tbray@watsol.waterloo.edu (Tim Bray) (01/20/91)
zs01+@andrew.cmu.edu (Zalman Stern) writes:
When I was doing development on a 530 (25 Mhz RIOS) I didn't notice these
problems. (My MIPS Magnum feels a little better, but at least part of that
is the losing X11 performance on the RIOS.)
Very very interesting. Is it well-established that X on the RS6K is slow?
How come?
Tim Bray, Open Text Systems, Waterloo, Ontario
jonathan@speedy.cs.pitt.edu (Jonathan Eunice) (01/31/91)
Zalman Stern (zs01+@andrew.cmu.edu) writes... | When I was doing development on a 530 (25 Mhz RIOS) I didn't notice these | problems. (My MIPS Magnum feels a little better, but at least part of that | is the losing X11 performance on the RIOS.) Leading Tim Bray (tbray@watsol.waterloo.edu) to ask... | Very very interesting. Is it well-established that X on the RS6K is slow? | How come? The currently-shipping AIXwindows is indeed under-optimized. One problem is that IBM is shipping X11R3 -- now a moderately old release -- while other workstation vendors ship X11R4. R4's better initial tuning, overhead-saving features like shared widgets, and its greater maturity are real wins. Also, I think IBM has had less experience than HP, DEC, Sun, et al in tuning X. As it's released from MIT, X is not the most highly optimized code you've ever seen (no slight intended to the fine folks who bring us X). Other vendors had a longer history of tuning their X packages, and I think that shows too. The extent of the problem depends seems highly user- and application- dependent. I use AIXwindows with 4-7 windows active, and new ones popping up all the time, and I haven't had much problem running X on a low-end RS/6000 M320 workstation with 16 MB RAM. I have not seen any solid benchmarks that indicate horrid IBM X performance, but I have heard reports of Island Graphics and (especially) Autodesk software being noticeably slow -- in the case of Autodesk, absolutely horrid. One possible explanation I have heard is that IBM tuned its software for the case of one active window, and ignored the much more common (at least among workstation users) multi-window scenario. This is clearly a brain-dead approach, making the rumor rather more dubious than usual. Take it as you will. I have also heard that IBM Xstations are slow, apparently for much the same reason AIXwindows is -- X11R3 vs R4, and tuning. Then again, it appears that many of the system vendors who've really gone after X terminals -- HP, IBM, DEC -- have all delivered goods that are slower, more expensive, heavier, etc than independents like NCD and Visual. This is beginning to change in the system vendors' second rounds of product. NB, though nothing has been announced, it is rumored that IBM will soon be shipping a much-enhanced X11R4 implementation. I would expect this to mean for both AIXwindows and the Xstation. (Interestingly, this may show the wisdom of runtime-loaded rather than ROM-based X terminals. Users of IBM, HP, etc terminals can just replace the X terminal manager software on one or a few file servers -- no running around changing tens or hundreds of ROM chips on every terminal.)
jsw@xhead.esd.sgi.com (Jeff Weinstein) (01/31/91)
In article <9837@pitt.UUCP>, jonathan@speedy.cs.pitt.edu (Jonathan Eunice) writes: > Also, I think IBM has had less experience than > HP, DEC, Sun, et al in tuning X. As it's released from MIT, X is not > the most highly optimized code you've ever seen (no slight intended to > the fine folks who bring us X). Other vendors had a longer history of > tuning their X packages, and I think that shows too. IBM used to have a group of X server hackers that had been working on tuning X11 since before X11R1 alpha about 4 years ago. The first port of the MIT X11 sample server to a non-DEC machine was to the IBM RT running AOS 4.3. This group still exists...we all work on the X group at SGI now. --Jeff Jeff Weinstein - X Protocol Police Silicon Graphics, Inc., Entry Systems Division, Window Systems jsw@xhead.esd.sgi.com Any opinions expressed above are mine, not sgi's.
mberkley@active.uvic.ca (Mike Berkley) (02/01/91)
> On 30 Jan 91 17:04:41 GMT, jonathan@speedy.cs.pitt.edu (Jonathan Eunice) said:
JE> (Interestingly, this may show the wisdom of runtime-loaded rather
JE> than ROM-based X terminals. Users of IBM, HP, etc terminals can
JE> just replace the X terminal manager software on one or a few file
JE> servers -- no running around changing tens or hundreds of ROM
JE> chips on every terminal.)
NCD offers the best of these combinations. A runtime-loaded server
that can be updated frequently and centrally, and a ROM-based server
that can be used if TFTP servers are temporarily unavailable.
Mike Berkley
University of Victoria
mberkley@sirius@UVic.CA
wlm@ibm.com (Bill Moran) (02/01/91)
One thing Jeff is too modest to mention is that the BSD RT used to make almost everyone else's color X look pretty silly. Eventually, other people caught up in terms of code, and their hardware got much faster. Bill Moran -- arpa: moran-william@cs.yale.edu or wlm@ibm.com uucp: uunet!bywater!acheron!khand!wlm or decvax!yale!moran-william ------------------------------------------------------------------------------- "I like liquor and women and chess and a few other things." Philip Marlowe