The situation with the gigE cards isn't entirely the fault of the OS (although the OS certainly isn't blameless). GigE cards simply generate *swamps* of interrupts. However, the nicer gigE cards give you the ability to turn on interrupt coallescing, where the card will queue up several packets before it sends an interrupt.
That does, of course, increase the latency, but unless you're in a clustering environment, you can generally handle a small hit in latency much better than you can handle a swarm of interrupts.
On the note of embedded PC architecture, yesterday I helped my boss crack open an old GPS to try and replace the interal battery. What did I see staring back at me? A 25 Mhz i386! It was, of course, in a *much* smaller package than the "regular" 386's were in, I suppose that Intel makes (or made) them on a smaller process for a while for projects like that. And I recall $15,000 Livingston PM3's which used an AMD 486. : )
surpassing my normal 18 month upgrade timetable and not looking like it's about to be retired any time soon
That is precisely the reason why dual-CPU machines are much more useful as desktops that the pundits would have you believe.
You'll hear over and over that "you won't use the other CPU", or "your apps won't take advantage of it." In reality, dual-CPU desktops are so much more responsive under load that they still feel "quick" much, much longer than their single-CPU counterparts will.
I have a dual Pentium 133. With NT4 on it, it's just as quick and usable as a P3/650 with Win98. CPU-intensive tasks do take longer, but the machine is still so responsive that you really don't notice it.
Here are the reasons: First, you *are* running more than one program at once, even if you don't notice it. On the Windows side, even with one app, you still have at least 30-40 different threads and/or programs running that you're not aware of - logging daemons, mouse daemons, graphics drivers, timers, and the copious other programs that X and/or Windows will launch. Then, if you're doing net or disk access, that's even more there.
The second reason is because of the interrupts. In a dual-CPU machine, one CPU can be getting hammered by interrupts, and you still have another to run other code, such as your GUI.
After having used dual-CPU workstations, I'll never build myself another single-CPU setup again. If that 2xP133 is still a nicely usable machine, I really can't imagine how long my 2xAthlonMP 1800 is going to last!
There are lots of air cleaners out there, intended for the woodworking industry. Very small particles of wood aren't good when you breathe them in. Here are a few from Grizzly, there are plenty of other manufacturers:
http://www.grizzly.com/products/item.cfm?itemnum be r=G5955
Sounds like a great tool for clustering, especially considering that the new motherboards have gigabit ethernet and a link directly to the northbridge/MCH.
There's just one problem with that... ethernet (even GigE) is *not* a good connection for clustering. Sure, the bandwidth is semi-decent, but the *latency* is the killer. Instead of a processer waiting a number of nanoseconds for memory (as with local memory), it'll end up waiting as much as milliseconds. That may not sound like much, but from nano to micro you jump seven orders of mangitude!
It looks like I was wrong this time. I know for a fact that my Tyan MPX boards have the option, but I double-checked, and my Asus A7M266-D does *not* have the scrubbing option. I may head over to their web site and see if there's a newer BIOS than the 1005A, and see if it supports it or not.
Asking hackers to look at your code to see how they would attack it. It almost smells of Open Source! A command-line server... sounds like Linux! I guess that imitation IS the most sincere form of flattery.
- Of course, having a small number of hackers look at your code still isn't as good as having a very large number look at your code.
I certainly couldn't find anything that says so. Of course, the *Athlon64* will only have a 64-bit (single-channel) memory interface, but the *Opterons* have 128-bit memory interfaces. All of AMD's docs have been saying so for quite some time, and the DECENT reviewers (read: Not Tom), which used the 128-bit memory interface enabled, showed significant improvements in memory bandwidth.
Both my Asus and Tyan MPX-based boards have the option available in the BIOS. I'll tell you, stuff a couple of gigs of RAM into the boards, turn on ECC, and the boot time sure seems excessive. Of course, it's actually doing valuable work, clearing out the memory so it can actually do the ECC, but still - waiting that long for the POST beep is a bit uncomfortable if you're not used to it.
The memory controller in the chip they tested handily beat out the dual-channel Nforce platform, and while it didn't beat the dual-channel DDR 400 memory of the P4, it wasn't too far behind.
Here's the cool part: That was an Athlon64. A desktop-oriented chip with a single-channel memory controller. The chips that are coming out on Tuesday are the "Hammer" multi-processing chips, targetted at servers, with *dual-channel* memory interfaces. Look at the memory numbers on the single-channel, and come close to doubling them. Now THAT will be impressive!
Here's hoping that the Hammers will really live up to the hype. Not only would they make good, solid servers, they'd definitely step up the competition with Intel, which means faster, cheaper chips for us.
Classic hardware review article. Take every bit of old-news, re-hashed, "everyone's seen it before" information you can find to pad out your 2-page article to at least 10 pages. Insert lots of huge pictures to pad it out even further. After all, you get paid by the banner impression, right?
Normally, most search engine's spidering methods are designed to be pretty nice to servers - such as only requesting pages once every 30 seconds or so.
However, I've seen times when the methods of some of the search engine spiders were foiled by such simple things as having a large number of virtual hosts on a machine. Combine that with a number of front-end machines all connected to the same database server, and things can get really nasty.
In one particularly bad incident, several fairly big-name search engines were spidering us simultaneously, and only hitting each domain name relatively infrequently. However, with 500+ on several front-end servers, and several search engines, we were getting something like 50-100 requests per *second* from the search engines. When those hits were to pages generated from the database, our servers kept up, but performance was definitely degraded.
So, where am I going? I see the potential for small bugs, weak algorithms, idiotic end-users, or even malicious end-users causing the same sort of havoc. Even if it weren't meant as an actual DDOS, it could certainly end up that way. And it would be much, much harder to prevent than merely blocking (or rate-limitting) requests from one company's spiders.
So what? My TNT2 lets me play very competitive Q3, DK2, Dungeon Siege, WC3, etc.. If you're telling me that it can pump out 60 FPS in Q3 with a few minor tweaks, but can't handle this guy's GUI, then maybe there's something wrong with his code.
Probably not, as the interconnect latency would kill you. It WILL, however, support multiple processers quite nicely. If your spanky new 3 GHz P4 isn't enough horsepower, no sweat - use TWO of them!
(not to mention what an 8-way Opteron system would be like...)
I would like to see more effort be put into original game genres instead of rehashing the old ones
They do, however, license out their engine, letting all of the companies who find that sort of genre profitable focus on the storyline, plot, etc. - and *still* deliver amazing graphics.
I would like to see more effort be put into original game genres instead of rehashing the old ones
Do you know which game genres get the most attention? The ones that make the most money. When the "original" genres are considered profitable, they'll see much more development.
I have nothing at all against said genres... but it's all simple economics.
Because the X-boxes have semi-decent hardware at a reasonable cost. Microsoft *loses* money selling them, in the hopes that they'll make it up later in the software.
So... you get a very decent machine at a pretty low price, and Micro$oft loses money in the process. That's a pretty nice idea.
If you don't see why it's a fight, then you don't understand economics!
Each company wants to succeed and make greater profits. In fact, they're *obligated* to their share holders to do just that. However, there's a finite market for these things, so they've all got to compete with each other - fighting.
Look at DEC's Alpha. A truly great product. But because the various companies that owned it weren't able to compete (fight) with it, look where it's at today.
steve
They can't beat them, so they're joining them!
on
Sun Considers Opteron
·
· Score: 2, Interesting
Sun has been hurting for a while - PC-based servers have been increasingly eating up Sun's market.
The Opterons are aimed squarely at a market segment that was hitherto tied to Sun and one or two other companies. If you wanted a highly-scalable 4- or 8-way 64-bit machine, you bent over, and Sun/IBM/DEC found your bank roll along the way.
Now, machines of those natures are coming from a commodity vendor. With a 128-bit DDR333 memory interface, each processer will have far more memory bandwidth than even the new Sun iiia's that were introduced today. And HyperThreading gives some pretty respectable inter-processor bandwidth. You think that Sun shouldn't be shaking in their boots? You bet they should.
In the end, they know that they're not going to win the lower end of the market. They simply can't compete with the economies of scale that AMD and Intel enjoy. Embracing the future is their only way to ensure that they keep at least a portion of that market.
While Suns do have their benefits, the company has been hurting because of several factors:
1. The cost boatloads of money. 2. Commodity hardware is catching up fast (and exceeding) with their lower-level servers 3. Software wedded to hardware.
The Opteron will give them extraoridinary value - a good, fast processer with buttloads of memory bandwidth at a far lower cost than the Sun processers. And it will let them offer competitively-priced low-to-midrange servers.
Oh, I don't know about that. Watson and crick came up with some realizations that were pretty important, even if you NEVER actually saw the molecule.
One of the realizations that they made was that there had to be a minimum of three identification units to code for each amino acid, and that more than three would be wasted. Now that sounds pretty simple, doesn't it? But even if you get to see the molecule, if you don't understand that you need to look at three sequential base pairs (a codon) together, and that you hence need to have the alignment right, then actually seeing the structure of the molecule doesn't get you very far.
For the curious, the reason that they needed a minimum of three base pairs was this:
You have four different ways to get a single base pair. If a single base pair were the functional unit, you could code for a total of four different amino acids. Way too low. With two sequential base pairs, you could code for sixteen. Still too low. If you use three sequential base pairs, you could code for up to 64 distinct amino acids, which more than covers the number of aminos that humans were known to be capable of producing.
Actually, the pages probably don't have a whole lot to do with it. Transmition is light. You can saturate a 100 mbit link without much hardware at all.
The real expense is in the database processing. The cost of performing even a relatively simple SQL query is generally a lot higher than the cost of serving out several large images.
Slashdot Mirror
Of course, they don't want to risk hurting sales of the Hammer, but it would still be nice to have more than one option, for crying out loud.
steve
The situation with the gigE cards isn't entirely the fault of the OS (although the OS certainly isn't blameless). GigE cards simply generate *swamps* of interrupts. However, the nicer gigE cards give you the ability to turn on interrupt coallescing, where the card will queue up several packets before it sends an interrupt.
That does, of course, increase the latency, but unless you're in a clustering environment, you can generally handle a small hit in latency much better than you can handle a swarm of interrupts.
On the note of embedded PC architecture, yesterday I helped my boss crack open an old GPS to try and replace the interal battery. What did I see staring back at me? A 25 Mhz i386! It was, of course, in a *much* smaller package than the "regular" 386's were in, I suppose that Intel makes (or made) them on a smaller process for a while for projects like that. And I recall $15,000 Livingston PM3's which used an AMD 486. : )
steve
surpassing my normal 18 month upgrade timetable and not looking like it's about to be retired any time soon
That is precisely the reason why dual-CPU machines are much more useful as desktops that the pundits would have you believe.
You'll hear over and over that "you won't use the other CPU", or "your apps won't take advantage of it." In reality, dual-CPU desktops are so much more responsive under load that they still feel "quick" much, much longer than their single-CPU counterparts will.
I have a dual Pentium 133. With NT4 on it, it's just as quick and usable as a P3/650 with Win98. CPU-intensive tasks do take longer, but the machine is still so responsive that you really don't notice it.
Here are the reasons: First, you *are* running more than one program at once, even if you don't notice it. On the Windows side, even with one app, you still have at least 30-40 different threads and/or programs running that you're not aware of - logging daemons, mouse daemons, graphics drivers, timers, and the copious other programs that X and/or Windows will launch. Then, if you're doing net or disk access, that's even more there.
The second reason is because of the interrupts. In a dual-CPU machine, one CPU can be getting hammered by interrupts, and you still have another to run other code, such as your GUI.
After having used dual-CPU workstations, I'll never build myself another single-CPU setup again. If that 2xP133 is still a nicely usable machine, I really can't imagine how long my 2xAthlonMP 1800 is going to last!
steve
There are lots of air cleaners out there, intended for the woodworking industry. Very small particles of wood aren't good when you breathe them in. Here are a few from Grizzly, there are plenty of other manufacturers:
m be r=G5955
http://www.grizzly.com/products/item.cfm?itemnu
steve
Sounds like a great tool for clustering, especially considering that the new motherboards have gigabit ethernet and a link directly to the northbridge/MCH.
There's just one problem with that... ethernet (even GigE) is *not* a good connection for clustering. Sure, the bandwidth is semi-decent, but the *latency* is the killer. Instead of a processer waiting a number of nanoseconds for memory (as with local memory), it'll end up waiting as much as milliseconds. That may not sound like much, but from nano to micro you jump seven orders of mangitude!
steve
It looks like I was wrong this time. I know for a fact that my Tyan MPX boards have the option, but I double-checked, and my Asus A7M266-D does *not* have the scrubbing option. I may head over to their web site and see if there's a newer BIOS than the 1005A, and see if it supports it or not.
steve
Asking hackers to look at your code to see how they would attack it. It almost smells of Open Source! A command-line server... sounds like Linux! I guess that imitation IS the most sincere form of flattery.
- Of course, having a small number of hackers look at your code still isn't as good as having a very large number look at your code.
steve
I certainly couldn't find anything that says so. Of course, the *Athlon64* will only have a 64-bit (single-channel) memory interface, but the *Opterons* have 128-bit memory interfaces. All of AMD's docs have been saying so for quite some time, and the DECENT reviewers (read: Not Tom), which used the 128-bit memory interface enabled, showed significant improvements in memory bandwidth.
steve
Of course I'm right. : )
Both my Asus and Tyan MPX-based boards have the option available in the BIOS. I'll tell you, stuff a couple of gigs of RAM into the boards, turn on ECC, and the boot time sure seems excessive. Of course, it's actually doing valuable work, clearing out the memory so it can actually do the ECC, but still - waiting that long for the POST beep is a bit uncomfortable if you're not used to it.
steve
The AMD MPX (dual Athlon) chipset also supports memory scrubbing.
A triangular manhole cover (well, an equilateral triangle anyway) is also immune to being dropped in the hole.
However, a ROUND hole is much, much more amenable to having an overweight public works employee climb through it than is a triangular hole.
steve
The memory controller in the chip they tested handily beat out the dual-channel Nforce platform, and while it didn't beat the dual-channel DDR 400 memory of the P4, it wasn't too far behind.
Here's the cool part: That was an Athlon64. A desktop-oriented chip with a single-channel memory controller. The chips that are coming out on Tuesday are the "Hammer" multi-processing chips, targetted at servers, with *dual-channel* memory interfaces. Look at the memory numbers on the single-channel, and come close to doubling them. Now THAT will be impressive!
Here's hoping that the Hammers will really live up to the hype. Not only would they make good, solid servers, they'd definitely step up the competition with Intel, which means faster, cheaper chips for us.
steve
Classic hardware review article. Take every bit of old-news, re-hashed, "everyone's seen it before" information you can find to pad out your 2-page article to at least 10 pages. Insert lots of huge pictures to pad it out even further. After all, you get paid by the banner impression, right?
Normally, most search engine's spidering methods are designed to be pretty nice to servers - such as only requesting pages once every 30 seconds or so.
However, I've seen times when the methods of some of the search engine spiders were foiled by such simple things as having a large number of virtual hosts on a machine. Combine that with a number of front-end machines all connected to the same database server, and things can get really nasty.
In one particularly bad incident, several fairly big-name search engines were spidering us simultaneously, and only hitting each domain name relatively infrequently. However, with 500+ on several front-end servers, and several search engines, we were getting something like 50-100 requests per *second* from the search engines. When those hits were to pages generated from the database, our servers kept up, but performance was definitely degraded.
So, where am I going? I see the potential for small bugs, weak algorithms, idiotic end-users, or even malicious end-users causing the same sort of havoc. Even if it weren't meant as an actual DDOS, it could certainly end up that way. And it would be much, much harder to prevent than merely blocking (or rate-limitting) requests from one company's spiders.
He's on a TNT2 for heaven sake!
So what? My TNT2 lets me play very competitive Q3, DK2, Dungeon Siege, WC3, etc.. If you're telling me that it can pump out 60 FPS in Q3 with a few minor tweaks, but can't handle this guy's GUI, then maybe there's something wrong with his code.
steve
Could D3 be made to run on two or more boxes
Probably not, as the interconnect latency would kill you. It WILL, however, support multiple processers quite nicely. If your spanky new 3 GHz P4 isn't enough horsepower, no sweat - use TWO of them!
(not to mention what an 8-way Opteron system would be like...)
steve
I would like to see more effort be put into original game genres instead of rehashing the old ones
They do, however, license out their engine, letting all of the companies who find that sort of genre profitable focus on the storyline, plot, etc. - and *still* deliver amazing graphics.
steve
I would like to see more effort be put into original game genres instead of rehashing the old ones
Do you know which game genres get the most attention? The ones that make the most money. When the "original" genres are considered profitable, they'll see much more development.
I have nothing at all against said genres... but it's all simple economics.
steve
Because the X-boxes have semi-decent hardware at a reasonable cost. Microsoft *loses* money selling them, in the hopes that they'll make it up later in the software.
So... you get a very decent machine at a pretty low price, and Micro$oft loses money in the process. That's a pretty nice idea.
If you don't see why it's a fight, then you don't understand economics!
Each company wants to succeed and make greater profits. In fact, they're *obligated* to their share holders to do just that. However, there's a finite market for these things, so they've all got to compete with each other - fighting.
Look at DEC's Alpha. A truly great product. But because the various companies that owned it weren't able to compete (fight) with it, look where it's at today.
steve
Sun has been hurting for a while - PC-based servers have been increasingly eating up Sun's market.
The Opterons are aimed squarely at a market segment that was hitherto tied to Sun and one or two other companies. If you wanted a highly-scalable 4- or 8-way 64-bit machine, you bent over, and Sun/IBM/DEC found your bank roll along the way.
Now, machines of those natures are coming from a commodity vendor. With a 128-bit DDR333 memory interface, each processer will have far more memory bandwidth than even the new Sun iiia's that were introduced today. And HyperThreading gives some pretty respectable inter-processor bandwidth. You think that Sun shouldn't be shaking in their boots? You bet they should.
In the end, they know that they're not going to win the lower end of the market. They simply can't compete with the economies of scale that AMD and Intel enjoy. Embracing the future is their only way to ensure that they keep at least a portion of that market.
steve
While Suns do have their benefits, the company has been hurting because of several factors:
1. The cost boatloads of money.
2. Commodity hardware is catching up fast (and exceeding) with their lower-level servers
3. Software wedded to hardware.
The Opteron will give them extraoridinary value - a good, fast processer with buttloads of memory bandwidth at a far lower cost than the Sun processers. And it will let them offer competitively-priced low-to-midrange servers.
steve
Oh, I don't know about that. Watson and crick came up with some realizations that were pretty important, even if you NEVER actually saw the molecule.
One of the realizations that they made was that there had to be a minimum of three identification units to code for each amino acid, and that more than three would be wasted. Now that sounds pretty simple, doesn't it? But even if you get to see the molecule, if you don't understand that you need to look at three sequential base pairs (a codon) together, and that you hence need to have the alignment right, then actually seeing the structure of the molecule doesn't get you very far.
For the curious, the reason that they needed a minimum of three base pairs was this:
You have four different ways to get a single base pair. If a single base pair were the functional unit, you could code for a total of four different amino acids. Way too low. With two sequential base pairs, you could code for sixteen. Still too low. If you use three sequential base pairs, you could code for up to 64 distinct amino acids, which more than covers the number of aminos that humans were known to be capable of producing.
steve
But virii *do* use DNA for code storage, and the article had absolutely nothing to contradict that.
What the article said is that prions do not use DNA for code storage. Prions != virii.
steve
Actually, the pages probably don't have a whole lot to do with it. Transmition is light. You can saturate a 100 mbit link without much hardware at all.
The real expense is in the database processing. The cost of performing even a relatively simple SQL query is generally a lot higher than the cost of serving out several large images.
steve