whitney@sdsu.edu (Roger Whitney) (12/13/89)
I have been using Smalltalk-80 v2.3 from ParcPlace on a Macintosh II computer with 5M of memory. When I received my upgrade to v2.5 I noticed a decrease in response of the windows. In particular the first time a window (say the browser) is created one to three seconds pass before the window appears. This is just long enough to be troubling. Response time improves after the first usage (as long the compiled version is still in the native-method zone?). I ran the standard benchmarks in v2.3 and v2.5 and found v2.5 slower in screen io. My questions are: 1) Am I the only one that experiences this delay? Have I installed v2.5 incorrectly? I assume that the delay is caused by compiling methods to native code for the first time, after which it is stored in native-method zone. 2) Does this means a similar delay occurs on other platforms? Is in particular does this occur on Mac IIcx or Suns? roger whitney whitney@sdsu.edu ucsd!sdsu!whitney
obrien@aerospace.aero.org (Mike O'Brien) (12/15/89)
In article <1989Dec12.202546.11965@sdsu.edu> whitney@sdsu.sdsu.edu (Roger Whitney) writes: > > I have been using Smalltalk-80 v2.3 from ParcPlace on a Macintosh II >computer with 5M of memory. When I received my upgrade to v2.5 I noticed >a decrease in response of the windows. >1) Am I the only one that experiences this delay? Have I installed v2.5 >incorrectly? > > I assume that the delay is caused by compiling methods to native code >for the first time, after which it is stored in native-method zone. > >2) Does this means a similar delay occurs on other platforms? Is in >particular does this occur on Mac IIcx or Suns? It is definitely the case that the VM in 2.3 was written in assembler, whereas the 2.5 VM is coded in C (for portability reasons). My understanding (i.e. this is rumor) is that the C compilers for the Mac are not exactly up to snuff -- at least, the one ParcPlace used wasn't. To find out if the n-method zone is causing your problems, just run the benchmarks twice, and throw away the first reading. -- Mike O'Brien obrien@aerospace.aero.org
trl2@bunny.gte.com (Tom Lanning) (12/15/89)
In article <1989Dec12.202546.11965@sdsu.edu> whitney@sdsu.edu (Roger Whitney) writes: > a decrease in response of the windows. In particular the first time a > window (say the browser) is created one to three seconds pass before the > window appears. This is just long enough to be troubling. Response time ... > 1) Am I the only one that experiences this delay? Have I installed v2.5 > incorrectly? I noticed this also on the Macintosh II. Tom Lanning @ GTE Laboratories Incorporated Apple Link: D5185, Phone: (508) 466-4221 USENET: trl2@bunny.gte.com
khaw@parcplace.com (Mike Khaw) (12/20/89)
In article <1989Dec12.202546.11965@sdsu.edu> whitney@sdsu.sdsu.edu (Roger Whitney) writes: > > I have been using Smalltalk-80 v2.3 from ParcPlace on a Macintosh II >computer with 5M of memory. When I received my upgrade to v2.5 I noticed >a decrease in response of the windows. >1) Am I the only one that experiences this delay? Have I installed v2.5 >incorrectly? > > I assume that the delay is caused by compiling methods to native code >for the first time, after which it is stored in native-method zone. Dynamic translation overhead is normally imperceptible to the user. On the Mac, the main reason for the performance difference between 2.3 and 2.5 is the change from a 68000 assembler implementation to a C implementation of the VM. The VM's performance is only as good as the code emitted by the MPW 3.0 C compiler. >2) Does this means a similar delay occurs on other platforms? Is in >particular does this occur on Mac IIcx or Suns? On Suns, you will notice a slight decline in speed, part of which is because version 2.5 (and 2.4) only runs within a standard Sunview window, whereas 2.3 could take over the screen. Also, the standard VI is somewhat larger, so on a system with not a lot of free RAM, version 2.5 is more likely to page than 2.3. However, the overall change in speed is not so noticeable on Suns as on the Mac. -- Mike Khaw ParcPlace Systems, Inc., 1550 Plymouth St., Mountain View, CA 94043 Domain=khaw@parcplace.com, UUCP=...!{uunet,sun,decwrl}!parcplace!khaw
pardo@cs.washington.edu (David Keppel) (12/20/89)
>[Slowdown more noticable on the Mac than the Sun]
Just for general entertainment, the following is part of a table from
``ParcPlace Newsletter, Autumn 1989, number 4. The article is on
Objectworkds for Smalltalk-80.
;-D on ( Lies, damn lies, statistics, benchmarks, ... ) Pardo
Processor Machine ST version Performacne
----------------------------------------------------------------------
SPARC Sun SPARCstation 330 2.5 600
SPARC Sun SPARCstation 1 2.5 465
68030 Sun 3/4700 2.5 335
MIPS DECstation 3100 2.5 328
SPARC Sun 4/110 2.5 293
MIPS DECstation 2100 2.5 265
80386 Sun386i 2.5 200
68020 Apollo DN4500 2.3 147
68000 Macintosh II-cx 2.5 120
80386 IBM PS/2 Model 70 (20Mhz) 2.5 110
68020 Sun 3/50 2.5 104
68020 Apollo DN580 Turbo 2.3 85
68000 Macintosh SE 2.5 19
[There's more, I got tired of typing]
Note: Benchmark statistics are approximate. They vary depdending on
[...] A performacne level of roughly 100 is considered to give a user
full interactive ``feel.''
--
pardo@cs.washington.edu
{rutgers,cornell,ucsd,ubc-cs,tektronix}!uw-beaver!june!pardokhaw@parcplace.com (Mike Khaw) (12/22/89)
pardo@cs.washington.edu (David Keppel) writes: >Note: Benchmark statistics are approximate. They vary depdending on >[...] A performacne level of roughly 100 is considered to give a user >full interactive ``feel.'' The Smalltalk-80 benchmarks are normalized to Dorado = 100. The Xerox Dorado was a microcoded 16-bit machine whose interactive Smalltalk performance was one of the first to be considered "reasonably responsive". The published ParcPlace benchmarks are intended to provide some idea how Smalltalk-80 on a particular machine stacks up against the Dorado. Your own subjective rating may vary, depending on host window system performance, host graphics hardware performance, etc. -- Mike Khaw ParcPlace Systems, Inc., 1550 Plymouth St., Mountain View, CA 94043 Domain=khaw@parcplace.com, UUCP=...!{uunet,sun,decwrl}!parcplace!khaw