Re: RAC and table partitioning
- From: Adam Musch <ahmusch@xxxxxxxxx>
- To: vasudevanr@xxxxxxxxx
- Date: Mon, 9 Jul 2012 23:51:57 -0500
In short, yes.
Inter-instance write-traffic is to be avoided where possible. In a
world of faster interconnects, inter-instance write traffic is less of
a problem than it once was.
Estimate the overhead cost of switching write-interest in a block at
one millisecond.
Assume you have the classic right-legged problem of doing inserts
where an index is populated by a sequence.
How frequently is the index block switched between two nodes if
processing 10 records per second on each node? 100? 100,000?
At what point does the overhead cost of inter-node interest take
longer than the Oracle service costs?
At what point does the overhead cost of inter-node interest in one
second take longer than one second?
On Mon, Jul 9, 2012 at 9:00 PM, Vasu <vasudevanr@xxxxxxxxx> wrote:
> Common sense says "data usage on RAC nodes- aligned to table partitions
> " should do better.
> Say, a table list partitioned on state column, thus dividing Txn activity
...
> My question is : Has anyone seen significant/dramatic performance gains by
> aligning application usage to table partitioning ?
> If so, what was the gain % (though it would largely depend on the workload
> , h/w etc )
>
--
Adam Musch
ahmusch@xxxxxxxxx
--
//www.freelists.org/webpage/oracle-l
Other related posts:
