Re: Oracle Parallel Server and 24x7 availability

"Paul Wright" <pwright_at_emailchannel.com> wrote in message news:F6VZ4.3546$1d5.23985_at_newsread2.prod.itd.earthlink.net...
> We are a start-up internet company and need 24x7 availability.

it seems like everyone needs 24x7 availability nowadays. you need to nail down what the real availability requirements are.
99.9% uptime? 99.99%? is "five nines" really possible ?

> I've been investigating into different options, but so far
 Parallel Server
> (OPS) seems to be the best option because at least both
 instances can be
> actively used for load-balancing and fail-over.

In an OPS cluster, if you lose one of the nodes, you will lose the client
connections to its Oracle instance. The client application will need to
participate in re-establishing the connection and transactions. Be sure
to verify what Oracle means by "fail over".

"load balancing" is a "nice to have" but doesn't really address your
availability requirements. if you do go with OPS, and depending on
your transaction mix, you may get better performance by distributing
the transactions such that activity for one application (set of tables)
is directed towards one node (instance) in the cluster.

> All other options seem to be expensive (EMC or advanced
 replication
> needs a duplicate server)

advanced replication would actually require _more_ hardware than OPS,
because you have two copies of the database, rather than one shared
copy. the EMC "storage solutions" are expensive, and only address
one part (storage) of a high-availability environment.

> But, does Parallel server really work as advertised?
> I've really never heard anything good about it. I've mostly
 heard that it
> doesn't work as advertised and the "raw device" requirements
 make
> maintenance a big deal.

doesn't everything ? (work as advertised, that is?) it really depends on
what you read into it. you definitely need to do some investigation as
to what OPS will and will not do for you. "raw partitions" aren't a huge
limitation, as long as you have a way to manage the logical volumes.
(the HP-UX logical volume manager and Veritas volume manager almost
make it easy to manager raw partitions.) for OPS, "raw partitions" are
required because the disks are shared by two (or more) hosts, and a unix
file system can't be shared by two (or more) hosts.

> I've also heard that performance is poor.

OPS does incur some overhead, due to the need to keep the single database "in sync" across all of the hosts (instances). that's extra
communication and between the hosts, and extra disk i/o that would
not necessarily be incurred in a non-OPS environment.

> So...can I get some responses from people using parallel
 server?
> Would you recommend it as a solution fro 24x7?
 

> Any books recommended by anyone that goes through the
 installation
> and configuration of parallel server?

HP has an in depth manual which details the steps required for planning,
configuring and implementing an OPS cluster, on HP-UX of course.

> We are deploying on Sun Solaris

check with Sun, see if they can help you out, or provide you with some
documentation.

> -----------------------------------------------------------
Received on Sat Jun 03 2000 - 00:00:00 CDT