Re: objective-c: how slow ?

[Chris Hanson]

At 3:00 PM -0500 8/31/01, Erik M. Buck wrote:
> I personally have never seen Java perform even half as fast as Objective-C.
> Every Java program I have used on the desktop suffered from slow buggy
> graphics and was prone to occasional halts/freezes presumably while garbage
> was collected.  Every Java implementation that I have seen as suffered from
> a large memory footprint for the VM and then problems resulting from the
> heap size not being large enough (Java heap sizes are limited for security
> so that a Java program can not crash a machine by allocating too much
> memory: (This is a job of the operating system and not a language, but what
> can you do ?)

Just as an aside, Apple has claimed that for equivalent code on the 
same hardware and OS (Mac OS X 10.0.x), a 100% Java WebObjects 5 
application will perform better than an Objective-C

I have not seen this in the real world (though I haven't done any 
comparisons), and I don't doubt the validity of Erik's real-world 
observations.

What I have seen the real world are Smalltalk implementations with 
*excellent* performance on ancient hardware.  QKS SmalltalkAgents had 
a full virtual machine architecture, full dynamic messaging, full 
garbage collection, etc. and performed so well on a 25MHz 68040 with 
12MB of RAM that a normal end-user application developed in 
SmalltalkAgents didn't differ in "feel" from a native Macintosh 
application.  It's an existence proof that something *like* Java can 
be done well.  (Sadly, QKS was just a little before their time.  They 
tried to make their "Agents Operating System" VM portable and license 
it to vendors in the early to mid 1990s, before Java and without 
Sun's marketing muscle.  It was a great idea, it's too bad they 
weren't successful.  At least they're still around!)

Apple and Connectix have also provided existence proofs that virtual 
machines don't have to be slow.  The 68040 emulator that's part of 
the Mac OS was fast enough on the original Power Macintosh hardware 
(60MHz PowerPC 601) that Apple didn't have to rewrite their device 
drivers for the PowerPC.  The second-generation "dynamic 
recompilation" (JIT) emulator ran 68040 code faster on a 1995-era 
PowerMac than any 68040 released to that date.  Connectix Virtual PC 
4, whose emulator was written by the same wizard as Apple's 
second-generation 68040 emulator, performs well enough on a 400MHz G3 
that I can run QuickBooks Pro on Windows 2000 and it feels like a 
233MHz Pentium.

There's no reason Java has to be slow -- that it is says more about 
Sun than it does about languages running on virtual machines, or 
about garbage collection, or about dynamic languages.

  -- Chris

PS - The question shouldn't be "Objective-C: How slow?" but rather 
"How fast?"  After all, Objective-C is native code and uses manual 
memory management.  True dynamic dispatch also tends to have better 
branch-prediction behavior on modern processors than C++ virtual 
function dispatch.  (Search comp.lang.smalltalk for a URL to the 
paper describing this discovery.  As instruction pipelines get 
deeper, this will become even more of an issue.)