Q. Can servers in the cluster be of different architectures?
A. No, all servers in a cluster MUST be either i386 OR x86_64
Just to be completely sure that I’m understanding this correctly, this is stating that all nodes in the cluster must be either 32-bit or 64-bit? Is it as simple as that?
So, can a 64-bit Opteron node, a 64-bit Xeon node and a 64-bit Pentium node all be mixed together with no problem?
Regarding the choice between AMD & Intel, are there any advantages of running InterWorx on either platform?
With all the data that could move between the cluster manager and many nodes, is the system bus bandwidth capacity a very important thing for the cluster manager? With Intel using FSB and AMD using hypertransport, would there be a significant advantage in having the cluster manager run on a server with the higher bandwidth hypertransport?
Yes. Really the only reason for this rule has to do with the RRD bandwidth graphs in SiteWorx. There is a single .rrd file for each domain on each node, and a 32 bit server can’t work with a RRD file created on a 64 bit server, and vice versa. The only clarification I can think to make is “The cluster manager and the nodes must be of the same arch”
Regarding the choice between AMD & Intel, are there any advantages of running InterWorx on either platform?
I personally have more experience with Intel, but I know that folks use AMD successfully without issue.
With all the data that could move between the cluster manager and many nodes, is the system bus bandwidth capacity a very important thing for the cluster manager? With Intel using FSB and AMD using hypertransport, would there be a significant advantage in having the cluster manager run on a server with the higher bandwidth hypertransport?
I can’t comment directly regarding hypertransport, since I know next to nothing about it. But, I can tell you that the main sources of traffic between the CM and the nodes are the NFS (mostly the nodes reading data from the CM), and the load balancing of the requests by the CM to the nodes. It is the nodes themselves that respond to the client’s load balanced web requests - the CM only passes along the request to the node. The CM is not involved in the (higher bandwidth, in the case of web traffic) response. Of course the CM itself can be a part of the load balancing policy, and in that case it will respond directly to the client, but there’s no CM -> node traffic in that case.
Regarding system bandwidth, I was referring to the motherboard front-side bus (or hypertransport for AMD) bandwidth, not data transfer bandwidth over the Internet.
Basically, the way I understand it, the FSB (front-side bus) can be a limiting factor / bottleneck in certain designs due to having less bandwidth for moving data through the system components.
So, in relation to an InterWorx cluster, could a cluster manager get to a point where it is sending so much data between the CM and the nodes that the system bus bandwidth becomes some sort of bottleneck?
If that could happen, maybe it would be best to go with an AMD based system with hypertransport so that the bus bandwidth would not be saturated as quickly (if at all).