the lighter instance makes fewer I/O calls and with OS buffering
the I/O calls are possibly faster on this instance)  be  *different*  on the
two nodes ?

The bean counters would want you to justify why you want 8 CPUs for the
HR users who are a much lighter load .   You might say "IF the Financials
server goes down, we want the HR server to be able to take the load of the
Financials Users".  Management might say "IF the Financials server goes down,
we reduce or stop the HR users and control the Financials users to only
critical users / reports for the few hours it takes to bring up that server,
but let's not waste 4 CPUs lying idle on the HR server  for 3   years just
because the Financials Server might go down for 2 to 4 hours once in those
3 years.  RAC does give us *continuous* uptime , we do not HAVE to have
*the same performance level* always  -- we can afford reduced performance
for 2 to 4 hours in 3 years as long as we still have 100% uptime. Let's just
buy 4 CPUs for the HR server and 8 CPUs for the Financials Server."

Wouldn't it be logical  to have differently sized servers running different
services even if in the same RAC Cluster ?  Then wouldn't System Statistics
be different between the instances ?

Hemant K Chitale

"First they ignore you, then they laugh at you, then they fight you, then you win" !" Mohandas Gandhi Quotes :


Other related posts:

  • » re System Stats in RAC across multiple nodes -- was RE: Varying plans on different nodes