now many nodes can a lan support?
James.P.Lux at jpl.nasa.gov
Fri Jan 10 13:28:11 EST 2003
It's basically a communications bandwidth problem...
What's your bottleneck? I would assume it's the link to the head in your
case, since all those 1KB packets heading out and 100B packets heading back
have to go through it.
1KB > roughly 10 kbits (by the time you add headers, etc.)
100B > roughly 1 kbit..
Clearly the "head to node" traffic will dominate the link, it's ten times
Figure that 10 Mbps wire can carry, comfortably, 5 Mbps..
5000 kbps/10 kb = 500 packets/second
Now you need to find out how long it takes for a node to process the
packet.. If it takes 1 second, you can support 500 nodes. If it takes 0.01
seconds you can support 5.
This is a very rough and ready estimate, of course... You need to address
such issues as:
1) synchronization of the nodes... if they all try to talk at the same
time, your throughput will drop (a lot). If node 2 tries to talk to the
head at the same time node 1 is, then either node 2 blocks, or its packet
gets held somewhere. In your application, the dominant traffic appears to
be from the head to the nodes, so it would inherently tend to be sequenced
2) variability of the processing time on the node...
This is all pretty standard queuing theory stuff. You could write a little
simulation program to try it all out. R.G.Brown's book at the Duke Brahma
site talks quite a bit about how to calculate parallelism and speedup
At 11:15 AM 1/10/2003 -0600, you wrote:
>My guess is this question has been asked before, but I've not been
>able to find it in the archive file. The question is given a typical
>10Mb/s lan how many nodes can a cluster support? Assume the cluster
>has its own switch, the head and nodes are connected in a star with
>the switch, the cluster lan is isolated from all other non-cluster
>network traffic, the only way to reach a node is through the head,
>ignore extra traffic from TCP handshakes and such, and the the
>data packet for a work unit is 1KB with a 100B results packet back
>to the head.
>How do I calculate this?
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf