I can't agree with you. Having just benchmarked Linux on a dual CPU Xeon, Solaris on a dual Opteron and NetBSD on single CPU configurations of both machines, I can say that the SMP support in Linux is:
Unpredicatable - performance degradation is not very linear as load increases, unlike Solaris and NetBSD.
Slow - a dual 2.2Ghz Opteron running Solaris outperformed a dual 3.4Ghz Xeon running Linux by a small margin.
Costly (in performance terms) - comparing the increase in performance between uniprocessor and SMP Linux kernels, under moderate load the SMP kernel was a far worse performer than the uniprocessor one - and this in a highly multithreaded application.
I must admit I was surprised at the poor showing of Linux, as I expected Solaris 9 to be less optimised for x86 architectures than Linux. I didn't have a chance to test Linux on the Opteron machine, so I wont write it off completely - perhaps the Xeon is a piss poor 64 bit processor. My other suspicion is that the Linux 2.6 kernel series has been a step down from the 2.4 one.
In a uniprocessor configuration with a native JDK 1.4.2, NetBSD outperformed Linux. I put this down to a cleaner design and implementation, as again, Linux performance degraded in a much less linear way.
They took a long time to get a UNIX solution out (some people wanted to not be totally stuck on proprietary VMS). Then when RISC took off and had much improved performance with less transistors, they were still stuck on VAX.
Most people using Unix in the early to mid 1980s ran it on a VAX - they licensed Unix from AT&T but used the BSD version. Digital brought out their own Unix variant in the 1980s (it was 4.3BSD based and called Ultrix). Ultrix disappeared when the Alpha came out with their version of OSF/1, which was eventually called Tru64. Any late adoption of Unix was not what sunk Digital.
Whether the Alpha was a latecomer to the RISC party is debatable as the PowerPC came later and is still rather successful. Performance wise, the Alpha was definitely the best of the era. However, DEC failed to push it as a viable mid-range server architecture, concentrating on the high end for too long.
As for Digitals PC line, I don't recall them being any better or worse than Compaq, which were considered top notch in the mid-nineties. In fact I particularily liked Digital laptops. They did try to push VaxStations as a viable alternative to the PC, whereas they were hardly an alternative to Unix workstations (down to whether you liked VMS I suppose).
my theory is:... Good products don't need much if any advertising.
That was the attitude over at Digital. Their head honcho believed that they would dominate by just having the best products, and that marketing was therefore a waste of time. Instead they got bought out by some commodity PC outfit called Compaq.
Great, so you've got a box chock full of cheapo components (IDE drives for Gods sake), no enterprise support and no certification that it'll run a decent operating system. If you want disposable servers, then that's up to you, but I prefer to pay a little more for a V20z that outperforms the Poweredge and runs Solaris.
I had bad experiences going from Digital to Compaq and then from Compaq to HP. Somewhere along the line, all the DEC expertise with clustering and RAID hardware just disappeared. Perhaps that expertise was what HP thought they were getting when they bought Compaq, but both companies RAID hardware screws up on a regular basis. I'm glad to be back on Sun hardware again...
So how come AMD does so much better with Opteron at clock speeds comparable to US IV? 1.4GHz Opterons are very fast indeed.
Depends on the application. The system I am currently building consists of three database servers in a pyramid, and a eight webservers. My team tried a number of configurations - including all Opteron boxes and all SPARC boxes. In the end the SPARC boxes outperformed the Opterons as database servers, but the Opteron boxes outperformed the SPARCs as webservers.
We also compared Solaris and Linux on the Opterons, as well as Linux on an equivalent priced Dell 1850. The Solaris/Opteron combination outperformed Linux by a small margin, and was much more consistent under load. Our conclusion has been that Solaris is much better at taking advantage of the dual processors in our Opterons.
As usual, TI can't fab a chip for toffee so Sun's having to sell some as 6-core (two broken cores disabled) and underclocked to 1GHz to get their money's worth out of the poor yield.
Uh, that's standard practice for processors. When a new design or fabrication plant starts up, the yield os low. Over time the yoeld goes up. Selling some of the yield that as lower spec chips is also standard practice, and is one of the main reasons you can overclock processors. If you by a slower Pentium chip, it may have actually been manufactured to run at a higher clock rate but sold as a lower speed chip because it failed tests at the highest clock rate.
Sun always boasted it had the second-largest CPU design team on the planet. So why is UltraSPARC so slow?
To do an Intel, and get the clock speed up, Sun would need Intel's massive financial resources. Then they could build fabrication plants dedicated to producing silicon with smaller tracks. As Sun doesn't have the capital or the fabs, they going to be working with Fujitsu on the new SPARC designs.
There is nothing wrong with the SPARC design, or the instruction set itself. In fact it's better than the Pentium which is handicapped by backwards compatablity with the poorly designed x86 instruction set.
I had to assume you were referring to Mao because the current Chinese regime has not murdered thousands of its own citizens, as the Iraqi regime did.
Depends on how narrowly you define murder, but the current Chinese regime has taken decisions that have killed hundreds of thousands. The flooding caused by ill advised dam projects, lack of even basic safety standards in major industries (particularily mining) and the low standard of healthcare despite a vast budget for military expenditure are examples of that.
As for not killing dissenters, they are sometimes killed, but the closed trials make it difficult to assess what they are charged with and how convincing the evidence is. Other dissenters are sent to labour camps, and some suffer the old Stalinist favourite of incarceration in mental hospitals - because you'd have to be mad to not want to live under a benevelent Communist party wouldn't you?
The DX-1 was manufactured in 1985, and only 140 of them were made. The DX-7 came out in 1983, the result of research into digital synthesis that Yamaha had been conducting since the late 1970's. If I recall correctly, Yamaha licensed the theoretical implementation of FM synthesis from a university and made it a practical reality.
Bullshit. Linux fanboys claim that their kernel runs on an architecture when it boots single user on one machine. Then the patches don't make it into the vanilla Linus kernel tree, and the porters lose interest. For example, check out the various MIPS, Vax and m68k ports - all stagnant. Even more "mainstream" architectures like Sparc are regularily broken in Linus sanctioned releases. On the distribution front, I've had Debian fail to install time and time agaon in non-x86 architectures. NetBSD by comparison is an *entire* operating system (not just a kernel) that is auto-built on all supported architectures on an almot daily basis, from a single source tree.
There is a lot of confusion around there about what exactly is open source, free, copyrighted and/or proprietary software. I suggest to everyone to read the Free Software Definition and the FAQs about the GNU GPL.
That's not "free" it's "Free" - it only applies to the FSF use of the word. To paraphrase Nietszche, "there are freedoms but no freedom".
Macs running OS10 are arguably genetic Unix (only arguably, because while they build on a BSD userland, and BSD is genetic Unix, the XNU kernel is not a descendent of any Unix system.)
MacOS X is built on top of the Mach kernel, which was developed from the BSD Unix kernel. See the opening parargraphs on the Darwin page, and the Unix timeline for 1985. Mach was forked from 4.2BSD.
I switched Automatic Updates off on the one Windows XP machine I have to deal with. The reason why? Because XP will automatically reboot the machine if the update tells it to. That makes XP unusable in a professional environment where I'm often running batch jobs that run for indeterminate lengths of time.
His reliance on static variables in functions is also yucky. Say hello to code that's never going to be thread safe. The path stuff the parent poster noted is the most breathtaking error though...
Having seen Windows 1.0, I'd argue that MicroSoft were either distinterested in, or unable to copy the the Apple interface. Windows 1.0 was more primitive than the GEM interface on an Atari ST, and in no way comparable to the Apple interface. This was probably down to the limited specifications of the PC.
Slashdot Mirror
Better kernel handling of multiprocessor systems.
I can't agree with you. Having just benchmarked Linux on a dual CPU Xeon, Solaris on a dual Opteron and NetBSD on single CPU configurations of both machines, I can say that the SMP support in Linux is:
I must admit I was surprised at the poor showing of Linux, as I expected Solaris 9 to be less optimised for x86 architectures than Linux. I didn't have a chance to test Linux on the Opteron machine, so I wont write it off completely - perhaps the Xeon is a piss poor 64 bit processor. My other suspicion is that the Linux 2.6 kernel series has been a step down from the 2.4 one.
In a uniprocessor configuration with a native JDK 1.4.2, NetBSD outperformed Linux. I put this down to a cleaner design and implementation, as again, Linux performance degraded in a much less linear way.
When I stopped laughing, I saw your followup post. I realised you weren't being sarcastic, and I started laughing again.
They took a long time to get a UNIX solution out (some people wanted to not be totally stuck on proprietary VMS). Then when RISC took off and had much improved performance with less transistors, they were still stuck on VAX.
Most people using Unix in the early to mid 1980s ran it on a VAX - they licensed Unix from AT&T but used the BSD version. Digital brought out their own Unix variant in the 1980s (it was 4.3BSD based and called Ultrix). Ultrix disappeared when the Alpha came out with their version of OSF/1, which was eventually called Tru64. Any late adoption of Unix was not what sunk Digital.
Whether the Alpha was a latecomer to the RISC party is debatable as the PowerPC came later and is still rather successful. Performance wise, the Alpha was definitely the best of the era. However, DEC failed to push it as a viable mid-range server architecture, concentrating on the high end for too long.
As for Digitals PC line, I don't recall them being any better or worse than Compaq, which were considered top notch in the mid-nineties. In fact I particularily liked Digital laptops. They did try to push VaxStations as a viable alternative to the PC, whereas they were hardly an alternative to Unix workstations (down to whether you liked VMS I suppose).
my theory is: ... Good products don't need much if any advertising.
That was the attitude over at Digital. Their head honcho believed that they would dominate by just having the best products, and that marketing was therefore a waste of time. Instead they got bought out by some commodity PC outfit called Compaq.
Great, so you've got a box chock full of cheapo components (IDE drives for Gods sake), no enterprise support and no certification that it'll run a decent operating system. If you want disposable servers, then that's up to you, but I prefer to pay a little more for a V20z that outperforms the Poweredge and runs Solaris.
I had bad experiences going from Digital to Compaq and then from Compaq to HP. Somewhere along the line, all the DEC expertise with clustering and RAID hardware just disappeared. Perhaps that expertise was what HP thought they were getting when they bought Compaq, but both companies RAID hardware screws up on a regular basis. I'm glad to be back on Sun hardware again ...
I even have a really old e250 that is still cranking along ;)
Where I work we've still got a bunch of SparcStation 5 and 20s in the server room. They just run and run.
So how come AMD does so much better with Opteron at clock speeds comparable to US IV? 1.4GHz Opterons are very fast indeed.
Depends on the application. The system I am currently building consists of three database servers in a pyramid, and a eight webservers. My team tried a number of configurations - including all Opteron boxes and all SPARC boxes. In the end the SPARC boxes outperformed the Opterons as database servers, but the Opteron boxes outperformed the SPARCs as webservers.
We also compared Solaris and Linux on the Opterons, as well as Linux on an equivalent priced Dell 1850. The Solaris/Opteron combination outperformed Linux by a small margin, and was much more consistent under load. Our conclusion has been that Solaris is much better at taking advantage of the dual processors in our Opterons.
As usual, TI can't fab a chip for toffee so Sun's having to sell some as 6-core (two broken cores disabled) and underclocked to 1GHz to get their money's worth out of the poor yield.
Uh, that's standard practice for processors. When a new design or fabrication plant starts up, the yield os low. Over time the yoeld goes up. Selling some of the yield that as lower spec chips is also standard practice, and is one of the main reasons you can overclock processors. If you by a slower Pentium chip, it may have actually been manufactured to run at a higher clock rate but sold as a lower speed chip because it failed tests at the highest clock rate.
Sun always boasted it had the second-largest CPU design team on the planet. So why is UltraSPARC so slow?
To do an Intel, and get the clock speed up, Sun would need Intel's massive financial resources. Then they could build fabrication plants dedicated to producing silicon with smaller tracks. As Sun doesn't have the capital or the fabs, they going to be working with Fujitsu on the new SPARC designs.
There is nothing wrong with the SPARC design, or the instruction set itself. In fact it's better than the Pentium which is handicapped by backwards compatablity with the poorly designed x86 instruction set.
If that was the only reason, they have now none left: Sun's Opteron servers have no serial, no parallel port and no PS2.
That's weird, I don't know what I plugged a PS2 mouse and keyboard into on my Sun V20z, but they sure seem to work.
Are you stalking me?
No crash here: Firefox 1.0.6 on NetBSD 2.0.2.
I had to assume you were referring to Mao because the current Chinese regime has not murdered thousands of its own citizens, as the Iraqi regime did.
Depends on how narrowly you define murder, but the current Chinese regime has taken decisions that have killed hundreds of thousands. The flooding caused by ill advised dam projects, lack of even basic safety standards in major industries (particularily mining) and the low standard of healthcare despite a vast budget for military expenditure are examples of that.
As for not killing dissenters, they are sometimes killed, but the closed trials make it difficult to assess what they are charged with and how convincing the evidence is. Other dissenters are sent to labour camps, and some suffer the old Stalinist favourite of incarceration in mental hospitals - because you'd have to be mad to not want to live under a benevelent Communist party wouldn't you?
mysql obviously made the decision that data integrity is more valuable then speed.
That would be a first for MySQL.
Isn't getting the jvm working on any BSD a horrible experience?
Havng downloaded three files, I installed a native JDK 1.4 onto NetBSD like this:
cd /usr/pkgsrc/wip/jdk14
make install
Tough I know ...
The DX-1 was manufactured in 1985, and only 140 of them were made. The DX-7 came out in 1983, the result of research into digital synthesis that Yamaha had been conducting since the late 1970's. If I recall correctly, Yamaha licensed the theoretical implementation of FM synthesis from a university and made it a practical reality.
The weird lead line in Good Vibrations sounds like a Theremin. It certainly predates Moog synthesisers.
Ernie Ball had some unused copies of MS software that they didn't know about and were fined $100,000.
Damn. So that's why a new scratchplate for my Stingray cost so much.
It's a myth that NetBSD runs on more than Linux.
Bullshit. Linux fanboys claim that their kernel runs on an architecture when it boots single user on one machine. Then the patches don't make it into the vanilla Linus kernel tree, and the porters lose interest. For example, check out the various MIPS, Vax and m68k ports - all stagnant. Even more "mainstream" architectures like Sparc are regularily broken in Linus sanctioned releases. On the distribution front, I've had Debian fail to install time and time agaon in non-x86 architectures. NetBSD by comparison is an *entire* operating system (not just a kernel) that is auto-built on all supported architectures on an almot daily basis, from a single source tree.
There is a lot of confusion around there about what exactly is open source, free, copyrighted and/or proprietary software. I suggest to everyone to read the Free Software Definition and the FAQs about the GNU GPL.
That's not "free" it's "Free" - it only applies to the FSF use of the word. To paraphrase Nietszche, "there are freedoms but no freedom".
Macs running OS10 are arguably genetic Unix (only arguably, because while they build on a BSD userland, and BSD is genetic Unix, the XNU kernel is not a descendent of any Unix system.)
MacOS X is built on top of the Mach kernel, which was developed from the BSD Unix kernel. See the opening parargraphs on the Darwin page, and the Unix timeline for 1985. Mach was forked from 4.2BSD.
I switched Automatic Updates off on the one Windows XP machine I have to deal with. The reason why? Because XP will automatically reboot the machine if the update tells it to. That makes XP unusable in a professional environment where I'm often running batch jobs that run for indeterminate lengths of time.
His reliance on static variables in functions is also yucky. Say hello to code that's never going to be thread safe. The path stuff the parent poster noted is the most breathtaking error though ...
Having seen Windows 1.0, I'd argue that MicroSoft were either distinterested in, or unable to copy the the Apple interface. Windows 1.0 was more primitive than the GEM interface on an Atari ST, and in no way comparable to the Apple interface. This was probably down to the limited specifications of the PC.