Supercomputer?

[lilwic]

So I was wondering...... Supercomputers like IBM's Watson consist of thousands of servers linked together to make one supercomputer. I have tons of random computers around my house which I don't use and I was wondering if I could link them like Watson and make one "supercomputer". (I did the parentheses because they wouldn't have nearly as much power as something like Watson)

[AshleyAshes]

It's called a cluster computer, and yes, that would be it, but it'd need highly specialized operating system setups and applications to even be useful.

[Sakkura]

You could do a mini-version of distributed computing. The connections between your different PCs wouldn't be as fast or direct as in a traditional supercomputer, but for massively parallel problems that doesn't matter.

[pagani123]

no you would not bee able take advantage of a combined power in a cluster system. Plus what would you use it for anyways supercomputers are used for highly specialized tasks 

[lilwic]

Just now, Sakkura said:

You could do a mini-version of distributed computing. The connections between your different PCs wouldn't be as fast or direct as in a traditional supercomputer, but for massively parallel problems that doesn't matter.

 

2 minutes ago, AshleyAshes said:

It's called a cluster computer, and yes, that would be it, but it'd need highly specialized operating system setups and applications to even be useful.

I was just thinking.... give the useless computers a use. So what would it require exactly? Or would that be too much for a forum post?

[lilwic]

1 minute ago, Maybach123 said:

no you would not bee able take advantage of a combined power in a cluster system. Plus what would you use it for anyways supercomputers are used for highly specialized tasks 

Uhhh...... idk kinda just a little project.

[moderategamer]

 

 

[Sakkura]

2 minutes ago, lilwic said:

 

I was just thinking.... give the useless computers a use. So what would it require exactly? Or would that be too much for a forum post?

It would require a relevant task, as well as the necessary know-how and software to set it up. Not a trivial task, at all.

 

It's probably more realistic to join up with an existing distributed computing project like Folding@Home or BOINC.

[RedWulf]

It's a great learning project, but to get any use out of it besides just random data sorting you'd need alot of nodes or some very high end ones. My tech club is finishing up one made out of 8 old dell optiplex 780s. Its complicated to an extent but to learn its good, our school is thinking of using it for analyzing test results that would strain a normal computer. But in reality anything you could make without shelling out thousands would just be for tinkering or sorting a shit ton of metadata. 

The Pi clusters are just because they're cheap and everyone and their brother is making them although most of them are more like distributed computing or grid than cluster or parallel from the tutorials I'vee seen on them.  

Another problem is it works better if the specs are the same, while I'm sure theres some way to adapt the various cluster database programs and normal parallel studios to account for uneven power, most divide the tasks up evenly meaning some nodes would finish quicker than others giving a great deal of down time.

I have to agree with the others though, the resources are better in a pool for research, but if you want to mess around with your own, message me and I'll send you some information on how we got started on our "beowulf" cluster. But again, even after set up it requires special programming to run which you'd need to learn.

[RedWulf]

59 minutes ago, AshleyAshes said:

It's called a cluster computer, and yes, that would be it, but it'd need highly specialized operating system setups and applications to even be useful.

Actually many just run on the server versions of Linux or even Windows, although some like Watson have their own flavor. As for the applications, yes, it requires special databases and some other reasource to control the nodes respectively by system. 

59 minutes ago, Sakkura said:

You could do a mini-version of distributed computing. The connections between your different PCs wouldn't be as fast or direct as in a traditional supercomputer, but for massively parallel problems that doesn't matter.

Distributed computing is far different from parallel or clusters, one the task is sent out mindlessly and is processed and received back as if its just null data. Parallel not only involves a shared network but also shared files, databases, often users, environments have to be edited, and the nodes are controlled entirely by the master(s) not just given a task. 

1 hour ago, Maybach123 said:

no you would not bee able take advantage of a combined power in a cluster system. Plus what would you use it for anyways supercomputers are used for highly specialized tasks 

highly specialized? they're used in everything from Disney movies as render farms to playing chess to solving massive equations/hashes to sorting millions of songs/data points to mechanical analysis. Just because they can't play gta v at 150fps doesn't mean the uses are restricted, if he goes to college for stats, digital art, computer science, or engineering he'll have a decent tool. Really with either a render farm or data sorting designed cluster the uses are vast, while the chess AI and the CAD analysis systems are admittedly limited.  

 

[AshleyAshes]

Just now, RedWulf said:

highly specialized? they're used in everything from Disney movies as render farms to playing chess to solving massive equations/hashes to sorting millions of songs/data points to mechanical analysis.

I don't understand why you said 'Highly Specialized?' with a question mark and then listed a series of highly specialized applications.

[RedWulf]

Just now, AshleyAshes said:

I don't understand why you said 'Highly Specialized?' with a question mark and then listed a series of highly specialized applications.

Data sorting nor rendering are highly specialized, hundreds of thousands of people and groups use them every day and many are interchangeable. The specializations come merely from the needed programming. I listed some specific examples, ask anyone whole dables in 3d modeling or video rendering or statistics if they wouldn't love a faster means of completing the related computations.    

[RedWulf]

6 minutes ago, AshleyAshes said:

I don't understand why you said 'Highly Specialized?' with a question mark and then listed a series of highly specialized applications.

I'll clarify, the previous person that I quoted tried to suggest that the cluster would be limited in uses by subjecting all clusters to the generalization that they are only used for a certain job. When in reality thats only because the people that own them only have that one need, clusters have an insane amount of applications as long as the user knows how to code it. I trying to show the broad range of clusters and the diversity that a typical number cruncher can be used for, unfortunately I don't know of a specific cluster thats labeled as general purpose or I would have just referenced it.  

[AshleyAshes]

1 minute ago, RedWulf said:

I'll clarify, the previous person that I quoted tried to suggest that the cluster would be limited in uses by subjecting all clusters to the generalization that they are only used for a certain job. When in reality thats only because the people that own them only have that one need, clusters have an insane amount of applications as long as the user knows how to code it. I trying to show the broad range of clusters and the diversity that a typical number cruncher can be used for, unfortunately I don't know of a specific cluster thats labeled as general purpose or I would have just referenced it.  

You're still describing specific tasks as opposed to general purpose computing...

[RedWulf]

Just now, AshleyAshes said:

You're still describing specific tasks as opposed to general purpose computing...

What do you consider general purpose, games and facebook? sorting data, solving mathematical problems, crunching numbers, and doing it all faster and more precise than a human is what computers were designed for and are used for outside of the home. With a single cluster, by only changing the data base and the algorithm, he could go from predicting hurricanes to sorting iTunes to measuring voter biases on every major topic.  

[HelplmChoking]

Sure you can do it. I connect several computers together sometimes to work on distributed rendering in 3dsMax.

[Sakkura]

8 hours ago, RedWulf said:

Distributed computing is far different from parallel or clusters, one the task is sent out mindlessly and is processed and received back as if its just null data. Parallel not only involves a shared network but also shared files, databases, often users, environments have to be edited, and the nodes are controlled entirely by the master(s) not just given a task.

I don't think cluster computing is relevant here as it generally requires each node to have identical capabilities. Distributed computing is more applicable.

[RedWulf]

4 hours ago, Sakkura said:

I don't think cluster computing is relevant here as it generally requires each node to have identical capabilities. Distributed computing is more applicable.

You can manipulate the configuration to work onvarious nodes as long as they all have enough power to run the OS and programs, but like with RAIDing with different drives, the weakest node will be the limiting factor. Although I'm sure some advanced clusters can offset appropriately. I think for a project and learning a cluster offers more, since its only a few computers and older ones neither configuration will be very efficient. Unless he joins a pool, but that cost bandwidth and you don't really get to enjoy the results.