Firstly: yes, Swing is a bit slow (i haven't used 1.3, so i can't comment on that); this seems to be mostly due to rather dim implementations (using the win32 GDI rather than DirectDraw, not using hardware acceleration, etc), which will get fixed in the not-too-distant future. I have to admit that Swing is an embarrassment to Java: by far the most slow and buggy package.
Secondly: cache locality? I'm sorry to see that you know nothing whatsoever about modern OO compiler research - it's fascinating 8). There were some lovely papers in an ACM SIGPLAN bulletin a while back on just this; you can build a VM that tunes memory layout to improve locality, eg by clustering objects which are accessed together (only possible if you have a sophisticated GCish memory model, like Java) and splitting up objects so that the hot parts are clustered together (where they can get cached) and the cold parts are somewhere else (only possible if the VM has control over the code, as object splitting requires code rewriting). I don't think any of this research has made it into production VMs yet (we're also waiting for funky new GC strategies), which is a shame.
Thirdly: sockets; guilty as charged! The lack of asynchronous IO in Java is a crime.
JSR-51 is sorting out bringing it in, but i don't know when it will make it into the releases. Saying "use a real operating system!" is not an answer, by the way - passing the buck doesn't help Java and doesn't help Java programmers (except that maybe Sun will sell more boxen, make more money and invest more in Java...).
Uh? I assume we have different definitions of 'benchmark'. Running Conways's 'life' (an intensively computational program), a fibonacci calculator and a fast Fourier transform program, with C and Java implementations, and measuring their performance is what i call a benchmark. Could you perhaps enlighten me as to why it isn't?
FWIW, the
fibonacci test (not a very real-world test, i admit) shows that the IBM and HotSpot JVMs are both faster than optimised C. I'm a Java zealot, and even i don't believe that!
Hell yeah! I installed Linux on my box after Win95 went apeshit. Having finally got my disk partitioned right and picked all the options in the configurator, i had a running system. Wonderful! Now the time came to add PPP support... after a week, i wiped the disk and reinstalled Win95; it should have been one of the saddest days of my life, but i was *so* glad to see the back of Linux. Rebuilding the kernel? Editing half a bazillion config files? Filling in the gaps in the docs with guesswork? All i want is bloody dialup!
To be fair, i did get dialup, but it was a raw connection to the server (ie log in and type shell commands), not PPP. Enough to read mail, but only just. Of course, if i had a more recent distro, and some help on hand, i'm sure all would be fine.
Seen that iCube or whatever they call it? flashy buggers. The double-headed G4s are pretty good too.
Anyway, point is, you can get one for $6.2 k that substantially out-specs our university mainframe (although it might have some sort of nuclear-powered disk stuff, RAID or summat, vs the G4's Ultra160 SCSI). 6.2 isn't that much cash, really.
The real question, of course, is how it stacks up against PC hardware. Something comparable from Dell costs 6.7... that's 2x450 G4 vs 2x733 PIIIX., the rest is identical (more or less). I'm not up on G4 vs PIIIX, but it seems like PC hardware isn't substantially cheaper, and a Mac is ginuwine RISC hardware.
My real hope is that an official UNIX on a Mac will help the PowerPC linux community produce something a bit more wonderful.
Slashdot Mirror
Firstly: yes, Swing is a bit slow (i haven't used 1.3, so i can't comment on that); this seems to be mostly due to rather dim implementations (using the win32 GDI rather than DirectDraw, not using hardware acceleration, etc), which will get fixed in the not-too-distant future. I have to admit that Swing is an embarrassment to Java: by far the most slow and buggy package.
Secondly: cache locality? I'm sorry to see that you know nothing whatsoever about modern OO compiler research - it's fascinating 8). There were some lovely papers in an ACM SIGPLAN bulletin a while back on just this; you can build a VM that tunes memory layout to improve locality, eg by clustering objects which are accessed together (only possible if you have a sophisticated GCish memory model, like Java) and splitting up objects so that the hot parts are clustered together (where they can get cached) and the cold parts are somewhere else (only possible if the VM has control over the code, as object splitting requires code rewriting). I don't think any of this research has made it into production VMs yet (we're also waiting for funky new GC strategies), which is a shame.
Thirdly: sockets; guilty as charged! The lack of asynchronous IO in Java is a crime. JSR-51 is sorting out bringing it in, but i don't know when it will make it into the releases. Saying "use a real operating system!" is not an answer, by the way - passing the buck doesn't help Java and doesn't help Java programmers (except that maybe Sun will sell more boxen, make more money and invest more in Java ...).
Uh? I assume we have different definitions of 'benchmark'. Running Conways's 'life' (an intensively computational program), a fibonacci calculator and a fast Fourier transform program, with C and Java implementations, and measuring their performance is what i call a benchmark. Could you perhaps enlighten me as to why it isn't?
FWIW, the fibonacci test (not a very real-world test, i admit) shows that the IBM and HotSpot JVMs are both faster than optimised C. I'm a Java zealot, and even i don't believe that!
Hell yeah! I installed Linux on my box after Win95 went apeshit. Having finally got my disk partitioned right and picked all the options in the configurator, i had a running system. Wonderful! Now the time came to add PPP support ... after a week, i wiped the disk and reinstalled Win95; it should have been one of the saddest days of my life, but i was *so* glad to see the back of Linux. Rebuilding the kernel? Editing half a bazillion config files? Filling in the gaps in the docs with guesswork? All i want is bloody dialup!
To be fair, i did get dialup, but it was a raw connection to the server (ie log in and type shell commands), not PPP. Enough to read mail, but only just. Of course, if i had a more recent distro, and some help on hand, i'm sure all would be fine.
Seen that iCube or whatever they call it? flashy buggers. The double-headed G4s are pretty good too.
Anyway, point is, you can get one for $6.2 k that substantially out-specs our university mainframe (although it might have some sort of nuclear-powered disk stuff, RAID or summat, vs the G4's Ultra160 SCSI). 6.2 isn't that much cash, really.
The real question, of course, is how it stacks up against PC hardware. Something comparable from Dell costs 6.7 ... that's 2x450 G4 vs 2x733 PIIIX., the rest is identical (more or less). I'm not up on G4 vs PIIIX, but it seems like PC hardware isn't substantially cheaper, and a Mac is ginuwine RISC hardware.
My real hope is that an official UNIX on a Mac will help the PowerPC linux community produce something a bit more wonderful.