First let me thank everyone who responded so quickly!!
A couple clarifications: IDS 9.40.FC5 is Informix Dynamic Server (we aren't
using Oracle)
The systems are not clustered...just "attached" by a gigabit network.
We still aren't sure *exactly* what was going on, but our DBA backed out
some changes to the Informix configuration file (that didn't have any
negative effects on the ES45), and it seemed to make all the difference in
the world.
The specific things that were changed were:
02:14:45 Onconfig parameter PHYSDBS modified from dbs_do to dbs_pl.
02:14:45 Onconfig parameter PHYSFILE modified from 64000 to 16000.
02:14:45 Onconfig parameter BUFFERS modified from 1200000 to 1000000.
02:14:45 Onconfig parameter PHYSBUFF modified from 512 to 64.
02:14:45 Onconfig parameter LOGBUFF modified from 512 to 64.
These changes were put in before we even purchased the GS1280, and on a whim
decided to take them out. Now the 1280 is considerably faster than the
ES45.
We are still looking into why these settings had such a horrible effect on
the GS1280, and not the ES45.
Most of the responses seem to think Memory is involved, especially with
respect to the very different architectures between the ES45 and the GS1280.
But anyway, I just wanted to post a summary and end this, because this
problem suddenly seems more hardware related than OS related.
Thanks again for the responses: Benjamin, Aaron, Charles, Thom, Dr. Thom,
Kevin, and Phil. :)
-----Original Message-----
From: tru64-unix-managers-owner_at_ornl.gov
[mailto:tru64-unix-managers-owner_at_ornl.gov]On Behalf Of Jonathan
Williams
Sent: Thursday, June 23, 2005 11:33 AM
To: tru64-unix-managers_at_ornl.gov
Subject: ES45 outperforming GS1280 - Tru64 5.1B
I was told to send out the following:
We are running IDS 9.40.FC5 and have the following 3 Alpha Tru64 systems
with Unix 5.1B, Patch Kit 4.
Royale: GS1280
8 - 1.3 EV7 CPU's
8 - gigs of memory
Broadhurst: ES80
8 - 1.0 EV7 CPU's
8 - gigs of memory
Majestic: ES45
4 - 1.25 EV6 CPU's
8 - gigs of memory
We have a multi-tier in house application so the database never has more
than 95 connections.
We use processor affinity (1-7 for 8 cpu and 1-3 for 4 cpu).
The systems are connected with a gigabit switch (Cisco 4006).
When using different configurations, the resulting throughput speed makes no
sense to us.
1) Royale application server - Majestic db server - fastest combo
2) Broadhurst application server - Majestic db server - 2nd fastest
3) Royale application server AND db server - much slower than the previous 2
configurations
4) Broadhurst application server - Royale db server - slower than
configuration #3
At all times the systems are at least 50% idle and usually 75% idle when
volume is heavy and timing tests were done.
All timings done with the exact same physical disks (2 spindles in a SAN we
configure to the db server).
We would have thought that configuration 3 would be by far the fastest.
We notice 'delays' when it appears that IDS completely stops for 4 or 5
seconds with configuration 4, but only 1 second
with configuration 3. 'Delay' is defined as all our applications stop (even
dirty read only) like they are all waiting for the db. There are more
delays than checkpoints.
It appears to us that Royale runs slower and communicates slower than
Majestic when it is a db server but is much faster than Majestic as an
application server.
Does anyone have any ideas why our fastest, newest computer Royale would run
slower than the 4 cpu with slower CPU's?
Thanks in Advance
Jonathan Williams
UNIX Systems Admin.
The Shubert Organization, Inc.
Received on Thu Jun 23 2005 - 19:32:26 NZST