First, thanks much to all the people that took the time to respond to
my initial query (copied below). I have included all responses here
even though it's made this email quite long, since a lot of detail and
some _tricks_ and comments are included in them that could prove
generally useful.
I believe X-clienting is an area where more complete information would
be beneficial to the NMR community (I still wish I had better number to
plan from), so please continue to update this group whenever you have
something new to add.
Overall, I can say that I wish I had the resources of DuPont
Central Research... since I don't, let me summarize some important
common points made by the respondents:
1. The current belief (no data to verify) is that memory on the
server is the most limiting factor as to the number of concurrent
logins that can be handled by the server; it would seem problems
are more likely at 64 Mb for getting multiple (2) users going
for 2D processing than at 96 or 128 Mb.
2. Software, both the primary application on the server and the X-client
software impact the performance substantially.
3. For some reasons (not clear why), Mac clients appear to not work with
XWINNMR at all.
4. XWINNMR has a couple other problems (single logins and system wide
serial processing; see below) that seem limiting.
5. Other application software seem to work fine with nearly any client.
I have settled on the following for our planning at UW-Madison, Chemistry:
Purchase SGI O2 or Sun Ultra 1 (minimum processors) with 128 Mb
memory and cache; we'll get as many as we can afford (likely only 2
or 3 in this phase). Plan on up to 5 total users, no more than 2
running 2D processing. We will have to experiment to see how X-clients
work: the big push for all this is to get processing, signups, and
acquisition checking going throughout the Department. With that said,
nearly every possible X-client exists (fun, fun :-).
It will be a while before we get everything in, but I will let you
know when things get going here.
Charlie
*******Original email**********************************************************
*******************************************************************************
Date: Fri, 28 Feb 1997 17:51:46 -0600
From: "Charles G. Fry" <fry@chem.wisc.edu>
Subject: Servers with X-windows
To: ammrl@bloch.cchem.berkeley.edu
X-Sender: fry@bigbird.chem.wisc.edu
Dear Colleagues,
I once again need your help in doing some planning. It turns
out that our lab has not yet moved to the world of X-windows
with regard to using UNIX boxes as servers, and having PC's or
Mac's log in as X-clients (e.g., to VNMR or XWINNMR).
We have done this a bit, but only so far as to see that when
working with Felix on an SGI Indy with 64Mb of memory, the
slowdown associated with even one X-client log in (we used
another Indy) was too much to be tolerated.
I would like to gather information about what configurations
of Sun or SGI servers the AMMRL group has, and how many clients
you have seen that that server can handle adequately.
If you can provide information, please include:
server type (including Sun/SGI, cpu, memory, network connection)
base application (VNMR, XWINNMR, Felix, modeling package...)
type of work being performed (e.g., # clients doing 2D, # doing 1D)
type of PC, Mac, or Unix workstation used for clienting
X-windows software used on PC or Mac
I will post a summary of responses.
Thanks in advance for your help!
*******************************************************************************
*******************************************************************************
I received a forward of your note from Betsy Mccord
(mccord@esvax.dnet.dupont.com). She forwarded it to me since I am the
dedicated computer support person for NMR at DuPont. In your note you were
asking about configurations of servers for running NMR applications back to PCs
and Macs running X-windows.
We currently use a PowerChallenge XL (12 processors, 2.5 GB memory and over 100
GB of disk space) as part of our NMR infrastructure. The machine is mainly
used by our computational chemists for that type of work, but NMR has added
some processors and disk space to the mix to accomodate their needs. We can
currently support about 150-200 interactive users and 8-10 heavy computational
jobs on our machine at any one time. I have found (except for poor coding on
the vendor's part in some applications) that the NMR applications in general
require very little power to run. One must be careful for runaway jobs
(especially in printing) on some of the applications.
As far as clients, we have them all (Macs, PCs, UNIX workstations of all
types). I have found that the Mac clients perform the worst. The only
X-windows application that I am aware of that is still being produced is eXodus
by White Pines. And it, according to our Macsperts is the reason for the
slow-down, not the Macs themselves. In comparing a PowerMac 7500 w/ 32 MB RAM
to a Pentium 133 MHz w/ 32 MB RAM running X-windows to display XWINNMR off of
our server, the Mac was unusable for some parts of the application (manual
phasing for one) and the PC was about like sitting at the Indy running the
application locally.
Hope this is helpful. Please let me know if you have any more questions.
Boyd
-- +-----------------------------------------+--------------------------+ | L. Boyd Reed, II | | | Specialist, Scientific Computing | Duct tape is like the | | DuPont Central Research and Development | Force. It has a light | | DuPont Experimental Station | side, a dark side, and | | Wilmington, Delaware | it holds the universe | +-----------------------------------------| together.... | | Internet: reedlb@boyd.es.dupont.com or | | | reedlb@esvax.dnet.dupont.com | -- Carl Zwanzig | | Telephone: (302)695-3723 | | +------------| My opinions are mine, not those of DuPont |-----------+******************************************************************************* *******************************************************************************
We regularly use X window servers on our PCs and Macs. Briefly:
The majority of the applications we run are off of a Sun Sparc10, which has 128 Megs of RAM and is on our coax network. We use it for 1D data processing (vnmr) and our electronic sign up. (Our sign up is very graphics heavy - it's written in tcl/tk.) Our PCs and Macs run anywhere from a 386 class to the newest, fastest hardware. We use eXceed for PCs and MacX for Macs.
Rarely do people on the console complain about slow downs due to users of the calendar or remote data work up. Mostly users complain of things on the X server being to slow (things take a while to start, things occasionally hang for a minute.) This is mostly due to network overload, I believe. I think the best speed up possible is to get more RAM for the workstation. But make sure memsize (in .login) is set appro- priately for the amount of RAM you have. Also make sure that you do NOT run a window manager (such as olwm) on the client machine. That will REALLY slow things down, and is not necessary.
Eileen Reilly-Horch NMR & UNIX System Administrator American Cyanamid Company P.O. Box 400 Princeton, NJ 08543 reillye@pt.cyanamid.com
******************************************************************************* *******************************************************************************
We are in the middle of transforming our NMR center in the chemistry department here at Columbia. We now have three Bruker D series spectrometers and will soon have four. They all have a SGI INDY 4600PC. The only software we will be running is Xwinnmr. I have set up the following configuration for "off-line processing" of NMR data:
Server: INDY 4400SC with 96MB memory, and thinnet ethernet connection. All spectrometers Indys have a data disk that is exported to the server for NFS. Clients: Tektronics netstation (X-terminal) xp417c software: only Xwinnmr # of clients: we hope to be able to serve about 5-6 clients simultaneously.
The first important point to consider is the client. I attempted to use PowerPC macs as the client using both MacX and Exodus software. I was told Exodus was much better so I concentrated on it. I found that, basically, the mac is worthless when the application is Xwinnmr. I consulted several times with a software engineer at Exodus (who had happened to work for Bruker, previously) and found that, under optimum conditions, the graphics are intolerably slow. Interactive phasing of a 1D spectrum was impossible. (More detail availble upon request)
Instead of pursuing PCs, I turned to X-terminals. The price of Tektronics netstation ($1700) was actually favorable, compared to Macs. They offer steep academic discounts, which they call "charitable contributions" (be sure to ask for that). The speed of Xwinnmr operations, on the netstation, was identical to the speed when using an actual UNIX box as a client, which is to say, it doesn't get any better. Using the netstation is, in most aspects, identical to using a UNIX box. They do have one important difference for Xwinnmr use, however, and that is that have only one colortable. To get around colormap allocation problems, one needs to start Xwinnmr from the x initialization files on the server. This is very easy and may actually be desirable. Overall, I find that the Tektronics netstation to be an extremely cost effective and high performance solution. The only drawback, of course, is that you don't have a mac or pc for other applications.
The second point regards the server. For a server with several Xwinnmr clients, I am told that having a cache and adequate memory are extremely important, more so than the speed or CPU.
Hope this helps.
Sincerely,
John Decatur, Ph. D. NMR Facility Manager Department of Chemistry 3000 Broadway, Mailcode 3179 Columbia University New York, NY 10027 212-854-2155 (voice) 212-932-1289 (fax) decatur@chem.columbia.edu
******************************************************************************* ******************************************************************************* We do this routinely with VNMR and and Spinsight. Only the person using the spectrometer is allowed to do data processing on a spectrometer host for the reason you stated. Others may only do file transfers. We have mostly Sun and some IBM RS6000 workstations which we access from Macs, PC's, and X-stations. All of these combinations work fine except when campus network traffic gets very heavy. Although we have not done careful studies to determine what load is acceptable, I would not recommend more than one or two users on, for example, a Sparc 20 with 64 Mb. RAM, if they are doing heavy 2D processing. We have a departmental Sun 4/660 Server with four processors where we allow free student access. This works fine with several students logged in, but they are often doing just 1D stuff. Give me a call if you'd like more detail about the system.
Charlie
--Charles L. Mayne University of Utah Dept. of Chemistry B103 HEB Salt Lake City, UT 84112 Voice: (801)581-7413 FAX: (801)581-4763 or 8433 E-mail: mayne@chemistry.utah.edu http://www.chem.utah.edu/atlas/facilities/nmr.html
******************************************************************************* *******************************************************************************
We've never seen any slowdown of the type you describe. We run VNMR simultaneously on the Sun's local display and on several remote PC's. In some respects the remote ones seem faster than the local VNMR (I believe this is due to splitting compute and graphics responsibilities among two computers instead of just one).
server type: Sun Sparc-10, 2 x 55 MHz Sparc CPU, 64 MB mem, 10 Mbps Ethernet base application: VNMR type of work: 2-3 people using VNMR (mainly 1-D) plus a few other X clients each (mailtools, xterms, etc.) type of PC: mainly PC's (66 MHz Pentium), sometimes other Suns X-server on PC: XVision
Our server's dual CPU certainly helps, but we had no slowdown even before we got the 2nd CPU. We also use our much slower Sparc-1/Sparc-2 spectrometers to host remote X clients occasionally, with no noticeable impact.
------------------------------------------------------- Steve McKenna stmckenna@amoco.com Amoco Corporation voice: 630-961-7846 room 600-1031, M/S F-9 fax: 630-420-5252 address: 150 W. Warrenville Rd., Naperville, IL 60563 mailing: P. O. Box 3011, Naperville, IL 60566 -------------------------------------------------------
******************************************************************************* ******************************************************************************* As requested, here is a summary of our X windows configuration:
X client platforms: SGI Indy (acquisition workstation), R4600SC CPU, 128 MB RAM, 10BaseT ethernet SGI Iris Indigo (processing workstation), R4000 CPU, 96 MB RAM, 10BaseT ethernet Applications: XWIN-NMR, ParaVision, Felix, NMR1/2 Type of work: Two clients on each machine at any one time. Mostly 1D and image processing. X servers (PC's): Power Macintosh 7500/100, Power Macintosh 8100/80, Macintosh Quadra 800's, Macintosh IICX 486/33 PC (Windows 3.1) X server software: Macintosh and Power Macintosh: eXodus II/PPC from White Pine Software PC's: HCL eXceed
Our experience is that X client connections work very well for NMR1/NMR2 and Felix provided that the Mac or PC's display has sufficient resolution and size to show a reasonable portion of the (virtual) SGI screen. This generally means displaying 1024 X 1280 pixels on a 17" or larger screen. XWIN-NMR and ParaVision do not work as well on a Mac-based X server since they seem to make use of some idiosyncratic SGI X window extensions or require color depth beyond what our Mac video controllers support. These Bruker software packages do, however, work perfectly with SGI-based X window servers, i.e. executing on one SGI and displaying on a different SGI. Note that the same user is not permitted to run more than one XWIN-NMR or ParaVision session on any given CPU, regardless of whether the display is local or remote. Also note that displaying X windows on a laptop computer via a modem connection requires use of a special, optimized transport in place of regular SLIP or PPP, since these would give unacceptably slow performance. Both White Pine and HCL sell matching client and server software to support special X-window optimized transports. . Overall, we find that the speed of our X window connections is quite acceptable and that everything runs fairly smoothly as long as there is sufficient memory on the Unix system to support the requisite number of user processes. In other words, a remote X window user does not use any more memory than a local user running the same program, but there has to be enough memory to support all concurrent users. In some cases, a given program may run faster in a client-server mode than displaying locally since there are now two CPU's available to it: one for "kernel" computations and one for managing the display.
Regards,
Ken Fishbein Facility Manager, NMR Unit NIH/NIA/GRC
Ph. (410) 558-8512 FAX (410) 558-8173 E-mail: fishbein@vax.grc.nia.nih.gov Address: Ken Fishbein NIH/NIA/GRC 4940 Eastern Avenue Baltimore, MD 21224 USA
******************************************************************************* *******************************************************************************
Most of our work is done directly at the workstations due to that very fact...X-Clients tend to be s.l.o.w. However, we (some of us) _do_ run NMR applications via X-Windows clients on PCs, so I'll respond to your request. Server type: Sun Sparc-2, 64Mb RAM, 10-Base-T Ethernet (TCP/IP) Base Application(s) that we use as X-Clients: Triad/Sybyl(Tripos), Spinsight3.x Type of Work: 1D&2D processing (Sybyl/Triad & Spinsight) Modeling (Sybyl) Client Configuration: Pentium-100 PCs, Micro X-Win32 (StarNet, Inc.) Windows NT-4.0 and Windows-95
Rich Shoemaker
-- Richard Shoemaker, Ph.D. Phone--(402) 472-6255 Professor, Instrument Guy, Skydiver FAX---- -6964 URL: http://wwitch.unl.edu/rshoe/ EMail-rskydiver@scientist.com******************************************************************************* ******************************************************************************* We have an SGI indigo 2 extreme or something like that, with 128 megs of RAM, and I regularly run nmrDraw (a graphical interface to the nmrPipe software that Frank Delaglio wrote, and which MSI markets) on the SGI from our Sun spackstation 5 downstairs, telneting in and setting my display to the Sun's display, etc. In this case, I'm not using an X emulator, I'm using the real thing, but I see no slow down at all, really.
We've experimented with various X emulators for windows on our pc, but because it's an old 496 DX2/66 with only 16 megs of RAM, it's an exercise in patience to wait for the screen to redraw.. we were using Xwin, which was free, and it worked technically.. I suspect that on a quick pentium or a quick powerpc, that x emulation would be fine. It would be interesting to see how many clients the SGI could handle - usually I'm the only user, so I don't notice if I'm hogging the CPU time. :-)
Sorry this isn't much information - it's going to depend a lot on how your unix host is tuned, and how the software interacts. I can tell you one thing for sure, though. If I bought a PC for the express purpose of having it be an X workstation for processing or using other X programs, I would without hesitation install Linux on it as an operating system. Linux is free, and even if you get the "Canned" versions like Red Hat, etc. they're still cheaper than MSWindows. Linux has the real X, not an emulator, so there is no overhead.. I am typing on a 486 DX2/66 running linux from home right now, under X, and it's quick as a wink. A pentium running linux would truly sing. Hope this helps, good luck!
- Jessica Dion University of Vermont
******************************************************************************* ******************************************************************************* just one comment about Bruker systems. It seems that XWINNMR queues processing functions such as xfb in a system wide manner. This can appear as a problem even on the fastest workstations when you have multiple users. We run classes with students on Xterms with up to five accessing one cpu at once; although the load average doesnt get too high problems to arise from the serial nature of data processing.
Just my two penneth
Graham
[ Dr. Graham Barlow, NMR Service Manager - gkb1@york.ac.uk ] [ Chemistry Dept. U of York, York, YO15DD. UK | Tel: 01904 432506 ] [ http://www.york.ac.uk/depts/chem/nmr/ | Mobile: 0402 181253 ] [ http://www.york.ac.uk/~gkb1/ | Fax: 01904 432516 ]
---------------------------------------------------------- Charlie Fry Tel: (608)262-3182 Director, MR Facility Fax: (608)262-0381 Chem. Dept., Univ. Wisconsin Madison, WI 53706 USA email: fry@chem.wisc.edu ----------------------------------------------------------