 |
>
Support > Domino Server Sizing and Capacity Planning
|
8.4.1 |
 |
Domino
Server Sizing and Capacity Planning |
Subject: Re: how
many can NT really host?
Date: Tue, 01 Oct 2002 13:13:11 -0700
From: Ron Herardian <rherardi@gssnet.com>
To: lnotes-l@ozzie.notesnic.net
References: 1
Gene,
I suggest looking at http://www.tpc.org. I have developed a related
measurement called Messaging Events Per Second (MEPS) for systems
like Domino which translates into TPC-Cs, thus if you have a statistical
capacity model of the planned Domino system it is possible to find
with reasonable accuracy the appropriate hardware requirements. We
have done this successfully for a number of clients. We do not use
Lotus' NotesBench or Server.Planner.
Published benchmarks are of limited use. Using these numbers requires
careful reading because in many cases the numbers are artificial in
the sense that they are based on unrealistic tests, e.g., a single
protocol or a single application. Ordinary benchmarks are inferior
to our methodology which is called Simulation Testing. We've modeled
and tested systems scaling to over 1 million users.
Another problem with benchmarks and 'rules of thumb' is that certain
variables do not remain constant with system size and complexity.
In other words, benchmarks and scalability guidelines are inherently
linear but the relationship of system size and complexity to performance
and scalability factors is not. At different orders of magnitude server
sizing equations can be completely different.
The key to using capacity modeling and simulation testing so that
real systems work in the real world is simply understanding, in a
quantitative way, how the planned system will be used. Messaging-only
numbers, for example, would not be useful for applications and applications
vary tremendously in terms of server processor load and network bandwidth
utilization, i.e., one application may be radically different from
another.
The ideal case is to acquire data about the user population and applications,
or from similar systems, and then use this data as the basis of modeling
and testing. Our rule of thumb is that 'if you want to do it right,
you have to do it yourself'.
Ron
--
-------- Original Message --------
Subject: how many can NT really host?
Date: 1 Oct 2002 18:55:34 -0000
From: "Gene Armstrong" <garmstro@csc.com>
Reply-To: lnotes-l@ozzie.notesnic.net
To: Multiple recipients of list LNOTES-L <lnotes-l@ozzie.notesnic.net>
According to various vendor server sizing sites,
many hardware vendors claim to be able to host many tens of thousands
of Notes users per Wintell server. (Some claims to be able to support
4,800 "heavy users" on a
Win2K box with eight(8) 900MHZ CPU's and 1 GB RAM.)
**These aren't just Notes Bench numbers, these are according to their
"server sizer" application. **
We have Windows NT running on Compaq 6 way 550 PIII Xeon's with 2GB
RAM and 240GB hard drive. We have reached an NT limit because of the
amount of data, and as such we are only hosting about 1,500 users
(about 400 concurrent users) per NT box. I know we can upgrade to
Win2K to remove the 180 GB open file issue. But I'm not sure where
the practical limit is.
Here's my question:
Has anybody in the real world hosted more than 2,000 active Notes
users on a single Wintel box? Did you have to disable anything to
do that? Has anyone some practical experience using a SAN attached
to NT to dramatically
increase the number of hosted users?
These are mail users with average mail files of 150MB who are heavy
C&S and Mail users. They also work on the server (no local replicas).
Thanks for you assistance.
Gene Armstrong
garmstro@csc.com


©1995-2005 by Global System Services Corporation (GSS). Portions
of this material are copyright ©1995-1999 by Ron Herardian
|