[GigaWatt]

OK, here is my problem, i have 2 older rigs at home, a 788MHz Celeron with 384MB of RAM and a 500MHz Celeron with 192 MB of RAM. i would like one of them to be the primer (master) machine and the other one to be some sort of a slave to the first one (preferably, 788MHz Celeron could be the master and the 500MHz one could be the slave, sort of speaking ). both PCs will be connected via Etherent and will share resources with each other (CPU and RAM). i don't plan on using the 500MHz rig, it will probably only have the basic components (probably only and older GFX card, the rig boots up without any keyboard attached to it, don't ask me how, hahahah ), so it will only be plugged to the power cord and the Etherent card of the 788MHz one. there will be no router or any other kind of data switching hardware between the 2 PCs. the LAN cards are both 100MBPS. both rigs have Win Server 2003 EE SP2 running on them (only because it's faster than XP)

now, my question is, is there any kind of software, or some kind of a service (or a plugin) that can enable workload (CPU or RAM) to be shared (i guess the more apropriate term would be "transfered" from the 788MHz one to the 500MHz one) over LAN?

i've done some searching on the web, and (as far as i was able to understand) this thing is called cluster computing, and it's only available for universities or large business. the idea is to have a large number of idle computers that can be used to process workload of one single CPU demanding task via an Ethernet network. one large cluster computing project is (for example) SETI@home.

now, if i understood correctly, these CPU demanding tasks (looking for UFOs is not an easy job ) are not Real-Time, so that means that the processing is done pretty much like rendering is done on some 3D application. you have one table (database) of equations with variables (putting it in simpler terms) that need to be solved and another table (database) that grabs the results of the solved equations. the computer farm (a group of networked computers using such software is called a computer farm) takes a portion of each table, processes it and returns it back to the main PC (the PC that demanded the processing). the other rigs share their unused CPU cycles with the main one and the processing is done much quicker than if it was done by only one rig.

but the data that the computer farms is processing is not Real-Time demanding. so the question is, can Real-Time processing be done via LAN and if it can be, is there some kind of a software that does this?

[Cow_160483]

Thats a very good question bro coz i also have my old rig just lying around and would like to do something similar to it and my main rig i.e hook the two buggers together, so if there is anyone who knows more on the matter that would be totally sweet . Thanks

[GigaWatt]

well, from what i have read on the net, this kind of thing is still in development and (as i wrote in my previous post) is still only available to larger companies and universities... bummer...

there are some multiplatfrorm pieces of software available, but most of them are written in Java, and (as we all know) Java applications can be pretty slow (the main problem is that when you make a multiplatform programming language, nothing is going to go smooth). the whole point is to have a small, low CPU demanding application that can make advantage of the CPU power of, lets say, a small office or a home network, something that can coordinate (for example) up to 10 rigs. now that would be an application worth paying for, especially if it saves me a lot money from buying a new rig .

[Vikingod]

Its a ways off. I remember reading that AMD is working intensively on this sort of stuff (potentially as a future gaming platform), but it's a long ways away.

[Matth]

The problem with realtime / near realtime clustering is the communication overhead.

For those interested in cluster experimentation, the most commonly available is a Linux beowulf cluster which can be assembled from a collection of surplus machines, or built as a few high power nodes using dedicated gigabit interconnects

[GigaWatt]

Originally Posted by Matth

The problem with realtime / near realtime clustering is the communication overhead.

For those interested in cluster experimentation, the most commonly available is a Linux beowulf cluster which can be assembled from a collection of surplus machines, or built as a few high power nodes using dedicated gigabit interconnects

i figured that that could be the problem, that is why i was asking if realtime processing over a 100MBPS LAN connection can be achieved, even if the LAN connection was dedicated.

i don't think that a small time lag could be a problem, as long as the time lag is not larger than the time needed for the primer rig to do the processing itself...

anyway, if anybody has some more info about clustering and load sharing under Windows, please write...

[GigaWatt]

update: there is a service under Windows Server 2003 EE called MSCS (Microsoft Cluster Service) and the service is controlled by an application (filed under Administrative Tools) called Cluster Administrator. there is a guide on the official Microsoft page about creating and configuring a server cluster under Windows Server 2003.

the guide is simple and to the point, but in the introduction part, there are a couple of things that i can't quite figure out. here are some quotes from the guide.

Server clusters allow client access to applications and resources in the event of failures and planned outages. If one of the servers in the cluster is unavailable because of a failure or maintenance requirements, resources and applications move to other available cluster nodes.

Server clusters do not guarantee non-stop operation, but they do provide sufficient availability for most mission-critical applications. The cluster service can monitor applications and resources and automatically recognize and recover from many failure conditions. This provides flexibility in managing the workload within a cluster. It also improves overall system availability.

does this mean that nodes in the service cluster can exchange workload between each other only if one of the nodes in the cluster fails, or they exchange workload between each other all the time? cuz if it's the first scenario the only scenario in a Windows service cluster, than this service is no good in transfering CPU load (workload) from one PC to another... :S