Ok, I didn't have much choice when I started, all I knew was that I wanted to do electronics and CS, and I wanted to become an engineer. Well, that left me with two choices here in Denmark.
Anyway, after my first year, my interests had moved completely away from electronics, and it was CS all the way.
However, I became rather dissatisfied with the CS department (at least with some of it), and numerics and _real_ computing has been my interest for the last years.
My point is, even though I've known ``exactly'' what I wanted since primary school, even CS is such a wide area, and you don't know what your real interests are going to be, before you found some subjects that weren't it.
If it's CS, find a university that does CS. Any university that does CS. You will end up doing stuff you didn't dream about anyway.
You have to get disappointed before you can be really happy. You have to hate subjects, before you can find the ones you love.
Didn't some (older?) SGI boxes require an IP connection to boot ? I think I heard that, havent seen it myself though...
It could be a nice feature to have in firmware. Connect to the vendors' ``bootup registration'' server, send a unique id of the hardware along with some information (like IP/subnet/routing).
The story I heard about the SGIs was, that it was an anti-theft measure. However, I wouldn't count on having Award or any of the other PC bioses hold an IP stack anytime soon though...
If someone wrote a daemon (regd) that was run right after the network came up, which did this registration, there could be a good chance that any asshole thief would at least boot the machine once before he wiped the drives.
So he created a macro that tells another piece of software to do something bad. Well if I tell anyone to go kill someone, and they do so, I wouldn't be the only one to face court actions would I ?
Dumb is, as dumb _does_.
I tell people on a regular basis, that I don't understand why so many people (including them) put up with systems that willingly will destroy everyting they work on. And get this, I either get blank stares, or some muttering about nobody wanting to switch to Linux....
Hell, if MS Word or any other product with just as little notion of security was ported to Linux, that would be just as bad. But why the fsck does people fail to see that security is just as needed in applications as it is in operating systems and front doors ?
I don't run netscape as root either. And my seti@home clients run with their own UID.
It all boils down to, if the source ain't open, you don't know what you got.
But I'll be the first to sell you the Eiffel tower if you tell me people will realize this.
I'm working on such a project. It's a distributed virtual machine that features automatic parallelization. It's currently in the very early stages of development, but so far it does parallelization of several test programs such as Fibonacci number calculation and others...
The homepage is at http://www.sslug.dk/TONS/, but you may find my report (finished earlier today) available from my homepage at http://ostenfeld.dk/~jakob/ an interesting read as it has a lot more information about what we're actually doing.
In short, we define a new language, a scheduler and some node servers to make up the distributed virtual machine, and we're in business.
However, it will take some time before this project becomes usefull though. But I beileve that we will be able to achieve very good performance (often comparable to hand-parallelized Fortran/C) in a large number of scientific computing applications. Read the report for the real argumentation:)
'wonder what kind of crap router they are running...
Router down for N hours. Oh, just 20 minutes before the router comes up, we'll just reboot the machine. Sure thing:)
Maybe the PPC guys ought to donate their crackbox to Microsoft and pre-configure it as a router. Then Win2K would really have it's chance to shine instead of being unreachable all the time because of a faulty router.
Well the whole thing isn't that much a surprise, but I'll give microsoft a point or two for entertainment value.
It's incredible that people cannot just choose the GPL or BSD licence variant that fits their needs the best, when they after all do create open source software.
Of course, perhaps they don't want people to use their stuff, and that's fair enough, it's too bad when the software is good and could benefit a lot of people, but it's fair enough if someone wants to prohibit others from using their creations.
Now someone may say that by choosing a licence that doesn't permit commercial use, binary distribution, or a licence like the rdist one, doesn't prohibit people from downloading the software themselves. Well it doesn't completely prohibit people from using it, but it does make the better software a second choice (or last resort), when so much other good software has a licence that allows a company like redhat/suse/... to back it.
Ironically, it may be the companies that will start a lot of the GPL projects in the future. We'll have anti-commercial shit^H^H^H^Hdevelopers changing their licences into ``everyone can use this software, but not if you use it in a way so that you make money'', or whatever, and the commercial parties will be left to write real Free software using GPL or similar.
Seems the US is the most Microsoft using region. Then Europe, then asia (dunno 'bout africa and others). Asia is far away from the states where Microsoft lives, both in miles and culture. No surprise that the irrational choice to just use microsoft stuff like everyone else is less widespread there.
Denmark where I live, used to be a very IBM loyal country. Banks used OS/2 when IBM was pushing that, and apple computers have never really been widely used. Sweden (right next to us), has always had a lot of apple machines. It seems that Denmark now is a Microsoft loyal country. The RIPE region has something like 50% of all web servers running Linux, and around 15% on windows. In denmark (which is a part of RIPE) those numbers are reversed. (What _is_ it with us and large firms?)
A friend I have went to Russia to tell some university people how they could solve their computing problems by buying PCs and putting Linux on them (that was a solution they could afford). But they weren't really interested in Linux, they wanted windows because that's what we use in the west. They smoke Marlboro there too:)
It's really fun to watch how culture and regional differences influence people's decisions. Meanwhile I'll just be an illoyal dane running the OS that works.
By turning copper into fibre, you only scale from the copper speed to somewhere nearer the light-speed (in vaccuum). You'll never _really_ scale.
As was pointed out in another post, another solution might be (temporarily) to decrease the length of the wiring. Still, in the long run, we need communication faster than the speed of light.
For an electric signal to travel thru less than 1 meter of copper, it still takes more than three nanoseconds (because we're travelling at at less than c).
Build your disks infinitely fast, with 1ps seek time, you'll still have more than several nanoseconds of latency. Build the same disks, but use superconductors/tunneling for transport, and you'll be rockin'.
I believe neither I, Hoare nor Einstin misunderstood anyting in this subject. Because neigher of us really cared/cares much.
As long as information travels at a speed slower than some fixed boundary, the information latency is going to become a bottleneck at some point in time.
Still, to get back to the original point of this discussion, disks have a long way to go, and CPUs, RAM etc. doesn't.
Expect superconducting/tunneling to become big, or expect disks to scale much better than CPUs/RAM/etc. in the future.
Or, perhaps you have some theory that I'm not aware of ? I'd appreciate to know about it.
Still, you have _miles_ of wire in your CPU alone.
And light travels thru copper wiring at about 0.6c.
Why do you think your IDE cables can't be any longer than [insert spec number here. it's less than 1 meter]. It's because of 0.6c and the latency that follows. The speed of light is a real limitation even in low end devices _today_.
Disks can improve latency for quite some time to come, because they're bound by the rotational speed rather than the speed of light, so no worries there. But the rest of the machine is dead in the water in this comparison.
I think we need CPU architecture to build upon single simple devices that communicate in a way that is not bound by the speed of light. Once the CPUs get there, the rest will follow, as is seen to fit. Read ``Communicating sequential process'' by Hoare et.al. and think about the communication happening way above c. That's sweet:)
But it doesn't either on CPUs or RAM (well it's done some, but it won't go on).
The fact is, information travels thru copper with less than the speed of light, and certainly never with more than this speed. We can only increase transfer rates (by using wider busses), we can't decrease latency, using current techonology that's bound by the speed of light.
Until we start using tunneling (that some physicists seem to believe can transfer information above the speed of light) we will not scale well in latencies.
Disk latencies, being so much larger than that of RAM and CPUs, will scale for some time to come. RAM and CPUs are almost dead in the water, until we get this breakthrough.
Until then, we can only re-arrange our problems to be transfer-rate bound, more than latency-bound. That requires thinking ofcourse.
I don't know much about physics, but I try to keep up by asking the right people the right questions. And there seem to be some possibilities in either tunneling or super-conducters, both which in some cases might allow information to travel above this limit of c. Meanwhile I'll be doing what I can do, to make problems rate-bound rather than latency-bound.
Drives is not a bottleneck. If you need performance, go with RAID.
I can transfer 34 MB/s from my disks (sustained), and around 100 MB/s from my main memory (RAM). As I see it, disks are not the problem here.
RAM today does some sort of parallel-like access to get the performance they get today. If you do the same with you disks, (and why wouldn't you, if performance is that important?) you get equal speedup. Also put more CPUs in your box, when the CPUs become the bottleneck.
Today in PCs, the PCI bus is the only real bottleneck, perhaps along with the memory subsystem. The memory subsystem has improved a lot over the last few years, but the majority of PCs still come with 32 bit 33MHz PCI busses.
When we start seeing motherboards with 64 bit 66 MHz PCI, that's when we'll again be looking at disks, cpus and memory as the bottleneck.
I'll probably be playing with vinum soon... But I'd be really surprised if there was any real performance difference between vinum and linux software raid.
The thing that kills disk performance, be it raid or not, is moving the disk heads, the speed of the disks and the busses, and memory bandwidth in the system.
Parity calculation and calculating what block goes to what disk, seems not to be an issue at all, with modern CPUs. A PII with linux raid-5 is able to do parity calculations on several hundred of megabytes per second. More than the memory/pci bus can do anyway.
However, AFAIK vinum does not yet implement raid-5. Since this guy wanted to play with raid-5 vinum may not be the choice for him. Please correct me if I'm mistaken here.
The phrase, intellectual property, are for a lot of us profane in itself.
The notion of intellectual property allows people to own ideas, which in itself is mistaken. Intellect, is not an idea, but a stream of ideas. An idea in itself is worthless, if it is not followed by a steady stream of new ideas. And the fact that we allow ownership of single ideas is simply mistaken, because one idea has infinitesimal value compared to the world of innovation we live in.
I think RMS did the right thing, by creating a licence that will satisfy everyone that acknowledges that innovation or intellect is not a matter of one idea. It is a matter of securing that enough people will be allowed to continually contribute new ideas to the original idea.
Owning one idea will only make further innovation suffocate. That is, ofcourse, a perfect situation for a lot of corporations and individuals who do not mind enslaving their users into this primitive proprietary world where problems are accepted because they will be fixed in the next release, which again probably will be available with a discount.
It's a sick world out there, but fortunately a lot of people have the clearsight to release their innovations in a way that acknowledges that they can never be the be all and end all of innvation by themselves.
Cooperation is the key word. And that just don't mix with the misguided desire not to share. If one wants to be alone with the idea and the profits, one will end up very alone with incomes long gone and an old idea.
I know someone who's running fluid dynamics simulations on two machines. One alpha (21164) and one dual PII.
Both machines are about a year old. The price tags where somewhere around $70K for the alpha and $3K for the PII.
When running two simulations on the PII and one on the alpha (one sim. per cpu), the PII is marginally slower than the alpha.
Ok, this is just one case. And I haven't seen the code and don't know which compiler options where used. Still, let's assume the code could be tuned to run twice as fast on the alpha, it's still one hell of a price/performance difference !
These results are far away from the SPECint/SPECfp numbers, but they are real world results achieved by real world users. I think that's what counts.
My point is, that although the alphas are really nice machines, and a new fast alpha will be faster than a new fast intel based machine, Compaq will have to get those price tags right.
I'd definitely get an alpha as my next box, if only I could afford it. A dual EV6, mmmmmmm... But there is no way in hell I can ever get that kind of money. I could buy a farm of PIIs for those money instead.
Linus himself is now working on a quad intel machine, instead of the dual alpha he had earlier. He said something like; the alpha is nice, but the intel machine is simply faster.
It would be great to see cheap alphas below 600 MHz, and really expensive ones above. That way us ordinary people could get our dirty hands on affordable alphas, and the companies that need the higher speeds can fund compaq by buying the high end ones. Much like intel is doing today.
FPGAs are well known technology. Wiring enough of them together, and programming them to do some specific task, will get that specific task done fast. No news there.
But it is a very special design. Reprogramming the FPGAs may be fast, but it is hard to program them to do a sequence of very different operations.
This is not quite unlike the Connection Machine (from Thinking Machines Corp.). A full CM has 64 thousand processors, but they can only do very specific tasks. If you program a CM to do matrix multiplication, it's lightning fast (or, at least it was in the days of the CM). But if you run a Perl interpreter, or any other not completely trivial or simple (matrix multiplication _is_ trivial and simple) piece of code on it, you will be _very_ disappointed.
Ofcourse these things are justified. Simple operations are done a lot in mathematical modelling. It will be very interesting to see what the supercomputer vendors can make up of a bunch of these FPGA boxes, wired to some standard processor boxes (to do the non-trivial stuff).
But don't think for a second, that we will be putting these things on the desktop, and have them running ``normal'' applications at a speed that is even comparative to a PII.
Why is it, that Linux users are always referred to as geeks ? I'm sick and tired of being referred to as a geek, by people who I consider more excentric than myself.
It seems that someone who writes code, or likes LaTeX, or has the ability to start a piece of software that doesn't have an icon on the desktop for it, are simply ``geeks''.
Anyone who's interested in anything, and I guess that sort of covers most people, knows something about that thing, that other's maybe don't. If a geek is someone who simply knows something about something which the observer doesn't, then most people are by definition ``geeks''. I wish the non-{Linux,UN*X} users would accept that fact, and stop referring to us as geeks.
It's damaging for the image of Linux, that people refer to it's user base as geeks, in a way that implies that most other people aren't geeks. They are, and so what. Get over that geek hump. Accept that Linux users knows something about something that non-Linux users don't, but don't imply that Linux is reserved for Linux-geeks only.
It's sick the way that people tend to put other people in categories, like geeks vs. non-geeks. Show me a non-geek and I'll show you someone who's lying about their interests.
Good thing with a full packet log though, running on a box with a non-affected kernel:)
Isn't this the first serious remote crash bug in the 2.2.x series ? There have been other bugs allright, and there still is, but I believe this is the first remote one.
That is not bad, if one thinks about the _huge_ changes that went into the 2.2 series from the 2.0 series. I'm pretty amazed we haven't seen a few more of these already... They may be coming though.
I would have expected a bug like this to appear sooner. And I would have expected more of these bugs. Well, either the developers are blessed with luck, or they are really skilled. We'll see which, in the next few months I guess. Luck don't last.
I really love the idea of sourceXchange and coSource. For development of new software, and for quick fixes and feature additions, this is surely the way to go.
So, I start this company and get paid $100K to develop some large piece of software, say, a complete 3D modelling and rendering package a'la 3D studio + Pixar renderman.
Great, the company that neede this got what they wanted, I got paid, and everyone's happy.
Until, a year later, someone wants not just a feature addon, but a larger change that will require some architectural changes in the software. Now what ? I'm of to Bahamas and can't be reached, and noone else really has any real insight into what goes on in my code.
How do we ensure, that a program developed thru these new efforts gets maintained, or at least, keeps it's maintainers ?
This is different from the current OpenSource development, because now, people write the software because they need it themselves, or because they truely love the idea of that piece of software they are creating. For those reasons, the software that currently gets developed, is likely to have a competent and loyal maintainer or ``head of development'' for many years.
With the get-paid-to-do-X-and-getouttahere, the software is not likely to hold this key person for very long time. I see a problem in that. But how do we solve it ?
I just wish that someone would install a real system on one of those boxes, and actually _compile_ some software (SPECfp for example), and then see how it worked. It would be a lot of fun just to see what Intel CPU egcs/pgcc should optimize for to get the best out of the K7.
Those ``Business benchmarks'' keep giving me the impression that those benchmarking guys are clueless. Why benchmark a CPU by measuring the disk throughput (or, we don't even know whether it's throughput, we just know it's business), instead of actually running something that is CPU bound for sure ?!
Also the tests that involve 3D graphics accelerators seem pointless to me. Sure, Quake may still use the CPU for something, but if this was supposed to be a CPU test and not a 3D graphics accelerator test, then why not let the CPU do the _entire_ job, instead of just what's left over from the 3D accel. ?
Just the fact that they posted the 3D and disk numbers, along with a notice saying that the disks and 3D accelerators weren't the same, makes me think that these people really have no idea of what they are doing.
Although I'm running on Intel boxes exclusively (oh, except for the PA7000 in the basement), I'd really consider my next box to be AMD K-something based. If they can ship fast CPUs at $250, and I can buy a motherboard that supports four or eight of them, then I don't care if the FPU is 10% slower than a Xeon. The price/performance is going to make this choice _really_ easy.
My previous box got around 40 BogoMIPS, my current one around 400, and I'll definitely be going for 4000 next time:)~ Eat me.
Re:Once again 4 intel ethernet cards
on
NOS Crossroads
·
· Score: 1
Channel bonding is already implemented in the Beowulf patches to the kernel (somewhere at cesdis.gsfc.nasa.gov). It does round-robin between a number of ethernet adapters.
How about the SMP scores in the test ? It seems that they rated the systems after how much each CPU was occupied... NT is famous for hot-potato'ing, swapping processes between CPUs for no reason (other than to pollute the L2 cache) what so ever. So it scores high. Linux gets the job done without using that much CPU, so it scores low. What a strange world this is...
I don't mean to dismiss the results of the benchmark as fake. But there are problems with these benchmarks. Everyone can configure systems to perform in any way, relative to eachother. A benchmark can be made to show anything one wishes to show. We need to see the technicalities behind these tests. It would be great if ZD and any others had links to a technical description of what they did, some page where they wheren't afraid of mentioning words that doesn't rhyme on ``icon'' and ``click''.
Oh, one last thing: The scorecard says RAID support, and Linux scores low. Well, if it had said Hardware-RAID support, it would probably have been true. But today, with Ingo Molnar's Software-RAID patches, Linux outperforms any hardware-RAID solution for a fraction of the cost.
From my personal experience, I can not believe that NT should beat Linux in performance on such a massive scale as the thing we're seeing here.
But the fact that Linux is faster than NT does not imply, that it is the operating system/kernel of choice by IT professionals. We use GNU/Linux, because it works great, it's fast, and we won't have to spend money on licences every time we need another box running. But that is not the way an IT professional with a near-million-dollar budget thinks. If Linux can't ``utilize'' (the word we interpret as ``hog'' or ``crawl on'') a huge machine, then Linux is just not the system of choice, for running the really expensive/important server jobs.
Imagine running a million-dollar operation on 1000 dollar hardware. It's possible, and a lot of you reading this, probably do so allready.
But it is just not the way many IT pros think.
We will keep improving Linux and the GNU system, because we need it and love it. But there will always be people with too much money and too little brains, that will interpret the fact that the system also runs on small machines, as a clear sign that it is a system for small machines only.
NT is often considered a high-end server OS, and I believe it is because it only can run on high-end server hardware.
Those people, the ones with small minds and even smaller ideas, will eat performance comparisons like the one we've seen here. They will actually believe the results without even the slightest sceptism.
And the rest of us will keep running the world on the OS that has proven it's worth, not in the press, but in the field.
Slashdot Mirror
It really doesn't matter that much where you go.
:)
Ok, I didn't have much choice when I started, all I knew was that I wanted to do electronics and CS, and I wanted to become an engineer.
Well, that left me with two choices here in Denmark.
Anyway, after my first year, my interests had moved completely away from electronics, and it was CS all the way.
However, I became rather dissatisfied with the CS department (at least with some of it), and numerics and _real_ computing has been my interest for the last years.
My point is, even though I've known ``exactly'' what I wanted since primary school, even CS is such a wide area, and you don't know what your real interests are going to be, before you found some subjects that weren't it.
If it's CS, find a university that does CS. Any university that does CS. You will end up doing stuff you didn't dream about anyway.
You have to get disappointed before you can be really happy. You have to hate subjects, before you can find the ones you love.
(Shit I sound old.
Didn't some (older?) SGI boxes require an IP connection to boot ? I think I heard that, havent seen it myself though...
:)
It could be a nice feature to have in firmware. Connect to the vendors' ``bootup registration'' server, send a unique id of the hardware along with some information (like IP/subnet/routing).
The story I heard about the SGIs was, that it was an anti-theft measure.
However, I wouldn't count on having Award or any of the other PC bioses hold an IP stack anytime soon though...
If someone wrote a daemon (regd) that was run right after the network came up, which did this registration, there could be a good chance that any asshole thief would at least boot the machine once before he wiped the drives.
Do I smell a project here ?
So he created a macro that tells another piece of software to do something bad.
Well if I tell anyone to go kill someone, and they do so, I wouldn't be the only one to face court actions would I ?
Dumb is, as dumb _does_.
I tell people on a regular basis, that I don't understand why so many people (including them) put up with systems that willingly will destroy everyting they work on. And get this, I either get blank stares, or some muttering about nobody wanting to switch to Linux....
Hell, if MS Word or any other product with just as little notion of security was ported to Linux, that would be just as bad. But why the fsck does people fail to see that security is just as needed in applications as it is in operating systems and front doors ?
I don't run netscape as root either. And my seti@home clients run with their own UID.
It all boils down to, if the source ain't open, you don't know what you got.
But I'll be the first to sell you the Eiffel tower if you tell me people will realize this.
I'm working on such a project. It's a distributed virtual machine that features automatic parallelization. It's currently in the very early stages of development, but so far it does parallelization of several test programs such as Fibonacci number calculation and others...
:)
The homepage is at http://www.sslug.dk/TONS/, but you may find my report (finished earlier today) available from my homepage at http://ostenfeld.dk/~jakob/ an interesting read as it has a lot more information about what we're actually doing.
In short, we define a new language, a scheduler and some node servers to make up the distributed virtual machine, and we're in business.
However, it will take some time before this project becomes usefull though. But I beileve that we will be able to achieve very good performance (often comparable to hand-parallelized Fortran/C) in a large number of scientific computing applications. Read the report for the real argumentation
'wonder what kind of crap router they are running...
:)
Router down for N hours. Oh, just 20 minutes before the router comes up, we'll just reboot the machine. Sure thing
Maybe the PPC guys ought to donate their crackbox to Microsoft and pre-configure it as a router. Then Win2K would really have it's chance to shine instead of being unreachable all the time because of a faulty router.
Well the whole thing isn't that much a surprise, but I'll give microsoft a point or two for entertainment value.
It's incredible that people cannot just choose the GPL or BSD licence variant that fits their needs the best, when they after all do create open source software.
Of course, perhaps they don't want people to use their stuff, and that's fair enough, it's too bad when the software is good and could benefit a lot of people, but it's fair enough if someone wants to prohibit others from using their creations.
Now someone may say that by choosing a licence that doesn't permit commercial use, binary distribution, or a licence like the rdist one, doesn't prohibit people from downloading the software themselves. Well it doesn't completely prohibit people from using it, but it does make the better software a second choice (or last resort), when so much other good software has a licence that allows a company like redhat/suse/... to back it.
Ironically, it may be the companies that will start a lot of the GPL projects in the future. We'll have anti-commercial shit^H^H^H^Hdevelopers changing their licences into ``everyone can use this software, but not if you use it in a way so that you make money'', or whatever, and the commercial parties will be left to write real Free software using GPL or similar.
I'd be interested in knowing a little about that prestige project at DTU, since I'm actually studying there...
/.'ed properly.
Hell, there might even be an english translation of the press release or whatever it is so it can be
Seems the US is the most Microsoft using region. Then Europe, then asia (dunno 'bout africa and others).
:)
Asia is far away from the states where Microsoft lives, both in miles and culture. No surprise that the irrational choice to just use microsoft stuff like everyone else is less widespread there.
Denmark where I live, used to be a very IBM loyal country. Banks used OS/2 when IBM was pushing that, and apple computers have never really been widely used.
Sweden (right next to us), has always had a lot of apple machines.
It seems that Denmark now is a Microsoft loyal country. The RIPE region has something like 50% of all web servers running Linux, and around 15% on windows. In denmark (which is a part of RIPE) those numbers are reversed. (What _is_ it with us and large firms?)
A friend I have went to Russia to tell some university people how they could solve their computing problems by buying PCs and putting Linux on them (that was a solution they could afford). But they weren't really interested in Linux, they wanted windows because that's what we use in the west. They smoke Marlboro there too
It's really fun to watch how culture and regional differences influence people's decisions.
Meanwhile I'll just be an illoyal dane running the OS that works.
By turning copper into fibre, you only scale from the copper speed to somewhere nearer the light-speed (in vaccuum). You'll never _really_ scale.
As was pointed out in another post, another solution might be (temporarily) to decrease the length of the wiring. Still, in the long run, we need communication faster than the speed of light.
For an electric signal to travel thru less than 1 meter of copper, it still takes more than three nanoseconds (because we're travelling at at less than c).
Build your disks infinitely fast, with 1ps seek time, you'll still have more than several nanoseconds of latency.
Build the same disks, but use superconductors/tunneling for transport, and you'll be rockin'.
I believe neither I, Hoare nor Einstin misunderstood anyting in this subject. Because neigher of us really cared/cares much.
As long as information travels at a speed slower than some fixed boundary, the information latency is going to become a bottleneck at some point in time.
Still, to get back to the original point of this discussion, disks have a long way to go, and CPUs, RAM etc. doesn't.
Expect superconducting/tunneling to become big, or expect disks to scale much better than CPUs/RAM/etc. in the future.
Or, perhaps you have some theory that I'm not aware of ? I'd appreciate to know about it.
Sorry to follow up on my own posing.
But light of course doesn't trave thru cooper. What I mean was: Information travels thru copper at around 0.6 times the speed of light.
Still, you have _miles_ of wire in your CPU alone.
:)
And light travels thru copper wiring at about 0.6c.
Why do you think your IDE cables can't be any longer than [insert spec number here. it's less than 1 meter]. It's because of 0.6c and the latency that follows. The speed of light is a real limitation even in low end devices _today_.
Disks can improve latency for quite some time to come, because they're bound by the rotational speed rather than the speed of light, so no worries there. But the rest of the machine is dead in the water in this comparison.
I think we need CPU architecture to build upon single simple devices that communicate in a way that is not bound by the speed of light. Once the CPUs get there, the rest will follow, as is seen to fit. Read ``Communicating sequential process'' by Hoare et.al. and think about the communication happening way above c. That's sweet
Seek time doesn't scale, ok you got that right.
But it doesn't either on CPUs or RAM (well it's done some, but it won't go on).
The fact is, information travels thru copper with less than the speed of light, and certainly never with more than this speed. We can only increase transfer rates (by using wider busses), we can't decrease latency, using current techonology that's bound by the speed of light.
Until we start using tunneling (that some physicists seem to believe can transfer information above the speed of light) we will not scale well in latencies.
Disk latencies, being so much larger than that of RAM and CPUs, will scale for some time to come. RAM and CPUs are almost dead in the water, until we get this breakthrough.
Until then, we can only re-arrange our problems to be transfer-rate bound, more than latency-bound. That requires thinking ofcourse.
I don't know much about physics, but I try to keep up by asking the right people the right questions. And there seem to be some possibilities in either tunneling or super-conducters, both which in some cases might allow information to travel above this limit of c. Meanwhile I'll be doing what I can do, to make problems rate-bound rather than latency-bound.
Drives is not a bottleneck. If you need performance, go with RAID.
I can transfer 34 MB/s from my disks (sustained), and around 100 MB/s from my main memory (RAM). As I see it, disks are not the problem here.
RAM today does some sort of parallel-like access to get the performance they get today. If you do the same with you disks, (and why wouldn't you, if performance is that important?) you get equal speedup. Also put more CPUs in your box, when the CPUs become the bottleneck.
Today in PCs, the PCI bus is the only real bottleneck, perhaps along with the memory subsystem. The memory subsystem has improved a lot over the last few years, but the majority of PCs still come with 32 bit 33MHz PCI busses.
When we start seeing motherboards with 64 bit 66 MHz PCI, that's when we'll again be looking at disks, cpus and memory as the bottleneck.
I'll probably be playing with vinum soon... But I'd be really surprised if there was any real performance difference between vinum and linux software raid.
The thing that kills disk performance, be it raid or not, is moving the disk heads, the speed of the disks and the busses, and memory bandwidth in the system.
Parity calculation and calculating what block goes to what disk, seems not to be an issue at all, with modern CPUs. A PII with linux raid-5 is able to do parity calculations on several hundred of megabytes per second. More than the memory/pci bus can do anyway.
However, AFAIK vinum does not yet implement raid-5. Since this guy wanted to play with raid-5 vinum may not be the choice for him. Please correct me if I'm mistaken here.
The phrase, intellectual property, are for a lot of us profane in itself.
The notion of intellectual property allows people to own ideas, which in itself is mistaken. Intellect, is not an idea, but a stream of ideas. An idea in itself is worthless, if it is not followed by a steady stream of new ideas. And the fact that we allow ownership of single ideas is simply mistaken, because one idea has infinitesimal value compared to the world of innovation we live in.
I think RMS did the right thing, by creating a licence that will satisfy everyone that acknowledges that innovation or intellect is not a matter of one idea. It is a matter of securing that enough people will be allowed to continually contribute new ideas to the original idea.
Owning one idea will only make further innovation suffocate. That is, ofcourse, a perfect situation for a lot of corporations and individuals who do not mind enslaving their users into this primitive proprietary world where problems are accepted because they will be fixed in the next release, which again probably will be available with a discount.
It's a sick world out there, but fortunately a lot of people have the clearsight to release their innovations in a way that acknowledges that they can never be the be all and end all of innvation by themselves.
Cooperation is the key word. And that just don't mix with the misguided desire not to share. If one wants to be alone with the idea and the profits, one will end up very alone with incomes long gone and an old idea.
I know someone who's running fluid dynamics simulations on two machines. One alpha (21164) and one dual PII.
Both machines are about a year old. The price tags where somewhere around $70K for the alpha and $3K for the PII.
When running two simulations on the PII and one on the alpha (one sim. per cpu), the PII is marginally slower than the alpha.
Ok, this is just one case. And I haven't seen the code and don't know which compiler options where used. Still, let's assume the code could be tuned to run twice as fast on the alpha, it's still one hell of a price/performance difference !
These results are far away from the SPECint/SPECfp numbers, but they are real world results achieved by real world users. I think that's what counts.
My point is, that although the alphas are really nice machines, and a new fast alpha will be faster than a new fast intel based machine, Compaq will have to get those price tags right.
I'd definitely get an alpha as my next box, if only I could afford it. A dual EV6, mmmmmmm... But there is no way in hell I can ever get that kind of money. I could buy a farm of PIIs for those money instead.
Linus himself is now working on a quad intel machine, instead of the dual alpha he had earlier. He said something like; the alpha is nice, but the intel machine is simply faster.
It would be great to see cheap alphas below 600 MHz, and really expensive ones above. That way us ordinary people could get our dirty hands on affordable alphas, and the companies that need the higher speeds can fund compaq by buying the high end ones. Much like intel is doing today.
FPGAs are well known technology. Wiring enough of them together, and programming them to do some specific task, will get that specific task done fast. No news there.
But it is a very special design. Reprogramming the FPGAs may be fast, but it is hard to program them to do a sequence of very different operations.
This is not quite unlike the Connection Machine (from Thinking Machines Corp.). A full CM has 64 thousand processors, but they can only do very specific tasks. If you program a CM to do matrix multiplication, it's lightning fast (or, at least it was in the days of the CM). But if you run a Perl interpreter, or any other not completely trivial or simple (matrix multiplication _is_ trivial and simple) piece of code on it, you will be _very_ disappointed.
Ofcourse these things are justified. Simple operations are done a lot in mathematical modelling. It will be very interesting to see what the supercomputer vendors can make up of a bunch of these FPGA boxes, wired to some standard processor boxes (to do the non-trivial stuff).
But don't think for a second, that we will be putting these things on the desktop, and have them running ``normal'' applications at a speed that is even comparative to a PII.
Why is it, that Linux users are always referred to as geeks ? I'm sick and tired of being referred to as a geek, by people who I consider more excentric than myself.
It seems that someone who writes code, or likes LaTeX, or has the ability to start a piece of software that doesn't have an icon on the desktop for it, are simply ``geeks''.
Anyone who's interested in anything, and I guess that sort of covers most people, knows something about that thing, that other's maybe don't. If a geek is someone who simply knows something about something which the observer doesn't, then most people are by definition ``geeks''. I wish the non-{Linux,UN*X} users would accept that fact, and stop referring to us as geeks.
It's damaging for the image of Linux, that people refer to it's user base as geeks, in a way that implies that most other people aren't geeks. They are, and so what. Get over that geek hump. Accept that Linux users knows something about something that non-Linux users don't, but don't imply that Linux is reserved for Linux-geeks only.
It's sick the way that people tend to put other people in categories, like geeks vs. non-geeks. Show me a non-geek and I'll show you someone who's lying about their interests.
Well, there goes 70+ days of uptime. Damn.
:)
:)
Good thing with a full packet log though, running on a box with a non-affected kernel
Isn't this the first serious remote crash bug in the 2.2.x series ? There have been other bugs allright, and there still is, but I believe this is the first remote one.
That is not bad, if one thinks about the _huge_ changes that went into the 2.2 series from the 2.0 series. I'm pretty amazed we haven't seen a few more of these already... They may be coming though.
I would have expected a bug like this to appear sooner. And I would have expected more of these bugs. Well, either the developers are blessed with luck, or they are really skilled. We'll see which, in the next few months I guess. Luck don't last.
Good work guys ! Also on the fix btw.
I really love the idea of sourceXchange and coSource. For development of new software, and for quick fixes and feature additions, this is surely the way to go.
So, I start this company and get paid $100K to develop some large piece of software, say, a complete 3D modelling and rendering package a'la 3D studio + Pixar renderman.
Great, the company that neede this got what they wanted, I got paid, and everyone's happy.
Until, a year later, someone wants not just a feature addon, but a larger change that will require some architectural changes in the software. Now what ? I'm of to Bahamas and can't be reached, and noone else really has any real insight into what goes on in my code.
How do we ensure, that a program developed thru these new efforts gets maintained, or at least, keeps it's maintainers ?
This is different from the current OpenSource development, because now, people write the software because they need it themselves, or because they truely love the idea of that piece of software they are creating. For those reasons, the software that currently gets developed, is likely to have a competent and loyal maintainer or ``head of development'' for many years.
With the get-paid-to-do-X-and-getouttahere, the software is not likely to hold this key person for very long time. I see a problem in that. But how do we solve it ?
Exactly !
Problem is, where's the SpecInt Icon in Win98 ?
Those people are helpless...
It'd be great to see AMD support the EGCS project, and eventually have Cygnus or whomever publish some _real_ benchmarks.
But then again, I must be dreaming...
I just wish that someone would install a real system on one of those boxes, and actually _compile_ some software (SPECfp for example), and then see how it worked. It would be a lot of fun just to see what Intel CPU egcs/pgcc should optimize for to get the best out of the K7.
:)~ Eat me.
Those ``Business benchmarks'' keep giving me the impression that those benchmarking guys are clueless. Why benchmark a CPU by measuring the disk throughput (or, we don't even know whether it's throughput, we just know it's business), instead of actually running something that is CPU bound for sure ?!
Also the tests that involve 3D graphics accelerators seem pointless to me. Sure, Quake may still use the CPU for something, but if this was supposed to be a CPU test and not a 3D graphics accelerator test, then why not let the CPU do the _entire_ job, instead of just what's left over from the 3D accel. ?
Just the fact that they posted the 3D and disk numbers, along with a notice saying that the disks and 3D accelerators weren't the same, makes me think that these people really have no idea of what they are doing.
Although I'm running on Intel boxes exclusively (oh, except for the PA7000 in the basement), I'd really consider my next box to be AMD K-something based. If they can ship fast CPUs at $250, and I can buy a motherboard that supports four or eight of them, then I don't care if the FPU is 10% slower than a Xeon. The price/performance is going to make this choice _really_ easy.
My previous box got around 40 BogoMIPS, my current one around 400, and I'll definitely be going for 4000 next time
Channel bonding is already implemented in the Beowulf patches to the kernel (somewhere at cesdis.gsfc.nasa.gov). It does round-robin between a number of ethernet adapters.
How about the SMP scores in the test ? It seems that they rated the systems after how much each CPU was occupied... NT is famous for hot-potato'ing, swapping processes between CPUs for no reason (other than to pollute the L2 cache) what so ever. So it scores high. Linux gets the job done without using that much CPU, so it scores low. What a strange world this is...
I don't mean to dismiss the results of the benchmark as fake. But there are problems with these benchmarks. Everyone can configure systems to perform in any way, relative to eachother. A benchmark can be made to show anything one wishes to show. We need to see the technicalities behind these tests. It would be great if ZD and any others had links to a technical description of what they did, some page where they wheren't afraid of mentioning words that doesn't rhyme on ``icon'' and ``click''.
Oh, one last thing: The scorecard says RAID support, and Linux scores low. Well, if it had said Hardware-RAID support, it would probably have been true. But today, with Ingo Molnar's Software-RAID patches, Linux outperforms any hardware-RAID solution for a fraction of the cost.
From my personal experience, I can not believe that NT should beat Linux in performance on such a massive scale as the thing we're seeing here.
But the fact that Linux is faster than NT does not imply, that it is the operating system/kernel of choice by IT professionals. We use GNU/Linux, because it works great, it's fast, and we won't have to spend money on licences every time we need another box running. But that is not the way an IT professional with a near-million-dollar budget thinks. If Linux can't ``utilize'' (the word we interpret as ``hog'' or ``crawl on'') a huge machine, then Linux is just not the system of choice, for running the really expensive/important server jobs.
Imagine running a million-dollar operation on 1000 dollar hardware. It's possible, and a lot of you reading this, probably do so allready.
But it is just not the way many IT pros think.
We will keep improving Linux and the GNU system, because we need it and love it. But there will always be people with too much money and too little brains, that will interpret the fact that the system also runs on small machines, as a clear sign that it is a system for small machines only.
NT is often considered a high-end server OS, and I believe it is because it only can run on high-end server hardware.
Those people, the ones with small minds and even smaller ideas, will eat performance comparisons like the one we've seen here. They will actually believe the results without even the slightest sceptism.
And the rest of us will keep running the world on the OS that has proven it's worth, not in the press, but in the field.