Using VAXStation 2000 as X-terminal

[Kurt J. Lidl]

>In article <PERF.90Nov15165838 at osiris.efd.lth.se>, perf at efd.lth.se (Per Foreby) writes:
>> We have 20 VAXstn 2000 which are never used because our Sun's are much
>> faster. Our DEC salesman came up with an idea: Buy som new DECstation
>> 2100's, put 24 Mb in each of them, and use the old 2000's as
>> X-terminals.
>> 
>> In this way, by purchasing 20 DECstation 2100, we could get 40 fast
>> workstations.
>> 
>> Comments? Has anyone tried this?
>
>The VAXstation 2000's will work fine as X-terminals *provided* you put
>enough memory in them.  I'd suggest an absolute minimum of 12MB, and
>preferably 16.  We've tried with 6MB (running DECwindows with only
>the server stuff running on the 2000), and it works, but painfully.
>Logging in is especially fun -- plenty of time for a cup of coffee,
>including brewing time.
>
>With more memory, they work fine.  But you have to consider the cost
>of adding memory vs. invoking DEC's trade-in deals on DECstations.

This is not our experience at all.  A 6 megabyte, diskless, monochrome
VaxStation 2000 makes a just dandy X-terminal.  In fact, I am typing this
article on one of these X-Terminals right now :-)

We run 19 of them off of one DecStation 3100, which functions as the
NFS server for them and as their MOP boot loader.  We have 3 SparcStation
1+'s that are the compute servers for the cluster of machines.  We get
awefully good performance on these machines.  Then again, we don't
run much of anything on the Vaxen, except for the X11R4 server (the
MIT one -- we did not like some of the "features" of the DecWindows
server).  The Dec3100 primarily does NFS traffic to the X-Terminals to
feed them their font files, which are all in one partition on the
machine.

I suppose the try gross thing about the setup (in a certain manner of
thinking) is that the machines all share the same root partition and
same 6 meg swap file...  The MOP packet that is sent on bootup says the
swap space is 6 megs -- the same as the amount of ram in the machines.
Thus, they never swap or page.  Thus, they can all share the same
swap space.   The root partition they write to, but primarily to
the X11 socket and the device drivers.  There is no file creation in
/tmp, so all the machines can stay happy.

It takes about 15 seconds from the time I hit return on the keyboard
(for the XDM display) to the point where my X setup is totally
up and running.  I can't get on that fast on the *console* of the
Sparcs that serve as the compute servers for the lab.

I'd be happy to show you how to setup your machines like this -- I would
get one 2100 or 3100 -- all it has to do is boot the Vaxen and then
you can use the compute engine of your choosing for running the client
programs.

We have compiled the X11R4 MIT server with GCC, with optimization, and
it goes pretty quick.  I think that we out-perform the DecWindows server
in one or two areas, are reasonably close in a lot of areas, and only are
significanyl slower than the DEC server in a few areas (this is via timings
from the "xgc" program.  I will be happy to send the results of our tests
(simple though they are) to people who want them.

The entire setup code we wrote could use a little cleanup, but it works.
We have been toying with the idea of spending some real development time
to bring the concept of a VaxStation 2000 as X-Terminal to a new height.
One that involves no Ultrix, just a packet filter on the Sparcs and
some kernel hacking.  You'll be hearing from us if this comes to
fruition...
							-Kurt
--
/* Kurt J. Lidl (lidl at eng.umd.edu) | Unix is the answer, but only if you */
/* UUCP: uunet!eng.umd.edu!lidl    | phrase the question very carefully. */