Huh. This software would be the most important thing that came out of Apple if it was integrated into the OS at a core level. 10.4 maybe?
However, what could this be used as far as NOW is concerned? I'm the IT guy for a video production company. Could this be used to offload projects that require heavy processing?
What is coming? (By the amazingly poor administration features there's lots of stuff missing) You can't do anything.
I assume Final or last beta will be available for WWDC so the Devs can work and get seminars. Then, come September, we'll be swimming in 10.4 Xgrid goodness.
As of yet, I have been unable to read any of the archives for the Xgrid list or re-sign up.
You can't even run a controller without being logged in. Which is useless when you have user switching and security concerns. I did read the docs. I just need information, and I'm just venting at the lack of it. I'm getting into the Sample Plug-In doc tomorrow.
It is admittedly a Preview product, but it would be nice to be able to fine tune the system. Like all the Docs say that a properly configured cluster of M machines will get the power of the combined Mhz of M machines. I'm Rendezvous is nice, but I can't imagine that it optimizes anything. So how do you optimize it?
One thing I did note while testing, was that the machine I was using as an Agent, with the screen saver activated, would consistently not exist. Since the graphic updated every 10 sec or something it would sometimes be at 0 for extended periods of time, while on my computer right beside it, the needle would continually jump from 1.25 to 2.25. I guess that's something to submit to Apple about.
To HOM, if you look in the Resources for Xgrid, there are additional "10x", "100x", "1000x" graphics as multipliers for the shown speed. I cannot imagine. So Big Mac would be at 2200 * 2 = 4.4 1000x.
You can't even run a controller without being logged in. Which is useless when you have user switching and security concerns. I did read the docs. I just need information, and I'm just venting at the lack of it. I'm getting into the Sample Plug-In doc tomorrow.
Are you *sure* about that? I have two towers sitting in the lab, running their own controllers and agents, and popping up Xgrid here in my office shows all three as possible controllers to connect to... and those two lab towers are showing the login panel.
Quote:
It is admittedly a Preview product, but it would be nice to be able to fine tune the system. Like all the Docs say that a properly configured cluster of M machines will get the power of the combined Mhz of M machines. I'm Rendezvous is nice, but I can't imagine that it optimizes anything. So how do you optimize it?
Er... unsure how you would want to optimize it beyond what Xgrid already does, finding services, figuring compute power, load levels, etc, and then dispatching jobs appropriately. What did you have in mind? Rendevouz only finds the machines in the first place, and has nothing to do with it after that.
Quote:
One thing I did note while testing, was that the machine I was using as an Agent, with the screen saver activated, would consistently not exist. Since the graphic updated every 10 sec or something it would sometimes be at 0 for extended periods of time, while on my computer right beside it, the needle would continually jump from 1.25 to 2.25. I guess that's something to submit to Apple about.
I gave up on getting anything meaningful from the gauge. On my three-node system, the tach consistently was barely bobbing above 0 since it was sending out jobs as fast as it was getting results. The smoooooooooth needle movement was actually too slow. Digital readout might be more useful for burst-style batch jobs. The tach may be great for massive ongoing computations though. Definitely give feedback however.
I was able to get Xgrid installed on the other compatible system, though I missed something with the password so it was show as offline until I could get back to it. As I remember, there was no one logged into the Controller computer, so maybe it just wasn't happy before, cuz I haven't changed anything on the system.
The plug-in architecture is simple, which makes it great and for command line scientific tools, it's more than enough.
There are other applications, that would be better suited for a naming scheme different than what is specified "command_args...."
Actually, how would this work outside of a plug-in, let's say, in Photoshop for a large render. Or even better, Final Cut Pro/Express or Compressor.
Just curious...how easy is this to implement into programs? I mean is it as easy as tying in rendezvous? It sure would be nice next when when I'm doing rendering or working on big apps if I could have my powebook connected to my G5 and get that extra processing power easily.
Ok well I'm sure wireless isn't the best thing to do it over, but the 733Mhz QS I was running the controller on, it's tachometer didn't go above the 1.25Ghz mark my machine runs at.
One thing to keep in mind - that tach is a smooth animation, so it takes time to 'get up to speed' then back down when jobs go out, and then get completed. Mine was bouncing around 1.4GHz on a 1.25GHz machine, but the speed was almost 5x faster. On the *right* task, it pegged at 5GHz easily. The problem was that the tach would get the 'go' signal, then get the 'done' signal so quickly that it couldn't get the needle up in time. :P Watch the clock, not the tach.
So is everyone in agreement that this could eventually find it's way into the system level of the OS, and therefore utilize multiple machines for any and all processor intensive tasks? Think about the possibilities, and think about the longevity this could give to old machines that are considered too slow.
'any and all'? Nope. That's what a couple of us have been trying to explain here.
Imagine you have a system of n gravitational bodies to solve. Each body interacts with all the other bodies. The calculations aren't involved, and pretty simple to conduct on each body once you have all the data.
Good candidate for Xgrid, right?
Wrong.
The calculation is quick and easy, but since all entities need to know about all the other entities, you have to transmit each result to *all* the other machines on *every* step.
Your quick calculation is now waiting on the network. D'oh. Very slow.
And that's for a relatively modest amount of data being transmitted. Imagine the problems with those that need a *lot* of data being passed around.
No, this is a technology for specific situations, but in those situations, it rocks.
Comments
However, what could this be used as far as NOW is concerned? I'm the IT guy for a video production company. Could this be used to offload projects that require heavy processing?
My.
God.
G4 PowerBook 1.2GHz, 512MB RAM. Nothing to sneeze at. First stage of Mandelbrot demo in local mode (local CPU only): 43.074 sec.
Add on a Dual 1.4GHz G4 and a Dual 2.0GHz G5, 1GB RAM each.
9.9533 sec.
Holy.
Crap.
~4.4x faster.
I'm in geek heaven.
Originally posted by Kickaha
Holy.
Crap.
~4.4x faster.
I'm in geek heaven.
The computer Chess and GO geeks are going to be all over this. Just you wait and see.
Originally posted by jginsbu
The computer Chess and GO geeks are going to be all over this. Just you wait and see.
Yep, you could put each legal move on a node. You'd prolly see linear scaling with this. Good call.
Like:
How do you write the plug-ins?
What is coming? (By the amazingly poor administration features there's lots of stuff missing) You can't do anything.
I assume Final or last beta will be available for WWDC so the Devs can work and get seminars. Then, come September, we'll be swimming in 10.4 Xgrid goodness.
As of yet, I have been unable to read any of the archives for the Xgrid list or re-sign up.
Originally posted by macserverX
Does anyone have more info?
Like:
How do you write the plug-ins?
What is coming? (By the amazingly poor administration features there's lots of stuff missing) You can't do anything.
Huh? What do you mean by that?
Read the documentation that comes with Xgrid, as well as the sample plug-in Xcode project. It's all in there.
It is admittedly a Preview product, but it would be nice to be able to fine tune the system. Like all the Docs say that a properly configured cluster of M machines will get the power of the combined Mhz of M machines. I'm Rendezvous is nice, but I can't imagine that it optimizes anything. So how do you optimize it?
One thing I did note while testing, was that the machine I was using as an Agent, with the screen saver activated, would consistently not exist. Since the graphic updated every 10 sec or something it would sometimes be at 0 for extended periods of time, while on my computer right beside it, the needle would continually jump from 1.25 to 2.25. I guess that's something to submit to Apple about.
To HOM, if you look in the Resources for Xgrid, there are additional "10x", "100x", "1000x" graphics as multipliers for the shown speed. I cannot imagine. So Big Mac would be at 2200 * 2 = 4.4 1000x.
Originally posted by macserverX
You can't even run a controller without being logged in. Which is useless when you have user switching and security concerns. I did read the docs. I just need information, and I'm just venting at the lack of it. I'm getting into the Sample Plug-In doc tomorrow.
Are you *sure* about that? I have two towers sitting in the lab, running their own controllers and agents, and popping up Xgrid here in my office shows all three as possible controllers to connect to... and those two lab towers are showing the login panel.
It is admittedly a Preview product, but it would be nice to be able to fine tune the system. Like all the Docs say that a properly configured cluster of M machines will get the power of the combined Mhz of M machines. I'm Rendezvous is nice, but I can't imagine that it optimizes anything. So how do you optimize it?
Er... unsure how you would want to optimize it beyond what Xgrid already does, finding services, figuring compute power, load levels, etc, and then dispatching jobs appropriately. What did you have in mind? Rendevouz only finds the machines in the first place, and has nothing to do with it after that.
One thing I did note while testing, was that the machine I was using as an Agent, with the screen saver activated, would consistently not exist. Since the graphic updated every 10 sec or something it would sometimes be at 0 for extended periods of time, while on my computer right beside it, the needle would continually jump from 1.25 to 2.25. I guess that's something to submit to Apple about.
I gave up on getting anything meaningful from the gauge. On my three-node system, the tach consistently was barely bobbing above 0 since it was sending out jobs as fast as it was getting results. The smoooooooooth needle movement was actually too slow. Digital readout might be more useful for burst-style batch jobs. The tach may be great for massive ongoing computations though. Definitely give feedback however.
I was able to get Xgrid installed on the other compatible system, though I missed something with the password so it was show as offline until I could get back to it. As I remember, there was no one logged into the Controller computer, so maybe it just wasn't happy before, cuz I haven't changed anything on the system.
The plug-in architecture is simple, which makes it great and for command line scientific tools, it's more than enough.
There are other applications, that would be better suited for a naming scheme different than what is specified "command_args...."
Actually, how would this work outside of a plug-in, let's say, in Photoshop for a large render. Or even better, Final Cut Pro/Express or Compressor.
Originally posted by Kickaha
Check the dev kit. There's a sample plugin that tells Xgrid how to hook into your app. Not too difficult looking, really.
This == Extent of my Cocoa Knowledge
But I'll download and try to check it out.
Originally posted by ast3r3x
This == Extent of my Cocoa Knowledge
Hey, then you know C too! (and PERL, and...)
Originally posted by blue2kdave
Hey, then you know C too! (and PERL, and...)
haha c++ and PHP
Imagine you have a system of n gravitational bodies to solve. Each body interacts with all the other bodies. The calculations aren't involved, and pretty simple to conduct on each body once you have all the data.
Good candidate for Xgrid, right?
Wrong.
The calculation is quick and easy, but since all entities need to know about all the other entities, you have to transmit each result to *all* the other machines on *every* step.
Your quick calculation is now waiting on the network. D'oh. Very slow.
And that's for a relatively modest amount of data being transmitted. Imagine the problems with those that need a *lot* of data being passed around.
No, this is a technology for specific situations, but in those situations, it rocks.
(Does PS support it yet?)