Re: 10.2.0.2 EE RAC, RHEL4 and LVM2 support?

On 07/24/2006 12:55:53 PM, Kevin Closson wrote:
>  
> Oops, I meant to say they may have added one, but they
> spent years arguing that the central lock manager was
> better than distributed. They just echo Veritas on that
> since theirs is central.

There is, of course, an easy point to be made here: central lock manager
is a single point of failure. If you lose it, your cluster is gone 
fishing. Central lock manager lists and locks all resources in the cluster.
That is the shop Oracle goes to when it needs a file block (or a group of
blocks) locked in such a way that the other nodes can not access it. So,
if the central lock manager is lost, there is no locking of disk resources.
Your cluster has just changed the plane of existence. It is, however, much
easier to program and, unless heavily burdened, more efficient. The trick is
to send an UDP packet asking for the master, much like yelling "Scooby Doo 
where are you". When master the answers, all subsequent requests are sent to 
the master. The whole thing is reduced to a very simple network protocol. Of 
course, if locking starts to get heavy, or if the node hosting the lock manager 
is very busy, response times will start growing.
I don't know if you've ever worked with OPS, but when starting an Oracle8i
instance in an OPS environment, lock manager would have to be reconfigured and
all other instances would freeze as well. Losing the central lock manager
will freeze the RAC database and it will have to be restarted. Distributed
lock managers, much more rare variety then the centralized ones, will survive
the crash of any node. HP has a very decent lock manager, but only on raw 
devices. No CFS of any kind. As far as I remember, Steve Adams used to work
for HP, he might know even more about their cluster manager. I was just a
user.



-- 
Mladen Gogala
http://www.mgogala.com

--
http://www.freelists.org/webpage/oracle-l


Other related posts: