Re: Parallel Server and availability?

mikegrof_at_rocketmail.com wrote :

>It sounds like you are talking about the Parallel Query Option. This is
a
>very different product.

IMHO OPS and PQ goes hand in hand (see my other posts in this thread).

>Can anyone confirm this for 7.x and 8.0? I think I read that 8i will
have
>this capability.

I can for OPS 7.3.4. But just think for a moment about what load balancing entails. Each of the nodes in a cluster needs to be the primary control of a set of raw devices. And of course service the other nodes' request for data from these disks. Then the node needs to run shadow processes, PQ processes and so on. So to load balance in such an architecture is very complex. And one I would rather have done "manually" via things like distributed TCP/IP, the PARALLEL clause of the CREATE TABLE statement, hints in SQLs and so on. As soon as Oracle joins this party with it's own method of load balancing.... Hell, we all know the problems we're having with the fuzzy logic of the CBO.

>What you are describing is a Fail-over setup. Yes, this has nothing to
do the
>OPS, but it's not related to my question.

IMO it has a lot to do with OPS. What I meant in my posting that this is not an OPS feature as such, but more of a requirement. <now I have confused myself! ;-)>

Let me try again:
If an OPS node goes down and the cluster or mesh disks controlled by that node can not be made available by a backup node to the rest of the cluster then OPS is useless. All the other OPS instances will have severe problems as they have lost some of the raw devices of the database.

>I know that some "clusters" work this way, but do they work with OPS?
It
>sounds like it should, but I haven't heard from anyone that is actually
>running it.

This is how I ran a 13 node massive parallel processor mesh. It's different though from a cluster of SMP boxes I think. And there's always arguments between the vendors about SMP clusters vs. MPP architecture.

In our setup node 2 was for example the fail-over of node 1. If node 1 goes down, it takes over the SCSI controller for node 1. Works great. Most times. Except when there's a hardware problem on the SCSI bus that caused node 1 to crash. So when node 2 takes over the bus - guess what happens! :-)

A guy at Oracle Tech Support I know locally told me that they installed and tested OPS successfully with a 2 node NT Enterprise Server cluster (this was sometime last year). I also tried it but failed to even get a cluster configured. NT did not like the hardware on my PCs at home...

IMHO "high availability" for OPS is simply a nice marketing slogan that sales uses. Not saying that this is not a nice feature of OPS (sure saved my butt more than once), but this is -not- the primary feature of OPS, or the reason why a company should even consider getting OPS. By the nature of the architecture of OPS, high availability of the hardware is a pre-requisite to running OPS. And very expensive. I would even say that it may be cheaper to use normal SMP hardware and normal Oracle and then add additional hardware and software if high availability is the core issue.

As I said in my previous posting, IMHO OPS is about parallel processing. Period. Which means multiple database engines running processes in parallel (which of course requires PQ too). But this is just my (technical) opinion from working with OPS for a couple of years. I'm sure that IT managers, sales and others will see it differently.

regards,
Billy Received on Fri Jan 29 1999 - 04:03:47 CST