Grid Computing...

[patri26281]

Hi all,
Right now, nearing the end of my college life, I'm searching for a good thesis project. I was just browsing the various bookmarks that I have made over the years and one of the book marks lead me to the operating system Inferno's website. With all the current talk of grid computing and distributed computing going around I thought this would make an interesting project.
Inferno was designed to run on multiple platforms, on it's own or in another OS, and most importantly it's small. However, it's designed to run Limbo apps, Limo being a language that is interpreted on the fily much like Java and C#. Here is my idea...
First, modify Inferno into a more usable OS, i.e. better GUI, more drivers, and etc.
Second, modify Inferno in such a way that instead of running Limbo and using the DIS virtual machine, make it run a Java JIT compiler, and the C# interpreter.

The way I see it is that, one would install Inferno on servers, any server would be fine since Inferno would be able to run on it. Thus this would turn processing power into a commodity, and making it possible to sell the CPU cycles.
Here's a new use for an old idea, NCs. That's right, Larry Ellison's network computer with a new twist. A user would install Inferno on his/her local machine. The the user would pay a subscription fee for processing power. Say $10/month = 1GHz worth processing power, $20/ month for incremental more processing power. Although the user would have local storage and processing power, when the extra power is needed the OS would automatically offload the extra load to the server; i.e. if the user is watching a DVD and doing some movie editing/ or image editing, the OS would intelligently off load the movie/image app to the server and leave the DVD on the client side for maximum performance. With the big push from MS, Sun, and the computing industry to .Net, and Java platforms the OS would be able to run most of the next generation apps, thanks to .Net's C#'s platform agnostic nature, and Java's write once, run anywhere promise.

So, would it work? Would the business model work? What would be the problems? I have figured that bandwidth is probably the #1 problem, security will be one, and any others?

 

[lifeguard1999]

Latency. It kills your idea. It is quicker for the local CPU to process the data then some computer half way across the country.

Cheap, fast CPUs. It kills your idea. You can thank AMD for that one.

Bandwidth. It kills your idea. Not everyone has broadband. Those that do, do not need your idea.

Then there is authentication (who are you?) and authorization (can you run here?) Throw in privacy concerns, hackers, .... and well, you get the idea.

Sorry to be a sour puss and spoil your idea. It is not personal. It is just that the outfit I am with finds that authentication (who are you?) and authorization (can you run here?) are HUGE factors in grid computing.

 

[rbhawcroft]

if you have a single mode fibre telco link to the other chip then fine, but a 2ghz chip has upto 4Gbps of bandwidth to crunch, so you may need say a Gigabit ethernet connection for each chip to draw up and return data. The best commercial applications would be using sales force computers during nighttime or even day time, or distributed comp processing in universities.

 

[rbhawcroft]

no view then?

to go on one use could be in server farms that produce large amounts on data quickly for example modeling explosions or so on and you could use GEth to farm the results out in terms of data bandwidth/ RAM costs and HDD costs, and for processing say within a few minutes and that i guess would reduce the size of server farm down. also for seti type applications or chip testing or whatever and benchmarking harware and software you could basically use this as a Gb LAN with added software. your investigation would be almost a business plan then if you could see something that made economic and tech sence.

 

[Shalmanese]

Its been considered, AFAIK, nobody has been stupid enough to try to implement it yet.