Re: Oracle Parallel Server

Hey, Brian.

I was running parallel server on a 4100 and a 4600 under VMS 5.5 a couple of years ago. I stopped that about a year ago when I upgraded to an Alpha 2100 for two reasons: 1) the Alpha had five times as much power as both 4000 class machines (which was more than enough at the time), and 2) Oracle did not support parallel server on multiprocessor Alphas. This has been fixed, and I am preparing to reimplement it for a couple of reasons. First, we want to be able to use the processor present in the "development" Alpha machine. Its idle 95 % of the time, which is really a crime. The second reason is to increase system availability (failover, autorecovery, etc).

In the past (the Vaxen), the parallel server implementation ran flawlessly. If you partitioned your users (i.e., segregated the users by the tables they access to minimize contention), performance was great (about 80% of the machine's power was usable, the rest going to overhead). However, there were alot of gotchas that you had to work through.

From what I have read about the latest version, alot of these problems have been removed (or at least decent tools exist for solving them).

Again, my experience is dated, but its was a very positive one. The automatic rollback/recovery is really quite seamless. In one case, we didn't even notice when we lost a node (well, for about 10 minutes anyway <g>).

Are you planning on using all machines at the same time, or will one be a backup machine in hot standby? Another option may be to designate one for batch jobs, while the other is for interactive users (a favorite configuration at my office).

Gimme a call at the office tomorrow if you want to discuss it further. Use private e-mail if you can't remember my phone number.

Matt Reagan
reagan_at_centocor.com (office)
mreagan_at_fast.net (personal)

In article <4hqilh$3j5_at_reeve.research.aa.wl.com>, Brian Laskey <laskey_at_aa.wl.com> wrote:

> Does anyone have any experience or thoughts about the Parallel Server
> Option? We are considering it on either a VMS Cluster or on HPUX
> with their new Distributed Lock Manager.
>
> We are looking at this for an application that "can't" go down. Would
> this help? The disk will be redundant.
  Received on Thu Mar 21 1996 - 00:00:00 CET