FOLLOWUP:TruCluster Production Server - Memory Channel

From: Wayne Sweatt <sweatt_at_dps.state.nm.us>
Date: Thu, 05 Aug 1999 12:31:16 -0600

This is a message that I sent back to an Oracle consultant
I thought I'd send it here too, for comments:
------------------------------------------------------------------------------

I have another question, as I sit here and wonder...

Maybe you could run this by your consultant and see what his view is on this.

The whole Digital TruCluster arrangement (and apparently most others) require, at the heart of
the cluster arrangement, a shared storage array. These are typically in the form of cabinet(s) which hold
a few various RAID sets. RAID 5, RAID 1 and maybe a JBOD(not RAID) or two. The gotcha here is SHARED.
The basic concept is that if one of the cluster member systems fails the other member can access the failed
member's data.
The problem is that in normal operation, they're still shared and must CONTEND for the bandwidth to/from
the drives on the shared path - Ultra SCSI - Differential, in our case.

The bandwidth contention is what worries me. We are going to have two AlphaServers in our TruCluster both
normally running their own instances. The plan is to have one server run 1 large instance and the other one to
run 4 much smaller instances. Currently they are running on older standalone AlphaServers.
It seems at this point that the increase in overall speed of the new Alpha Server is lost when they are clustered
due mainly to the fact that they have to share application storage and the fact that all this cluster-wise software
has to run on both CPUs as well (TruCluster PSE, LSM, Memory Channel, etc.)

Now, what could the solution be?
* Add a second shared UWSCSI bus, and split up the storage so that one server's normal storage is on one bus,
and the other server's on the other? I know that TruCluster Production Server can support more than one shared SCSI bus.

I also have a OPS related question:
Does Oracle Parallel Server support 5 instances running on TruCluster Production Server?

And a migration to raw device question:
What is the migration method for moving what is now data and ctrl files sitting on an AdVFS filesystem to
raw devices? I have never done anything with raw character devices/disks. I/We had planned to "cold copy"
the current .dbf, .ctl , et al over to the new cluster.

Thanks,

***********************
Wayne Sweatt
Principal Software Analyst
Litton / PRC
505.827.9288
***********************
Received on Thu Aug 05 1999 - 18:34:46 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:39 NZDT