Vince said:
aaaaa00 said:
I don't want a pointer to a PR fluff piece. Give me a real world example of something that could not be implemented in .NET (just for example). Code would satisfy me.
What does the Globus toolkit actually offer me TODAY, over competing solutions that already exist and are widely deployed in the industry?
Who said it had to be utterly revolutionary? You're creating your own hype and then outlashing when it doesn't live up.
I recall speculation on this board that GRID was going to extend to millions of PS3s and be applied to distribute low latency game functions on the client side to multiple machines.
I believe myself and many others basically concluded it was going to be a server-side clustering solution.
Of course, I may be misremembering.
Vince said:
It's a technology, it's open source, it works - they will build on it as stated before. Whats the problem? Ohh, I know...
Just because a technology is open source does not instantly make it BETTER than what exists today.
The problem I have is not that Globus exists, it is with the attitude and automatic assumption that it is BETTER than what we already have, because 1. IBM is behind it, and 2. It is "open source".
Vince said:
And Vince, how come it is when someone questions your world-view, your first impulse is to call them an idiot, almost as if you take any criticism directed at a TOOLKIT, personally?
Why are you reacting with such vehemence? Why do you act as if you have a personal stake in this?
I hate stupidity, in our nice conversations I've stated this before.
So do I, but I don't call people "idiots" when I believe their thinking is flawed.
Vince said:
You're smarter than that, you're better than your first post - why did you make it?
I don't believe what I've said in this thread so far has been stupid.
Ok maybe I went a little overboard with my "hive mind" remark. ;-)
Vince said:
Not a magical distributed processing system to connect up millions of PS3s into one huge hive mind.
If it can do it server side, what prevents an extention of this on a distributed scale foir various non-time sensitive processing -
Even for simple tasks, reliably distributing general processing and storage tasks (even reasonably non-time critical ones) onto millions of potentially insecure, untrusted (mod chips), potentially unreliable nodes (yank power plug), joined by unreliable, latency-filled, and bandwidth-limited connections would be architecturally difficult.
People have been working for years to solve just the distributed storage part of the problem, without total success. (Freenet, Farsite, etc.)
Current solutions for Internet distributed processing work precisely because they are extremely focused and limited in scope, and they attack a very narrow subset of problems. (Things like protein folding, distributed.net, etc.)
Clustered servers are a much easier problem to attack, precisely because you control the hardware, the interconnect, and the software environment.
With control over these, a lot of problems simply go away or are minimized. You can guarantee the interconnect is fast. You can take steps to minimize the possibility that the network connection vanishes because someone yanked the network cable. You can guarantee no one yanks the power at a critical instant. You can guarantee no one accidentally (or maliciously) corrupts the data you just wrote to disk A on Node B. So on and so forth.
Vince said:
or what prevents it's use to moderate the sharing of computing and storage resources inside Cell, et al.
Inside a single machine composed of Cells, why bother with Globus at all?
Why invoke a location-server when you damn well know where that object is? Why encode everything in XML? Why pass around method calls using a horrible inefficient message based protocol? Spend time marshalling function arguments all over the place? Why?
It is like flying to the moon and back in order to get to the local supermarket. Flying to the moon might be useful if your final destination is Mars, but if you're just going down the street, why waste all that effort?
Within a single machine, Globus is a waste of time, especially on a console, where time is always at a premium. SOAP is a waste of time. DCOM is a waste of time. Distributed CORBA is a waste of time.
It's not like the Globus Toolkit actually helps you partition your problem, nor does it help you with synchronization, scheduling, or debugging (all the hard parts of writing a highly parallel app).
Vince said:
Instead, you take it to an extreme [eg. "huge hive mind"] and come across in the wrong way - if this is intentional I do not know.
I may be misremembering, but I do recall a discussion here at B3D where that very notion came up, and some people thought it would indeed be feasible for time-critical tasks to be broken up and farmed out to many PS3 clients.
Still, ok, I admit I may have gone a little overboard with the hive mind comment.
Peace?
