Yikes... I have a hard time imagining that being very fast unless apps are coded to avoid thrashing memory pages.</strong><hr></blockquote>
I can confirm that CMS clustering is completely transparent to applications. You can (and I have) added a new VAX to a cluster while it is running, and processes automagically migrate onto the new machine until it's evenly loaded with the other machines.
DEC (OK, HP...) has extremely low latency, high bandwidth interconnects for clustering. Not cheap, mind you, but after using VMS clustering (and Digital UNIX clustering), even hearing Wolfpack described as clustering is nauseating. Way back when I worked at Delphi (the online service) which ran an amazing number of VAXen as a single cluster, and it was a beautiful environment from an operational perspective. Even now, nobody's even close.
myahmac & others, its not that I think clustering is a bad idea, but THIS clustering idea doesn't hold water. Yes, it would be cool if Apple (or someone else) were to create a method where I could take my existing machines (which is, in my case 2 500 MHz G3s, 1 700 MHz G3, and 1 450 MHz G4) and somehow end up with a massively parallel 2.3GHz 'puter -- but it won't. Ever. There's these persnickety things called physics, ya know...
Even if it was close, even if you could get *some* benefit, you won't get it using your existing machines. And then consider the hassle of maintaining a network where you're not just passing standard TCP/IP packets, but you're trying to get clustering work done? True 'nuff, this is something that we should get in the Xserve, but it's then a pro-level solution, that will cost you pro-level bucks, and is a DEDICATED CLUSTERING SOLUTION. This is not sharing emacs, or imacs, or whatever.
I think zero-latency, zero-maintenance peer-clustering would be great. I also think unlimited bandwidth would be swell. And I wish Elvis would finally come out of hiding and Walt Disney would defrost. Ain't gonna happen, folks. Not that it's a bad idea, just that it's fiction.
<strong>I can confirm that CMS clustering is completely transparent to applications. You can (and I have) added a new VAX to a cluster while it is running, and processes automagically migrate onto the new machine until it's evenly loaded with the other machines.</strong><hr></blockquote>
Does an entire process migrate, or can threads migrate individually? The former is much more efficient because then you have seperate address spaces and environments which migrate between macines. Thread-level has much bigger potential performance problems because thrashing could occur where multiple machines in the cluster fight over who owns the same byte of memory. It is thread-level computing that is needed to accelerate most of the computational algorithms since they work on the same data set. Either of these schemes is only going to work on a dedicated, high speed interconnect network.
<strong>ok I have had enough of this. Scott, john, you guys are so full of crap you don't even realize it. The reason i am so upset is because IF we had something like that at my school, we could have had an awesome video vs. a barely put together pos. And now i will explain.
I live in Clear Lake. Easily a high middle class area. The School District however is poor as **** .
At my HIGHSCHOOL there are around 5 maclabs with 20-30 G4's. one of the labs has dual 450's. We dont get te nice G4's we have an old 350. Thats right a 350 with 512 MB of RAM. With Block Scheduling etc. Do you honestly think i would need more computers if i hadaround 100 not doing anything. This number does not include the G4 or G3 that every teacher has hooked up to the network. I could max out the network before i even began to use all the proccessing pwer of the school, let alone the district because we are all hooked up via fibre optics. Now that you know the hardware ill show you the problem. My little team had to produce a video that had to be just as good as a professional's because the district could afford anymore for their little project. Shattered Dreams was a recreation of a drunk driving incident at school. We Film as being real and its supposed to make everyone cry etc. we had one day to shoot, and edit the enitre thing. The longgest part came not from editing, deciding how we wanted things to look, or even going out for pizza at 3 am. but waiting for our dinky little g4 to render transistions, layers etc. If We had "wolf" then life would have beedn so much easier. especially with all the thing we try. you say only pro's would want it and they can afford clusters. i am telling you from expierence that kids use and max computers everyday trying to learn this stuff. and we need something fast. I am no longer at Clear Lake but the kids there want to try blue screens and etc next year, and when i tried that at home it took forever to render. so trust me at public schools "wolf" would be a god send.
</strong><hr></blockquote>
Oh goody! Line um up so I can shoot um down. Check this... Alright so your little video took too long to render. You had 1 day to make this whole thing right? So during that time you would be MORE efficient running around booting up a school full of G4s? Speaking of which, didn't you say the school district was "poor as ****". Hum.. I must have grown up in the Ghetto. We had maybe 12 G4s in my school TOTAL. (counting the library too). Further that with how practicle this would be in similar situations. You could never orchastrate this on short notice so the value addition you are building saying it would have been sooo much easier doesn't work. In a bigger busienss you'd have to send around a memo to remind everyone to leave there machines on. (or hvae them all on times [which would take MORE time to setup]) Then folks would forget the memo and turn them off anyway. Then you have to go turn them all on.
I'm sorry you video was "a barely put together pos"....but don't blame it on lack of clusters.
So what is the big deal? We are all 'full of crap' because we believe that consumers will NOT see ANY benifit from this? And the professionals will NOT buy it?
<strong>myahmac & others, its not that I think clustering is a bad idea, but THIS clustering idea doesn't hold water. Yes, it would be cool if Apple (or someone else) were to create a method where I could take my existing machines (which is, in my case 2 500 MHz G3s, 1 700 MHz G3, and 1 450 MHz G4) and somehow end up with a massively parallel 2.3GHz 'puter -- but it won't. Ever. There's these persnickety things called physics, ya know...
Even if it was close, even if you could get *some* benefit, you won't get it using your existing machines. And then consider the hassle of maintaining a network where you're not just passing standard TCP/IP packets, but you're trying to get clustering work done? True 'nuff, this is something that we should get in the Xserve, but it's then a pro-level solution, that will cost you pro-level bucks, and is a DEDICATED CLUSTERING SOLUTION. This is not sharing emacs, or imacs, or whatever.
I think zero-latency, zero-maintenance peer-clustering would be great. I also think unlimited bandwidth would be swell. And I wish Elvis would finally come out of hiding and Walt Disney would defrost. Ain't gonna happen, folks. Not that it's a bad idea, just that it's fiction.</strong><hr></blockquote>
- You don't need to buy it, it comes with every copy of MacOSX.
- It communicates using standard IP protocols.
- Not leaving machines on just isn't a valid reason to not implement a technology like this.
I really don't understand the forceful rejection of this concept. Its not intrusive, it doesn't cost anything, and its advantage that the competion doesn't have (yet). Do you guys object to Apple providing a standard plug-in API? A standard driver framework? OpenGL? A standard printing interface for network printers? The latter two are particularly apt comparisons: would we have a variety of powerful 3D accelerators, or network printers to choose from if those system services hadn't be standardized? No. Now Apple might be doing the same thing for heavy-compute applications and you're all up in arms about what a terrible idea that is. Are you all on the Microsoft WolfPack project team, or what? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" />
<strong>What are you guys on about?!!</strong><hr></blockquote>
People claiming that this colaboration 'thing' will replace the cluster 'thing'. Using this new technology on an Xserver/Raid array is fine. Using this for personal use and expecting huge results and easy integration? Not in the cards.
PS, Programmer, you are on this board like 24/7. What line of work are you in where you're allowed such freedom? Sign me up.
- You don't need to buy it, it comes with every copy of MacOSX.
- It communicates using standard IP protocols.
- Not leaving machines on just isn't a valid reason to not implement a technology like this.
I really don't understand the forceful rejection of this concept. Its not intrusive, it doesn't cost anything, and its advantage that the competion doesn't have (yet). Do you guys object to Apple providing a standard plug-in API? A standard driver framework? OpenGL? A standard printing interface for network printers? The latter two are particularly apt comparisons: would we have a variety of powerful 3D accelerators, or network printers to choose from if those system services hadn't be standardized? No. Now Apple might be doing the same thing for heavy-compute applications and you're all up in arms about what a terrible idea that is. Are you all on the Microsoft WolfPack project team, or what? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" /> </strong><hr></blockquote>
Those that are agueing against the technical feasibility of the idea are presenting some good points, worth reflecting. Those that are arguing against the concept seem to be they most strenuosly arguing. I really can't understand why. If you don't like the idea, fine, you don't like the idea. But to slam the concept seems irrational. I suppose Apple shouldn't implement any solutions for customers, other than these people. the don't see the value of a particular concept, so it must be bad. No one should try and come up with a easy to setup, inexpensive clustering solution, because everyone has billions of dollars to throw at custom solutions.
Lighten up people. The idea is simply being hashed around. Coming up with techncal reason against the concept ok, arguing against the concept just because it doesn't help you seems strange...if it doesn't help you, fine...if it doesn't affect you then, why is it such a bad idea. This is not a bad idea, only the implementation may be shoddy, hence the discussion.
<strong>I will put forth a bet that he is a programmer </strong><hr></blockquote>
Gee, good guess. Long hours, long build times, and fasting typing makes for a large amount of AI posting. That and I want a new PowerMac soon and the lack of rumours is annoying.
Gee, good guess. Long hours, long build times, and fasting typing makes for a large amount of AI posting. That and I want a new PowerMac soon and the lack of rumours is annoying. </strong><hr></blockquote>
I'll second that (the last part). Hey, I also do most of my posting on AI while some processor-intensive plugin is running, too.
I'll second that (the last part). Hey, I also do most of my posting on AI while some processor-intensive plugin is running, too.</strong><hr></blockquote>
So if this clustering technology is true, does that mean you guys will go away?
I wonder if M$ came out with this first if all the whiners would then be b!tchin about why Appple didn't have clustering technology when it uses a U*ix base.
<strong>But they haven't seen it in a Wintel product, ergo it's a bad idea. I swear we're becoming an insular community with a serious inferiority complex.</strong><hr></blockquote>
Yeah, instead of Apple having the "Not invented here" complex, we've got a community with a "Not invented there" complex. <img src="graemlins/oyvey.gif" border="0" alt="[No]" />
People claiming that this colaboration 'thing' will replace the cluster 'thing'. Using this new technology on an Xserver/Raid array is fine. Using this for personal use and expecting huge results and easy integration? Not in the cards.
</strong><hr></blockquote>
And of course, connecting computers to a network to share resources is a bad idea isn't it? Go back to 1979. Please. <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
Apple is in the business of making computing easy. One of the reasons we haven't see this yet is because of the difficulty in getting wintel machines to do it. Apple has the advantage since they own both hardware/software. We haven't seen it from Apple yet because OSX is just now reaching the stage where it's possible.
Granted, it doesn't make sense for a research house to forgo a dedicated cluster in favor of spare cycles found on the admin's G4, but why is it such a bad idea to make those cycles available? I think there would be lots of folks (like me!) who would like to get a little 'oomph' without a lot of $.
Rather than a dedicated task that a cluster performs, this could be a generalized pool of clock cycles. Think about the penny jar you see next to a cash register. If you need one, take one, if you have one, leave one. Same with spare cycles. It could be a continually changing resource. If there are cycles available, they get used. I really keep thinking of gnutella when I look at this. The number of systems connected at anyone time is constantly changing.
Can I afford a cluster? no. Can I afford an i/eMac/PowerMac? yes.
We network computers to share resources. Bandwith, printers, burners, etc. Why not share CPU cycles while you're at it? Besides, what other product could make everyone's system faster, just be being placed on the network? Certainly wouldn't be a wintel machine.
Heh, if you think about it, Apple could very quietly make Sun's old slogan/philosophy true. "The network is the computer"
I'm sure I've been rambling. And I'm sure everyone will promptly blow holes in this, but...
Comments
The biggest problem with most clusters (dec/sun whatever)is that if a problem does occur then the entire cluster can hang.
Dobby
<strong>
Yikes... I have a hard time imagining that being very fast unless apps are coded to avoid thrashing memory pages.</strong><hr></blockquote>
I can confirm that CMS clustering is completely transparent to applications. You can (and I have) added a new VAX to a cluster while it is running, and processes automagically migrate onto the new machine until it's evenly loaded with the other machines.
DEC (OK, HP...) has extremely low latency, high bandwidth interconnects for clustering. Not cheap, mind you, but after using VMS clustering (and Digital UNIX clustering), even hearing Wolfpack described as clustering is nauseating. Way back when I worked at Delphi (the online service) which ran an amazing number of VAXen as a single cluster, and it was a beautiful environment from an operational perspective. Even now, nobody's even close.
Even if it was close, even if you could get *some* benefit, you won't get it using your existing machines. And then consider the hassle of maintaining a network where you're not just passing standard TCP/IP packets, but you're trying to get clustering work done? True 'nuff, this is something that we should get in the Xserve, but it's then a pro-level solution, that will cost you pro-level bucks, and is a DEDICATED CLUSTERING SOLUTION. This is not sharing emacs, or imacs, or whatever.
I think zero-latency, zero-maintenance peer-clustering would be great. I also think unlimited bandwidth would be swell. And I wish Elvis would finally come out of hiding and Walt Disney would defrost. Ain't gonna happen, folks. Not that it's a bad idea, just that it's fiction.
<strong>I can confirm that CMS clustering is completely transparent to applications. You can (and I have) added a new VAX to a cluster while it is running, and processes automagically migrate onto the new machine until it's evenly loaded with the other machines.</strong><hr></blockquote>
Does an entire process migrate, or can threads migrate individually? The former is much more efficient because then you have seperate address spaces and environments which migrate between macines. Thread-level has much bigger potential performance problems because thrashing could occur where multiple machines in the cluster fight over who owns the same byte of memory. It is thread-level computing that is needed to accelerate most of the computational algorithms since they work on the same data set. Either of these schemes is only going to work on a dedicated, high speed interconnect network.
<strong>ok I have had enough of this. Scott, john, you guys are so full of crap you don't even realize it. The reason i am so upset is because IF we had something like that at my school, we could have had an awesome video vs. a barely put together pos. And now i will explain.
I live in Clear Lake. Easily a high middle class area. The School District however is poor as **** .
At my HIGHSCHOOL there are around 5 maclabs with 20-30 G4's. one of the labs has dual 450's. We dont get te nice G4's we have an old 350. Thats right a 350 with 512 MB of RAM. With Block Scheduling etc. Do you honestly think i would need more computers if i hadaround 100 not doing anything. This number does not include the G4 or G3 that every teacher has hooked up to the network. I could max out the network before i even began to use all the proccessing pwer of the school, let alone the district because we are all hooked up via fibre optics. Now that you know the hardware ill show you the problem. My little team had to produce a video that had to be just as good as a professional's because the district could afford anymore for their little project. Shattered Dreams was a recreation of a drunk driving incident at school. We Film as being real and its supposed to make everyone cry etc. we had one day to shoot, and edit the enitre thing. The longgest part came not from editing, deciding how we wanted things to look, or even going out for pizza at 3 am. but waiting for our dinky little g4 to render transistions, layers etc. If We had "wolf" then life would have beedn so much easier. especially with all the thing we try. you say only pro's would want it and they can afford clusters. i am telling you from expierence that kids use and max computers everyday trying to learn this stuff. and we need something fast. I am no longer at Clear Lake but the kids there want to try blue screens and etc next year, and when i tried that at home it took forever to render. so trust me at public schools "wolf" would be a god send.
</strong><hr></blockquote>
Oh goody! Line um up so I can shoot um down. Check this... Alright so your little video took too long to render. You had 1 day to make this whole thing right? So during that time you would be MORE efficient running around booting up a school full of G4s? Speaking of which, didn't you say the school district was "poor as ****". Hum.. I must have grown up in the Ghetto. We had maybe 12 G4s in my school TOTAL. (counting the library too). Further that with how practicle this would be in similar situations. You could never orchastrate this on short notice so the value addition you are building saying it would have been sooo much easier doesn't work. In a bigger busienss you'd have to send around a memo to remind everyone to leave there machines on. (or hvae them all on times [which would take MORE time to setup]) Then folks would forget the memo and turn them off anyway. Then you have to go turn them all on.
I'm sorry you video was "a barely put together pos"....but don't blame it on lack of clusters.
So what is the big deal? We are all 'full of crap' because we believe that consumers will NOT see ANY benifit from this? And the professionals will NOT buy it?
No need to get spiteful just because your wrong.
<strong>FWIW, I'm on your side, Brisby.</strong><hr></blockquote>
Thanks! (some folks just don't seem to get it do they?)
<strong>myahmac & others, its not that I think clustering is a bad idea, but THIS clustering idea doesn't hold water. Yes, it would be cool if Apple (or someone else) were to create a method where I could take my existing machines (which is, in my case 2 500 MHz G3s, 1 700 MHz G3, and 1 450 MHz G4) and somehow end up with a massively parallel 2.3GHz 'puter -- but it won't. Ever. There's these persnickety things called physics, ya know...
Even if it was close, even if you could get *some* benefit, you won't get it using your existing machines. And then consider the hassle of maintaining a network where you're not just passing standard TCP/IP packets, but you're trying to get clustering work done? True 'nuff, this is something that we should get in the Xserve, but it's then a pro-level solution, that will cost you pro-level bucks, and is a DEDICATED CLUSTERING SOLUTION. This is not sharing emacs, or imacs, or whatever.
I think zero-latency, zero-maintenance peer-clustering would be great. I also think unlimited bandwidth would be swell. And I wish Elvis would finally come out of hiding and Walt Disney would defrost. Ain't gonna happen, folks. Not that it's a bad idea, just that it's fiction.</strong><hr></blockquote>
Elegantly put. I agree.
- You don't need to buy it, it comes with every copy of MacOSX.
- It communicates using standard IP protocols.
- Not leaving machines on just isn't a valid reason to not implement a technology like this.
I really don't understand the forceful rejection of this concept. Its not intrusive, it doesn't cost anything, and its advantage that the competion doesn't have (yet). Do you guys object to Apple providing a standard plug-in API? A standard driver framework? OpenGL? A standard printing interface for network printers? The latter two are particularly apt comparisons: would we have a variety of powerful 3D accelerators, or network printers to choose from if those system services hadn't be standardized? No. Now Apple might be doing the same thing for heavy-compute applications and you're all up in arms about what a terrible idea that is. Are you all on the Microsoft WolfPack project team, or what? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" />
<strong>What are you guys on about?!!</strong><hr></blockquote>
People claiming that this colaboration 'thing' will replace the cluster 'thing'. Using this new technology on an Xserver/Raid array is fine. Using this for personal use and expecting huge results and easy integration? Not in the cards.
PS, Programmer, you are on this board like 24/7. What line of work are you in where you're allowed such freedom? Sign me up.
<strong>What are you guys on about?!!
- You don't need to buy it, it comes with every copy of MacOSX.
- It communicates using standard IP protocols.
- Not leaving machines on just isn't a valid reason to not implement a technology like this.
I really don't understand the forceful rejection of this concept. Its not intrusive, it doesn't cost anything, and its advantage that the competion doesn't have (yet). Do you guys object to Apple providing a standard plug-in API? A standard driver framework? OpenGL? A standard printing interface for network printers? The latter two are particularly apt comparisons: would we have a variety of powerful 3D accelerators, or network printers to choose from if those system services hadn't be standardized? No. Now Apple might be doing the same thing for heavy-compute applications and you're all up in arms about what a terrible idea that is. Are you all on the Microsoft WolfPack project team, or what? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" /> </strong><hr></blockquote>
Those that are agueing against the technical feasibility of the idea are presenting some good points, worth reflecting. Those that are arguing against the concept seem to be they most strenuosly arguing. I really can't understand why. If you don't like the idea, fine, you don't like the idea. But to slam the concept seems irrational. I suppose Apple shouldn't implement any solutions for customers, other than these people. the don't see the value of a particular concept, so it must be bad. No one should try and come up with a easy to setup, inexpensive clustering solution, because everyone has billions of dollars to throw at custom solutions.
Lighten up people. The idea is simply being hashed around. Coming up with techncal reason against the concept ok, arguing against the concept just because it doesn't help you seems strange...if it doesn't help you, fine...if it doesn't affect you then, why is it such a bad idea. This is not a bad idea, only the implementation may be shoddy, hence the discussion.
<strong>
PS, Programmer, you are on this board like 24/7. What line of work are you in where you're allowed such freedom? Sign me up.</strong><hr></blockquote>
I will put forth a bet that he is a programmer
<strong>I will put forth a bet that he is a programmer
Gee, good guess. Long hours, long build times, and fasting typing makes for a large amount of AI posting. That and I want a new PowerMac soon and the lack of rumours is annoying.
<strong>
Gee, good guess. Long hours, long build times, and fasting typing makes for a large amount of AI posting. That and I want a new PowerMac soon and the lack of rumours is annoying.
I'll second that (the last part). Hey, I also do most of my posting on AI while some processor-intensive plugin is running, too.
<strong>
I'll second that (the last part). Hey, I also do most of my posting on AI while some processor-intensive plugin is running, too.</strong><hr></blockquote>
So if this clustering technology is true, does that mean you guys will go away?
MSKR
<strong>But they haven't seen it in a Wintel product, ergo it's a bad idea. I swear we're becoming an insular community with a serious inferiority complex.</strong><hr></blockquote>
Yeah, instead of Apple having the "Not invented here" complex, we've got a community with a "Not invented there" complex. <img src="graemlins/oyvey.gif" border="0" alt="[No]" />
<strong>
People claiming that this colaboration 'thing' will replace the cluster 'thing'. Using this new technology on an Xserver/Raid array is fine. Using this for personal use and expecting huge results and easy integration? Not in the cards.
</strong><hr></blockquote>
And of course, connecting computers to a network to share resources is a bad idea isn't it? Go back to 1979. Please. <img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
Apple is in the business of making computing easy. One of the reasons we haven't see this yet is because of the difficulty in getting wintel machines to do it. Apple has the advantage since they own both hardware/software. We haven't seen it from Apple yet because OSX is just now reaching the stage where it's possible.
Granted, it doesn't make sense for a research house to forgo a dedicated cluster in favor of spare cycles found on the admin's G4, but why is it such a bad idea to make those cycles available? I think there would be lots of folks (like me!) who would like to get a little 'oomph' without a lot of $.
Rather than a dedicated task that a cluster performs, this could be a generalized pool of clock cycles. Think about the penny jar you see next to a cash register. If you need one, take one, if you have one, leave one. Same with spare cycles. It could be a continually changing resource. If there are cycles available, they get used. I really keep thinking of gnutella when I look at this. The number of systems connected at anyone time is constantly changing.
Can I afford a cluster? no. Can I afford an i/eMac/PowerMac? yes.
We network computers to share resources. Bandwith, printers, burners, etc. Why not share CPU cycles while you're at it? Besides, what other product could make everyone's system faster, just be being placed on the network? Certainly wouldn't be a wintel machine.
Heh, if you think about it, Apple could very quietly make Sun's old slogan/philosophy true. "The network is the computer"
I'm sure I've been rambling. And I'm sure everyone will promptly blow holes in this, but...
<img src="graemlins/smokin.gif" border="0" alt="[Chilling]" />
[ 06-25-2002: Message edited by: tsukurite ]
[ 06-25-2002: Message edited by: tsukurite ]</p>
<strong>
tsukurite
----------
If we get the G5, what will we talk about?
]</strong><hr></blockquote>
The G6?
<strong>
The G6?</strong><hr></blockquote>
Or else b1tch about it's performace. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> <img src="graemlins/oyvey.gif" border="0" alt="[No]" />