next »

[vsloathe]

First a brief explanation about my process in getting to this point, why I have chosen the pieces of technology we are going to use, etc.

I chose the Single System Image (SSI) model of clustering mainly because it allows for little to no modification of current code, and has the highest "just works" factor of any clustering solution. As far as your OS and any applications are concerned, an SSI cluster may as well be one machine. You shouldn't have to change anything to make use of the power of SSI clustering.

I decided to use openSSI as the clustering solution, and they offer simple installation solutions for Red Hat/Fedora Core, Debian and Knoppix. Of those choices, Debian was the most palatable for the sorts of things I will be doing on this cluster (mainly command line PHP, Python, C), and I'm also a big apt fan so the choice was simple. I used the Debian 4.0 netInst CD (180MB download). 2 of the machines are 64 bit and 2 are 32 bit. They should all cluster together quite nicely in spite of the disparate architectures.

Network:
I'm just using a 100MB/s switch I had lying around, but I plan on upgrading to a gigabit switch and installing an extra gigabit NIC in each box. Each box will have 2 NICs - one attached to the cluster switch and one attached to a router to the outside world.

UPDATE: So far, I have the master node up and running Debian. I'll be net booting and configuring the rest of the nodes this evening. Updates and pictures to follow shortly.

[nutballs]

Im interested in how this goes. because frankly, I don't like managing multiple machines with "specific roles". I like the idea of having just 1-big-ass-box even its made of a few smaller ones. Even at the cost of efficiency. 1 node dies, you replace and replicate, and done.

Am I understanding correctly that SSI is basically that? the "machine" gets bigger as you add more nodes? Or is this just a clone type cluster?
does this work for a LAMP setup?

any recommended links so I can read up?

I guess I am technically trying to build a grid-cluster.

[vsloathe]

Am I understanding correctly that SSI is basically that?

Yes, though it depends on the implementation. MOSIX and OpenMOSIX are the most prominent and famous examples of an SSI system, so you should read a little about those. Note that OpenMOSIX is no longer supported which is why I chose OpenSSI.

the "machine" gets bigger as you add more nodes? Or is this just a clone type cluster?
does this work for a LAMP setup?

Yes and yes. Even the master node can go down and your cluster will be fine. The setup is fairly easy but every machine needs to have PXE (boot from ethernet) enabled.

any recommended links so I can read up?

http://openssi.org/
http://www.mosix.org/

I guess I am technically trying to build a grid-cluster.

Well, a cluster is a cluster - kind of. You're basically looking at either Beowulf (good for math stuff, not SSI, requires code specifically written for it), or Mosix/SSI. Most "cloud" hosting setups are basically an SSI image with a bunch of Xen nodes, since Xen is so extensible and widely hardware compatible.

I'm building the cluster not so much for reasons of failover (though that is a plus, I'm going to use NFS to combine every single disk in the cluster into a giant data warehouse), but more performance. I'm looking at aggressively expanding my spider operations and I need a beefy machine with plenty of horsepower that's able to make use of a ton of bandwidth (separate cable IP on a 24MB/s pipe).

[nutballs]

cool cool.

I am thinking the same thing to some extent. 1 thing that might be interesting is to use Amazon EC2 to build a large cluster, that runs massive fast, but then can be turned off at the end of a run. but in my case, its mostly a desire to have only ONE setup.

[nutballs]

related thoughts and conversation. small cheap machines that can do this.

I have been trying to find a viable replacement to a mac-mini. I like the footprint and the low wattage, but, they are a tad expensive. also to use a standard linux install, instead of OSx, which although is pretty nice, doesn't really fit into an "enterprise" type scenario i think.

So. any thoughts on contenders for mini-nodes?
I think 2gb limit seems to be normal for what I have looked at, at least that is priced reasonably. And thats probably fine for a cluster/grid.

[vsloathe]

I need to post an update anyway.

After a lot of futsing, I went ahead and set them all up running ClusterKnoppix.

I have concerns with ClusterKnoppix because it makes use of OpenMosix, which is no longer maintained. But I installed and configured apt on them and they're 110% less fuss to configure and manage than they were when I was trying to set them all up with Debian and OpenSSI.

As far as cheap cluster nodes go, I'll have to get back to you. You don't need homogenous nodes to run any Mosix based cluster software. 2 of the boxes are AMD64 - one turion one athlon - and 2 of the machines are intel - one P4 and one P4EHT. The nodes all have differing quantities of RAM and such as well. I'll post a screenshot of the load-balancing monitor app when I get home.

[vsloathe]

Ok, pricing out cluster nodes:

Nice platform:
http://www.newegg.com/Product/Product.aspx?Item=N82E16856167036
Pick your processor:
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010340343%201051733662%204027&name=%2475%20-%20%24100
And some memory:
http://www.newegg.com/Product/Product.aspx?Item=N82E16820148160

Buy yourself a gigantic hard disk to stick in the master node, and you can have yourself an army of drones for just over $200 a piece with 3 cores and 4GB RAM each.

[nutballs]

jesus that thing is huge. I could fit like 8 mac minis in there.

i guess you missed that qualifier

ill post what I find as well.

[nutballs]

what about centOS5 cluster? I was just installing a centOS5 VM and noticed during the graphical install a choice for Clustering, as well as Storage Clustering.

Im gonna go read about it, but figured I would post first.

[nutballs]

I just realized I can probably mock this whole thing up under VM. duh.
I might have to give it a whirl.

[mampy]

I believe Knoppix is SSI or some port of openSSI
but cluster(ed) Knoppix(ii) is by far the easiest to set up:

http://openssi.org/cgi-bin/view?page=contrib/KNOPPIX/README.html


I think version 3.6+ of knoppix has port of openssi and is debian based:

http://distrowatch.com/table.php?distribution=clusterknoppix

I am most curious to hear of how you come along with your project..