20000407: Diskless Sun workstations and McIDAS

>From: Erick Lorenz <address@hidden>
>Organization: UC Davis
>Keywords: 200004072133.PAA04215 Sun diskless workstation Sunray

------- Forwarded Messages

Date: Fri, 07 Apr 2000 14:34:09 -0700
To: address@hidden
Subject: Diskless workstations and McIDAS

I recently attended a seminar given by our math department in which
they demonstrated and extolled the virtues of equipping a student
computer lab with diskless workstations.  They were able to assemble
these units for about $400 (not including the monitor).  Each station
contains a PROM that boots the unit up to the point where it can then
download Linux from a server.  They each have enough memory to run
Linux and any software without swapping over the network.  They also
used a dedicated LAN for internal communication.  And of course the
server(s) are very powerful with lots of disk space.

This got me to thinking whether such a system could work with McIDAS.
Unlike trying to run X-terminals off of a server, this gives each
station its own CPU and memory resident software.  Is there anything
about the inner workings of McIDAS that would make this a bad idea?

On the good side I can see (1) cheap hardware, (2) ease of maintenance,
and (3) security.  The math people related how crackers would find the
addresses of the workstations and try to break in but couldn't because
there was no local file system to mess with.

What do you think?

Erick Lorenz, UCDavis


------- Message 2

From: "James D. Marco" <address@hidden>
Date: Fri, 07 Apr 2000 23:21:11 -0400
To: Erick Lorenz <address@hidden>
Subject: Re: Diskless workstations and McIDAS

Erick,

I'm not much better with McIDAS than a novice.  But, as a computer
scientist, I'm not sure the trade off in network bandwidth, the added
expense of a good server, and loss of general reliability (the server
becomes a single point of failure for the entire lab) is worth it. I
conceede that administrative chores will be easier.

Yes, I am sure many people have different views, circumstances and
priorities, so I'm sure there will be a lot of discussion here.

My thoughts:
        1) Expense - PC/Linux systems can be had quite inexpensivly.  
        These include enough disk space & memory for McIDAS.  The HP 
        systems currently being 'pushed' on the network will not cost 
        more than $500 (not including monitors.)  For a 15 seat lab, 
        the additional cost of the Server upgrades is apprximatly equal 
        to the difference between diskless workstations and a PC based 
        system. (Much of the cost savings of diskless workstations will 
        be offset by the additional memory needed to prevent network 
        swapping and the higher performance -cpu and disk- that is 
        required.)
        
        2) Maintenence - Depends on what you mean by maintenence.  There
        are two forms: Individual machine and overall Lab.  Hardware 
        maint will be more with more full machines. But, in a stand-alone
        scenario, the reduncy of the stand-alone systems will give you 
        greater overall LAB reliability and availaibility.  The loss of a 
        single machine, or perhaps a few of the machines, may force 
        doubling-up by students, but the lab will not go down 'cuz a 
        drive fails. In a diskless      scenario, a hard-drive crash or 
        motherboard failure will effectivly shut down the lab. Overall
        reliability in the lab will be much better with distributed 
        stand-alone systems. This is an old arguement: Distributed 
        System versus Centralized Server. There is no one answer that 
        is totally best.  (I use a couple of redundent servers, and 
        independent workstations. Even if I loose the LDM server, the
        faculty/students can use old data till I perform a fail-over to 
        the backup server.) Software maintenence is not as bad as you 
        expect. A single image of a lab X-Station is all that's needed. 
        They should all be identical with home directories mounted on 
        the server. You need    regular backups of the server and home 
        directories, in either case. 

        3) Security - Slightly higher concerns with stand-alone systems 
        because they ARE full computers, as you say. But, this is more 
        of a maintenence concern. A hacked server in a diskless setup 
        will, again, shut down the entire lab. (A good admin would shut 
        down the lab if a hack was discovered, in either case. Soo, the
        actual difference in security is the implementation on 15 machines
        rather than one.  But the 'identical image' should minimize this.)   
     
I have found that network bandwidth, even with 100BTX, more of a 
limitation than CPU/disk space.  At 12-15 sessions of McIDAS over the nets
to a single server, the network utilization is high enough to be a noticable
slowdown. A diskless setup will NOT decrease net utilization.  If/when I get
McADDE running properly, it promises to relieve most of this.
        
                        'Jest my 2 cents...errruh...2 bits, now-a-days,
                                                jdm
                  
James D. Marco, address@hidden, address@hidden
Programmer/Analyst, System/Network Administration, 
Computer Support, Et Al. 
Office:                 1020 Bradfield Hall, Cornell University 
Home:                   302 Mary Lane, Varna    (607)273-9132
Computer Lab:   1125 Bradfield          (607)255-5589


------- Message 3

From: address@hidden
Date: Sat, 8 Apr 2000 06:04:35 +0000 (GMT)
To: "James D. Marco" <address@hidden>
cc: Erick Lorenz <address@hidden>, address@hidden
Subject: Re: Diskless workstations and McIDAS

I have installed McIDAS-X and McIDAS-XCD sucessfully and have ran it on
3 in-house built (from pieces) 166 Mhz Pentiums running Solaris x86.
It has always worked fine and these have been really bottom of the
barrel machines. Our main machines are a 400Mhz PII and an Ultra Sparc
but even the 400Mhz PII is by no means the bare minimum in my
experience.  My daily use machine is a 166 Pentium dualbooted with NT
and Solaris.  When I am in Solaris I run McIDAS and get data from the
SPARC in the other room via the ADDE server (or data from our other
machine in New Mexico).  It works great and is imminently usable even
on the old Pentium (IDE disk and 96MB RAM).

The future of Solaris x86 always seems to be in doubt, but Sun has
recently reaffirmed its support for Intel.  I would give it
consideration as well. I have built, installed and run McIDAS on Linux
as well but have encountered bugs, quirks, etc that more experienced
Unix-types can get around and probably enjoy tinkering with..but I find
Solaris x86 to be a no pain deal.  Much easier to get running perfectly
right out of the box.  I have really been happier with our Solaris
Intel system than our Solaris SPARC (although it is really an
Ultrasparc clone, not a genuine Sun).

In short I agree with James, several minimal PC's should work
great.

Robert Mullenax

------- Message 4

From: "James D. Marco" <address@hidden>
Date: Sat, 08 Apr 2000 08:38:29 -0400
To: address@hidden
Subject: Re: Diskless workstations and McIDAS

I probably should have described in more detail the setup I am using,
my apologies.  Robert described good workable system in some detail.
Solaris is in use at Unidata.  Hence McIDAS-X is a bit prejudiced
towards it; if you do not have a lot of experience with some flavour of
Linux, choose Solaris. (I am in the process of evaluating RH 6.2 and it
seems a little more friendly than past versions.  The downfall is the
Gnome/Enlightenment GUI.  It's way too ponderous for older machines.
The old FVWM is MUCH easier on the CPU, faster on the system.) The
difference between the two, otherwise, is cost...Linux can be
downloaded for free.  Soo much for OS.

The main LDM/McIDAS machine is a P233 with 2 9g UWSCSI drives. This is
capable of serving up to 12 simultaneous sessions of McIDAS. 4 of the
sessions are set up as non-interactive batch jobs.  The Fail-Over
server is a PPro200, with 1 9g IDE drive. It mounts the LDM data over
NFS and serves up to 8 McIDAS sessions out to the lab client machines.
The network traffic is mostly here, since the display data is moved
twice: NFS to the server, X-Windows traffic to the remote client.

The client machines are a real mixed bag.  All hand-me-downs from all
over the campus.  They range from old 486 to Pentium 200's.  All run
Windows NT with Exceed X software.

BTW:
My students are not usually Unix people, soo I have a script menu and
batch files set up for them. I can make the files available if anyone
wants them.

                        Hope this helps!
                                        jdm             

James D. Marco, address@hidden, address@hidden
Programmer/Analyst, System/Network Administration, 
Computer Support, Et Al. 
Office:                 1020 Bradfield Hall, Cornell University 
Home:                   302 Mary Lane, Varna    (607)273-9132
Computer Lab:   1125 Bradfield          (607)255-5589


------- End of Forwarded Messages

From: "Mike Schmidt" <address@hidden>
Date: Tue, 11 Apr 2000 11:06:41 -0600
To: address@hidden
Subject: SunRay clients

Tom,

I looked over the brief discussion regarding SunRay clients in the
mcidas-x archives.  There was a heated discussion about the same
topic on comp.unix.solaris and the ease of management seems to
emerge again and again as of the few positives.  There seemed to be
general agreement that if you have/use/like Xterminals currently,
SunRay hardware is the next step to take...

mike