home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.unix.admin:5048 comp.windows.x:16710
- Newsgroups: comp.unix.admin,comp.windows.x
- Path: sparky!uunet!stan!cy
- From: cy@solbourne.com (Cy Foughty)
- Subject: Re: Xterminal-Server ratio wanted
- Message-ID: <Bun4Cu.Ft5@solbourne.com>
- Organization: Solbourne Computer, Inc., Longmont, CO
- References: <1836euINNaqs@early-bird.think.com> <1992Sep12.201312.3019@tybse1.uucp> <PD.92Sep15183212@herts.x.co.uk>
- Date: Tue, 15 Sep 1992 22:12:30 GMT
- Lines: 81
-
- In article <PD.92Sep15183212@herts.x.co.uk> pd@x.co.uk (Paul Davey) writes:
- >
- >One of our distributors was telling me the other week of a pair of
- >sites using multiple processor Data General machines.
- >
- >They are running 100 - 150 X terminals per box with no problems.
- >
- >Software is mixed office automation stuff with X.desktop.
- >
- >The DG hosts have 96 or 128Mb RAM mind you.
- >
- >Response is apparently excellent.
- > Deleted junk.....
-
- I can believe that. The question that needs to be asked is "how many
- ethernet segments are used to support those 100 Xterminals?".
-
- I have performed some tests using a network monitor that also can
- generate traffic. The rules I came up with usually give a good response
- time on Xterminals.
-
- Xterminal Configuration Rules
-
- 1. Separate ethernet segments for Xterminals and ONLY Xterminals!!!!!!
- Ratio: 1 ethernet segment for
- every 100 Xterminals with no more than 40 to 50 Xterminals having
- heavy graphics programs running. If ALL of your Xterminals are running
- a lot graphics then reduce the ratio to 1 enet segment for every 30-50
- Xterminals. You'll need to experiment.
-
- 2. Because the X protocol trys to use commands instead of transfering images
- use RISC based Xterminals such as the NCD 19c (19" color). The draw functions
- will be performed faster. This is one of the most important rules!!!!!
-
- 3. When setting up the Xterminals DO NOT USE NFS for anything!! NFS is a pig!!
- Use the tftp protocol!
-
- 4. Xterminals require SYMMETRIC MULTIPROCESSOR SERVERS for good response time.
- A SparcStation 2 or and HP 7xx are not good Xterminal servers. I/O on the
- server side must be fast. Also configure the SYMMETRIC MULTIPROCESSOR SERVER
- with enough memory and cpu's. This can only be determined by analysis of the
- applications running on the server. Suffice to say uniprocessor boxes and
- ASYMMETRICAL MULTIPROCESSOR SERVERS (can you say "SUN") are the worst for
- Xterminals.
-
- 5. Broadcast packets will kill the network. Xterminals seem to show it more.
- Keep the segments free from broadcast packets. Like, xdm access should be
- direct rather than through broadcast packets. Use routers to clean up the
- segment if needed.
-
- Comments:
-
- I have used workstations for years and also Xterminals. There are cases
- where workstations are the solution. However, Xterminals provide great
- service and you have access to the entire server resources. This is good
- only when the server is a SYMMETRIC MULTIPROCESSOR SERVER that has the
- memory, cpu's, and I/O controllers on separate boards. Because if the
- server goes down, all that is needed is to pull the failing board and
- reboot the server. Performance might decrease while awaiting the new
- board, but the system will be running.
-
- While developing software on a VAX with several hundred other engineers
- the performance went in the gutter. Compiles took hours instead of minutes.
- The solution the MIS department came up with: buy a bigger VAX. OK, fine.
- for about 6 months, then the same thing again! Same solution! The VAX
- upgrades were in the millions of dollars!! The UNI-processor VAXes
- with I/O to multiple processes chokes. There is a bottleneck, it is
- called I/O!
-
- With SYMMETRIC MULTIPROCESSOR SERVERS that have multiple I/O channels
- (not all do) most of the bottlenecks vanish.
-
-
- Xterminals work great if setup correctly with the correct server!!!!!!
-
-
- --
- Cy Foughty Solbourne Computer
- Work: (214) 450-5993 15851 North Dallas Parkway Suite 675
- La Casa: (214) 578-8837 Dallas, TX 75248
- ARETC: Association for the Revival of Electron Tube Computing
-