ADSM-L

Re: ADSM Experiences

1995-09-07 03:29:56
Subject: Re: ADSM Experiences
From: "Keith A. Crabb" <KEITH AT UHUPVM1.UH DOT EDU>
Date: Thu, 7 Sep 1995 02:29:56 CDT
On Tue, 5 Sep 1995 13:53:00 -0500 paul (p.) shields said:
>Actually we were hoping to limit ourselves to 2 server, but our performance
>evals seem to suggest this is impossible. We currently see 500-1000Kb/s on
>our eval machine going to a disk storage pool. Unless we can get performance
>in the 2000-2500Kb/s range, we will need more than 2 servers. What we are
>trying to figure out is where is the bottleneck and how can we eliminate it?
>The only bottleneck we can?t eliminate in our test environemtn is the
>database, and we are trying to guage the impact of going from one disk
>mirrored to another to a system of 6 stripped disks mirrored to another set
>of 6. This is all with a Model 39 RS/6000 machine with a FDDI interface
>driving ADSM.
>Paul Shields
>pshlds AT bnr DOT ca
>Bell Northern Research

Forgive my obtuseness but, it appears that the performance problem is then
the rate at which data is delivered to your disk storage pool?  The 500-
1000Kb/s is for a single machine (session) or for several at once?

Have you considered that the problem may be with the clients ability to
transmit data rather than the server?

I'm use to seeing average data transfer rates of 250-500Kb/s from
a number of clients with client compression enabled, and the compression
does slow the average transfer rate quite a bit.  My DEC Unix Alpha
workstation running a migrated version of the Ultrix client does manage
to get better than 1500Kb/s (even with compression) all by its self.
A quick run while I typed this did a rate of 1613Kb/s while 3 other
machines are running as well.  Now all this is going to a VM server but,
I would find it hard to believe that I'm getting better performance from
an old single CPU 9121 w/4.5MB channels and dasds without caching or
dasdfw on thickwire ethernet than you would be from your RS/6000 and FDDI.
Heck 25% of the CPU is just TCPIP trying to keep up with all the traffic
being generated by ADSM (and I've learned from this exactly why IBM makes
TCPIP offload solutions for VM :).

---
Keith A. Crabb         Keith AT UH DOT EDU
Keith A. Crabb         Keith AT UH DOT EDU
University of Houston  Operating Systems Specialist +1-713-743-1530
<Prev in Thread] Current Thread [Next in Thread>