Linux cluster Vs 1 good CPU
 

hey guys maybe you can help me with this one.
I wonder what is better : 2-3 PCs(for a start) with a crappy 2Ghz Celeron (or any other cheap cpu) +GigaBit ethernet or a good new AMD 64 ???



Any information will be appreciated and i can't be more specific i guess because i don't know much

Most programs aren't optimized to run on multiple processors, let alone multiple machines.

On the other hand, the more boxes you have, the more processes you can have running that have more processor time... One box can be compiling, one box can be browsing, one box can be playing music, one box can be serving, etc.

On the other hand, the more boxes you have, the systems you have to maintain.

On the other hand, the more boxes you have, the less you are to have a catostrophic failure and not be able to do anything.

On the other hand, the more boxes you have, the more hardware you have to buy.

How many hands is that, so far?

If you want to go with the cluster, you should know that it doesn't usually work as if it is one big computer. As I mentioned first, most programs aren't written to run on multiple processors. But, the more processors you have, the more programs you can have running. A good thing to look into would be grid-computing. It's a cluster that does some automatic load balancing... Intensive jobs get passed to the computer with the best processor.

well then maybe i am confused with the meaning of cluster..
I thought it's like 3little cpus = 1 big CPU and that's how things work... meaning that if a dvd-rip proccess needs the 75% of the combined CPU power it gets it and remains the 25% of the CPU power.. not like 1 process uses 1 cpu...
i get a little confused here :confused: :confused: ..could anyone give me some basic info (simple stuff so that i have't got to dig through GooGle) ???

I hope this helps out a little. (Here I go)

Clusters only make apps run more reliable/faster if the app is cluster aware. For the most part Cluster run in an "active -passive" state meaning that one node processes the requests while the other node sits idle/dormant. Meaning that if the active node fails the backup node will kick in and you might only lose a few transactions. There are some “active - active” clusters but these are very highly specialized apps that are expensive and complicated to configure.

Grids share the computing load among multiple computers/nodes. Now this adds more complexity to the mix. These are VERY specialized apps. These for the most part are number crunchers and simulators that are usually custom wrote for a specific purpose. For example the Earth Similar by NEC in Japan or weather model generators by NSSL (National Severe Storms Laboratory). There are also some relatively less expensive and easier to configure (I use those term loosely) apps. An example I can think of is Oracle (which does grids and clusters)

So if just wanted to rip some MP3’s, while re-compiling your kernel, while rendering a movie there is no one stop magic app/component, I’ve ever seen or heard of that will do this. What I basically mean is, there nothing to take some generic app and spread it’s load across a grid or cluster and make it work.

If there is such a thing let me know and I’ll use my test 10g grid and test it. (I love working at large Companies!)

Sorry I'll stop rambling on now....


This link may also help shed some light on it for you.
http://lcic.org/

Your are right, a cluster is 3 cpus into 1 big cpu. Thats the whole meaning of cluster. Buisnesses use them all the time, be it for the internet, intranet, or fileserver. The whole point is this:
If client a is using word processing and client b is accessing files on the server, client b gets precidence over client a. So on a larger scale the server if at near full capacity would put client a on hold because in a server enviroment client b is more important. Which is why clusters are so invaluable, now client a won't get shoved to the side when the first computer is at 100% processing compacity. There is no such thing as node this node that.

think of it like this, if you make shoes, and you hire 3 more people to help you make shoes, you wont make any single shoe 4 times as fast, but you will make 4 times the amount of shoes in the time it used to take to make only 1.

thats like a cluster, work in a cluster must be devided up between the nodes by the cluster aware program. thats why tasks such as audio encoding cant be done any faster for a single track, but you can do twice as many tracks in the same time if you have a cluster.

What is your level of experience with Linux? If you are relatively new to it, perhaps the best thing to do would be to get a solid, midrange machine, install Linux, then learn Linux. Once you're comfortable, acquire a second machine, set up your cluster, and expand as much or as fast as you're comfortable with. You'll have a lot more fun and enjoy it a lot more if you don't bite off more than you can chew on day one.

Good luck with it either way and welcome to LQ -- J.W.

Tnx for the replies guys, i've got it all cleared up now ;) :)


@J.W. : i've been working with linux since kernel 2.4.22 and slack9.0 i think :D
i did the right thing and started with slack :study: :study: :study: :cool: :D

Excellent. And if you ask me, Yes, you definitely did the right thing by starting with Slack :) Right on. Anyway I didn't mean for my question to be seen as a challenge, but rather just as a measure that would help people to respond appropriately. Your original post came across (to me anyway) that you were brand new to Linux, and therefore I figured that for a newibe, it would be a better experience to gradually come up to speed with Linux rather than to try to build a cluster as your very first project. Sorry if that didn't come across the right way.

In any event, congrats on solving the issue -- J.W.