Oh, I know how many different interpretations of performance there could
be for SunRay servers.. #1 is a pretty simple one, and easy enough to
figure out the max # of reasonably performing clients.
2 is way harder.. what WindowManager do you allow your users to have? I
would push them off of Gnome/KDE/JDS, and more towards blackbox, or
icewm, or something really lightweight. Also, what are the users doing
on the box.. IC design with Cadence is a way different load then a
programmer, then a secretary..
Ideally though, you could make up a "typical" user for the #2 case, and
have X numbers of users that can run with no noticeable performance
degradation.. Differences in memory for the typical workload would be
nice to have graphs and benchmarks for too..
In the end however, no benchmark(et)ing will be right for your case, but
it would be nice to have a way of comparing the different types of
servers... i.e. T1000s make great servers for case1, but go x4100 for #2
in general cases?
Ivar Janmaat wrote:
I don't know if benchmarks are the proper way of showing the systems
capability.
Maybe a graph with number of users versus performance would also be
interesting.
First of all there are two ways to use a Sunray system:
1. Just a session manager. Rdesktop, Citrix or NX are used to run the
desktops on other systems.
2. Run a JDS (full desktop) on the sunray server.
If you look at the memory, cpu and disk usage there is a big difference
between 1 and 2.
So maybe two benchmarks are needed?
Secondly the scalability of a Sunray server is very important.
If you add more users the user experience (performance) should not degrade.
Lets run a benchmark for every user and increase the number of users.
Make a graph of the number of users (x) versus the benchmark results (y).
The ultimate sunray server would probably be the one which has a
horizontal line in the graph.
The benchmark stays the same when you increase the number of users.
Ivar
Paul Greidanus wrote:
That's an interesting thought.. is there any resonable way to quantify
SunRay performance in a way that you could compare Opteron vs
UltraSparcIII vs Niagara based systems? Would be cool to have the
SimDacenter tool have a metric for this as well...
Ivar Janmaat wrote:
It would be nice if Sun would provide some more information about
what to expect from their servers in a sunray environment when they
launch the systems. Maybe a Sunray benchmark is needed for this?
Or might sun focus on the niagara II to be the real sunray server?
_______________________________________________
SunRay-Users mailing list
[email protected]
http://www.filibeto.org/mailman/listinfo/sunray-users
_______________________________________________
SunRay-Users mailing list
[email protected]
http://www.filibeto.org/mailman/listinfo/sunray-users
--
Paul Greidanus
CAD Administrator / Systems Administrator
Center of Excellence in Integrated Nanotools University of Alberta
[EMAIL PROTECTED] 780-492-7368
http://www.cein.ualberta.ca
_______________________________________________
SunRay-Users mailing list
[email protected]
http://www.filibeto.org/mailman/listinfo/sunray-users
