20000322: running multiple McIDAS sessions off a server

>From: "Ka Kit Lai" <address@hidden>
>Organization: CCNY
>Keywords: 200003221803.LAA13132 McIDAS-X memory multiple users server LDM 
>decode

Ka,

>      This is Ka Kit from CCNY weather station. I have a question regarding 
>"halo" performance. Prof. Hindman ran a class yesterday and try to run 
>Mcidas on 8 PC mount to halo.

Given the computing power of your machine (halo), this is inadvisable.

>But he can only be able to run 5 PC with the 
>Mcidas running sluggishly, another additional PC had slow down the halo 
>dramatically, almost impossible to run mcidas.

I am assuming that you are running McIDAS on halo itself.  This means that
you will be asking the machine to support 8 concurrent Xwindow sessions
and all of the McIDAS processing that is requested by each user.  It would
take a machine with a lot of processing power to keep up with this.

>Halo currently has 128 MB 
>memory and we plan to upgrade it to 256 MB. Do you have any sugggestion of 
>how much memory we need in order to run Mcidas (or gempak) on 11-15(max) PC?

I would say that the amount of memory needed would be far in excess of
256 MB; perhaps it would be greater than 1 GB.  The real thing to worry
about is the CPU speed.

>      Also, Halo is running on a Sparc 20, do you think that would be the 
>cause of the problem?

Yes, it is definitely not a fast machine.

>Should we buy another faster machine to replace this 
>one,

A SPARC 20 is now a pretty old, slow machine especially if one is trying
to support a number of users concurrently.

>any suggestion to what machine we should replace?

I would say that the model of running the desired program (e.g., McIDAS-X
or GEMPAK) from a single machine is automatically doomed to failure.
Can the PCs you are using be upgraded to run Solaris x86 or Linux so that
McIDAS-X/GEMPAK could run directly on them.  This way, the SPARC 20 would
be used as a data server.

>Sometimes Halo is 
>running sluggishly even for the console(like when I login to halo locally), 

Right.  It is doing all of the data ingestion and decoding.  Again, it is
a pretty slow machine.

>should we avoid running the mcidas/gempak on the halo locally?

If this is possible, then the answer is yes.

>I need to 
>occassionally reboot the machine in order to improve this problem, what will 
>be the problem?

I think that if you ran 'top' while trying to support all of those users,
you would find that the system is totally maxed out.  Adding more memory
will help a little, but this is not the ultimate solution.

>    Thank you very much and I look forward to your reply.

You are welcome.

Tom Yoksas