The very reasons interfaces are changing to serial is that it's very problematic to keep signals synchronized in parallell. So it's either "fast" serial, or "slow" parallell, and it looks like serial is winning. While fast parallell obviously would be the best, don't hold your breath for it.
Hypertransport is a good example of a serial/parallel interface. To get more bandwidth, you add more links in parallel, each of which is a serial link capable of carrying the whole traffic on its own, just slower.
Yeah, but what drive actually delivers 320 Mbytes/second? As long as the connection between the controller and each drive can keep up with the drive, the connection is fast enough.
Scsi is a bus. I have a box here with 5x10K drives, at 49 MB/s each, easily able to saturate its ultra 160 bus. These days, that box is nothing special.
When their stock rises 40% on a worthless lawsuit.
Actually, their stock rose steadily and dramatically over the two week period just before the lawsuit. Looks like insider trading, and a matter requiring the involvement of the SEC.
Don't underestimate the gravity of this move by Sun. This is a MAJOR blow to linux, because Sun holds large clout with big companies who worry about Linux's stability and support issues.
It's no major blow to Linux, it's a self-inflicted wound for Sun. I predict they'll about-face by tomorrow, once they figure that out.
Unfortunately, these dinosours give us something that we need: Choice.
SCO/Caldera doesn't give anybody anything that they need, except perhaps for existing, locked-in customers. When they go extinct as they inevitably will as a result of this misguided strategem, nobody in the Linux community will miss them a bit, quite the contrary.
Yep. I've got the impression that IBM has only recently started contributing scalability-related stuff, i.e. their work will be seen in 2.6 series of kernels.
What they did is contribute people. The Linux work done by those people is original. I know this to be true, since I've been working with them from the start (on the VM/scalability front) and we have the irc logs where the ideas were born and developed, including ideas originated by non-IBM people and developed further by IBM people. IBM also contributed the patented RCU lockless sharing algorithm and contributed a license as well, as required by the GPL. Besides that, I don't know of a single instance of IBM people contributing anything other hard work, original engineering and creativity.
What I was trying to say is more or less what someone in the article such much better than I - That each of the two patches is around an 80% solution, and the two together are a 95% solution.
Yes, that's accurate. Sorry, I was a little agressive about your otherwise informative post. Now, what's really exciting about all of this is how Linus pulled this completely new, elegant technique out of his sleeve, almost instantly.
Needing 100 Mb/s or Gb/s of bandwidth to run X clients over network means X is barely suitable for cable modem or DSL
It works perfectly well at 5 MB/s, i.e., over my wlan. You don't want to do this with an animation, but I run graphical editors, mailers etc. this way all the time and it's scarcely distinguishable from running locally.
Actually, Linus's patch doesn't improve things any better than the scheduler patch it is Linus's patch combined with the scheduler patch that make it such a huge improvement. Again... its the COMBO patch that's arousing so much excitement.
Where do you get your facts from? Read the thread again. Linus's patch by itself produces a dramatic improvement, so does Ingo's for different loads. The two together provide improvement over a wider range of loads.
How about someone setting up a list for Linux Hackers willing to spend time migrating SCO customers to the linux platform for free. To fight back I am willing to spend some of my time.
It could work out nicely for you too, if the company you help then decides to come back to you for some paid work. Feel the quality of the goods and all that.
I totally agree. Every experience I've had with Caldera/SCO products was horrible... granted, it was their Linux products (I know nothing of their Unix).
I can help fill in the blanks. I also had horrible experiences with their DR Dos product *after* it was supposedly open-sourced, then closed again. I wanted to fix some flaws in their implementation of MOVE, but they basically told me to take a hike - neither would they fix the flaws, nor would they give me the source so I could contribute the fixes myself.
There was a short, glorious moment when Caldera were the good guys - nothing like getting a little revenge on Microsoft for their strategems re DR Dos. It passed. I no longer have the slightest respect for Caldera, oops, SCO.
To think that they could go from copying a 500mhz chip to producing chips that can keep up with 3ghz x86 chips (and this is assuming that the x86 market stands still for a few years) seems to me a bit of a stretch.
Funny, I recall almost precisely the same thing being said about Japanese dram production, round about the time of 16K (that's bits) drams.
The US is screwed software wise, largely due to patents. We'll be importing more than we export in a decade.
A decade is a huge overestimate. Just look at what's happening in India, Russia and China. Hint: much top-grade open source code is coming out of these countries already. They aren't hobbled by patents+lawyers stupidity.
*If* Google wants to keep their trademark, and there are good reasons for them to do so, then this is exactly what they need to do, whether you like it or not.
Whether Google Inc. likes it or not, "to google" has already become a verb. By leaning on dictionary authors, who must merely document the truth, they can only burn karma, and even so it can only delay the inevitable. Clearly, the correct solution for them is to accept "google" as a word and direct their legal efforts towards defending "Google".
Just for the hell of it, the other day I typed 'Windows' into Google. I got (about) 57,600,000 results.
Then I typed 'Linux'. I got (about) 53,700,000 results.
Now, one could write a whole book on how unscientific those statistics are, but it was still interesting to see a (damned near) 1:1 ratio.
It's closer to 1:1 than you think. The query "windows -microsoft -nt -xp -98 -2000" gives almost exclusively links to the kind of windows that people have in their walls, and there are 4,730,000 of those, more than making up the difference you saw.
This may be unscientific for a variety of reasons, but reproducibility is not one of them
The distro has been dead for years, but I have yet to see a distro that is as easy to install as Corel Linux.
It's not dead, it was purchased from Corel by a group backed by Linux Global Partners, renamed Xandros, and developed further. I had them send a copy to my brother for Christmas.
If you ask him for his thoughts on something like Linux, he'll correct you and say, "Well, I can't tell you that, but I'll tell you about my thoughts on GNU/Linux...
Personally, I'd prefer listen to that remark, which at least shows some humor, than your aggressive attack, which only shows bile.
I really think that the 4-way system niche is so small that even AMD went to try to fill it, it would not be worth their investment.
AMD certainly will cater to that niche, since that's what hypertransport was designed for (among other things). However, 4 ways don't make sense for rendering applications. A networked cluster is significantly cheaper for the same throughput.
I haven't had the slowdown problems you're describing. I don't see exactly the same times, but the difference is rather small.
OK, I reran the tests, and the difference is small as you say. For root_fs, a 105 Meg file, copied to a 500 Mhz laptop from a dual 1 Ghz server over 100 MHz ethernet, I get:
time scp deathray:bigfile. real 0m20.926s
time scp -c blowfish deathray:bigfile. real 0m16.672s
If you're only offering files to a group of users who you can give passwords to, you could even use SCP. (Secure copy...uses sshd on the server side)
Scp is a horribly slow way to transfer a file, 5 times slower than http or ftp (which are nearly the same, for big files) on my network. Use scp if you need the security, or if you don't care about bandwidth sucking and just want the nice easy interface.
To get it to work I have to run a driver from the GATOS project that is from their CVS, patch it heavily so it supports things like XV, and run it on a bleeding edge copy of XFree just to get the semi-crudy TV out to work. It's the worst time I've ever had getting X to work correctly in years and years.
But remember how video codecs were like that a couple years ago, now they're faster, more reliable and better quality than on Windows, in spite of the intentional obstacles in the way.
ATI seems to think they support Linux, but they don't really do anything, as far as I can tell.
Yes, it smacks of evil. Well, continuous gentle pressure on ATI is the best medicine, and do like you did, buy from the competition. They'll get the message. The only wrong thing to do is stay silent.
In what way was my post a troll?... (somebody abusing moderation for whatever reason)
Hey, I didn't Linux was a particularly good choice. But, you've answered your own question - a mobile phone doesn't need USB (though it would be handy for docking your phone and syncing contacts), IDE drivers, AGP support or anything like that. I'm willing to bet that this "Linux" kernel is little more than a scheduler and some memory management. However, Linux wouldn't be my first choice for a mobile phone OS.
You missed a big one: these days, a mobile phone needs a tcp/ip stack. And don't forget, this phone does video. Also, virtual memory (as in protection, not as in swapping) is something you really want in a device like this, as it's going to be running applications of varying quality. That's just off the top of my head.
The embedded device of today is a lot more like a computer and a lot less like a toaster.
Hey, I didn't Linux was a particularly good choice. But, you've answered your own question - a mobile phone doesn't need USB (though it would be handy for docking your phone and syncing contacts), IDE drivers, AGP support or anything like that. I'm willing to bet that this "Linux" kernel is little more than a scheduler and some memory management. However, Linux wouldn't be my first choice for a mobile phone OS.
You missed a big one: these days, a mobile phone needs a tcp/ip stack. And don't forget, this phone does video. Also, virtual memory (as in protection, not as in swapping) is something you really want in a device like this, as it's going to be running applications of varying quality. That's just off the top of my head.
The embedded device of today is a lot more like a computer and a lot less like a toaster.
Slashdot Mirror
The very reasons interfaces are changing to serial is that it's very problematic to keep signals synchronized in parallell. So it's either "fast" serial, or "slow" parallell, and it looks like serial is winning. While fast parallell obviously would be the best, don't hold your breath for it.
Hypertransport is a good example of a serial/parallel interface. To get more bandwidth, you add more links in parallel, each of which is a serial link capable of carrying the whole traffic on its own, just slower.
Yeah, but what drive actually delivers 320 Mbytes/second? As long as the connection between the controller and each drive can keep up with the drive, the connection is fast enough.
Scsi is a bus. I have a box here with 5x10K drives, at 49 MB/s each, easily able to saturate its ultra 160 bus. These days, that box is nothing special.
When their stock rises 40% on a worthless lawsuit.
Actually, their stock rose steadily and dramatically over the two week period just before the lawsuit. Looks like insider trading, and a matter requiring the involvement of the SEC.
Don't underestimate the gravity of this move by Sun. This is a MAJOR blow to linux, because Sun holds large clout with big companies who worry about Linux's stability and support issues.
It's no major blow to Linux, it's a self-inflicted wound for Sun. I predict they'll about-face by tomorrow, once they figure that out.
Unfortunately, these dinosours give us something that we need: Choice.
SCO/Caldera doesn't give anybody anything that they need, except perhaps for existing, locked-in customers. When they go extinct as they inevitably will as a result of this misguided strategem, nobody in the Linux community will miss them a bit, quite the contrary.
Yep. I've got the impression that IBM has only recently started contributing scalability-related stuff, i.e. their work will be seen in 2.6 series of kernels.
What they did is contribute people. The Linux work done by those people is original. I know this to be true, since I've been working with them from the start (on the VM/scalability front) and we have the irc logs where the ideas were born and developed, including ideas originated by non-IBM people and developed further by IBM people. IBM also contributed the patented RCU lockless sharing algorithm and contributed a license as well, as required by the GPL. Besides that, I don't know of a single instance of IBM people contributing anything other hard work, original engineering and creativity.
Honestly, though, I don't think will effect their Linux strategy either. It's just a short-term marketing/PR stunt.
I'd call it a pretty serious loss of karma points. And they were doing so well...
What I was trying to say is more or less what someone in the article such much better than I - That each of the two patches is around an 80% solution, and the two together are a 95% solution.
;-)
Yes, that's accurate. Sorry, I was a little agressive about your otherwise informative post. Now, what's really exciting about all of this is how Linus pulled this completely new, elegant technique out of his sleeve, almost instantly.
I think it was AA that said that.
Andrew Morton, actually
Needing 100 Mb/s or Gb/s of bandwidth to run X clients over network means X is barely suitable for cable modem or DSL
It works perfectly well at 5 MB/s, i.e., over my wlan. You don't want to do this with an animation, but I run graphical editors, mailers etc. this way all the time and it's scarcely distinguishable from running locally.
Actually, Linus's patch doesn't improve things any better than the scheduler patch it is Linus's patch combined with the scheduler patch that make it such a huge improvement. Again... its the COMBO patch that's arousing so much excitement.
Where do you get your facts from? Read the thread again. Linus's patch by itself produces a dramatic improvement, so does Ingo's for different loads. The two together provide improvement over a wider range of loads.
How about someone setting up a list for Linux Hackers willing to spend time migrating SCO customers to the linux platform for free. To fight back I am willing to spend some of my time.
It could work out nicely for you too, if the company you help then decides to come back to you for some paid work. Feel the quality of the goods and all that.
I totally agree. Every experience I've had with Caldera/SCO products was horrible... granted, it was their Linux products (I know nothing of their Unix).
I can help fill in the blanks. I also had horrible experiences with their DR Dos product *after* it was supposedly open-sourced, then closed again. I wanted to fix some flaws in their implementation of MOVE, but they basically told me to take a hike - neither would they fix the flaws, nor would they give me the source so I could contribute the fixes myself.
There was a short, glorious moment when Caldera were the good guys - nothing like getting a little revenge on Microsoft for their strategems re DR Dos. It passed. I no longer have the slightest respect for Caldera, oops, SCO.
To think that they could go from copying a 500mhz chip to producing chips that can keep up with 3ghz x86 chips (and this is assuming that the x86 market stands still for a few years) seems to me a bit of a stretch.
Funny, I recall almost precisely the same thing being said about Japanese dram production, round about the time of 16K (that's bits) drams.
The US is screwed software wise, largely due to patents. We'll be importing more than we export in a decade.
A decade is a huge overestimate. Just look at what's happening in India, Russia and China. Hint: much top-grade open source code is coming out of these countries already. They aren't hobbled by patents+lawyers stupidity.
*If* Google wants to keep their trademark, and there are good reasons for them to do so, then this is exactly what they need to do, whether you like it or not.
Whether Google Inc. likes it or not, "to google" has already become a verb. By leaning on dictionary authors, who must merely document the truth, they can only burn karma, and even so it can only delay the inevitable. Clearly, the correct solution for them is to accept "google" as a word and direct their legal efforts towards defending "Google".
Just for the hell of it, the other day I typed 'Windows' into Google. I got (about) 57,600,000 results.
Then I typed 'Linux'. I got (about) 53,700,000 results.
Now, one could write a whole book on how unscientific those statistics are, but it was still interesting to see a (damned near) 1:1 ratio.
It's closer to 1:1 than you think. The query "windows -microsoft -nt -xp -98 -2000" gives almost exclusively links to the kind of windows that people have in their walls, and there are 4,730,000 of those, more than making up the difference you saw.
This may be unscientific for a variety of reasons, but reproducibility is not one of them
My favorite is their 4-way SMP system that fits into a briefcase.
To put it in simple terms, SMP means several processors share the same memory. I doubt that the four separate motherboards shown are capable of this.
With Hypertransport, this little bit of fiction might become reality in the not too distant future.
The distro has been dead for years, but I have yet to see a distro that is as easy to install as Corel Linux.
It's not dead, it was purchased from Corel by a group backed by Linux Global Partners, renamed Xandros, and developed further. I had them send a copy to my brother for Christmas.
If you ask him for his thoughts on something like Linux, he'll correct you and say, "Well, I can't tell you that, but I'll tell you about my thoughts on GNU/Linux...
Personally, I'd prefer listen to that remark, which at least shows some humor, than your aggressive attack, which only shows bile.
I really think that the 4-way system niche is so small that even AMD went to try to fill it, it would not be worth their investment.
AMD certainly will cater to that niche, since that's what hypertransport was designed for (among other things). However, 4 ways don't make sense for rendering applications. A networked cluster is significantly cheaper for the same throughput.
I haven't had the slowdown problems you're describing. I don't see exactly the same times, but the difference is rather small.
.
.
OK, I reran the tests, and the difference is small as you say. For root_fs, a 105 Meg file, copied to a 500 Mhz laptop from a dual 1 Ghz server over 100 MHz ethernet, I get:
time scp deathray:bigfile
real 0m20.926s
time scp -c blowfish deathray:bigfile
real 0m16.672s
time wget http://deathray/bigfile
real 0m15.945s
If you're only offering files to a group of users who you can give passwords to, you could even use SCP. (Secure copy...uses sshd on the server side)
Scp is a horribly slow way to transfer a file, 5 times slower than http or ftp (which are nearly the same, for big files) on my network. Use scp if you need the security, or if you don't care about bandwidth sucking and just want the nice easy interface.
To get it to work I have to run a driver from the GATOS project that is from their CVS, patch it heavily so it supports things like XV, and run it on a bleeding edge copy of XFree just to get the semi-crudy TV out to work. It's the worst time I've ever had getting X to work correctly in years and years.
But remember how video codecs were like that a couple years ago, now they're faster, more reliable and better quality than on Windows, in spite of the intentional obstacles in the way.
ATI seems to think they support Linux, but they don't really do anything, as far as I can tell.
Yes, it smacks of evil. Well, continuous gentle pressure on ATI is the best medicine, and do like you did, buy from the competition. They'll get the message. The only wrong thing to do is stay silent.
In what way was my post a troll? ... (somebody abusing moderation for whatever reason)
Hey, I didn't Linux was a particularly good choice. But, you've answered your own question - a mobile phone doesn't need USB (though it would be handy for docking your phone and syncing contacts), IDE drivers, AGP support or anything like that. I'm willing to bet that this "Linux" kernel is little more than a scheduler and some memory management. However, Linux wouldn't be my first choice for a mobile phone OS.
You missed a big one: these days, a mobile phone needs a tcp/ip stack. And don't forget, this phone does video. Also, virtual memory (as in protection, not as in swapping) is something you really want in a device like this, as it's going to be running applications of varying quality. That's just off the top of my head.
The embedded device of today is a lot more like a computer and a lot less like a toaster.
Hey, I didn't Linux was a particularly good choice. But, you've answered your own question - a mobile phone doesn't need USB (though it would be handy for docking your phone and syncing contacts), IDE drivers, AGP support or anything like that. I'm willing to bet that this "Linux" kernel is little more than a scheduler and some memory management. However, Linux wouldn't be my first choice for a mobile phone OS.
You missed a big one: these days, a mobile phone needs a tcp/ip stack. And don't forget, this phone does video. Also, virtual memory (as in protection, not as in swapping) is something you really want in a device like this, as it's going to be running applications of varying quality. That's just off the top of my head.
The embedded device of today is a lot more like a computer and a lot less like a toaster.