|
|
SCL Cluster Cookbook Simple Cluster Recipe |
This is a very brief example of how to build a small cluster with
little emphasis on equipment selection. If nothing else, it will give
you an idea of the work involved in actually constructing the
cluster.
Ingredients:
Gather Pentium II PCs. Make sure PCs are unplugged from power. Remove covers as necessary, insert two Fast Ethernet network interface cards into PC that will serve as the cluster server and connection to the rest of your network. Insert one Fast Ethernet card into each of the other three PCs. Install any other necessary hardware, such as additional memory. Install covers back on computers. Plug PCs into power.
Take four Category 5 cables and connect Fast Ethernet hub or switch to each of the four cluster computers with the cables. Use the remaining Category 5 cable to connect the "extra" Ethernet card in the server to your institution's network (assuming that your institution's network is an Ethernet or Fast Ethernet). Plug in the Fast Ethernet hub's or switch's power cord.
Using the Linux CD and instructions that come with it, install Linux on each of the PCs. Be sure to install the GNU C compiler and C libraries on the server. When you configure TCP/IP, it is suggested that you assign IP addresses of 192.168.1.1, 192.168.1.2, 192.168.1.3, and 192.168.1.4 to the first, second, third, and fourth PC, respectively. The first PC should be your server node (the one with the additional network card). The extra network card in the server node should be assigned an IP address by your institution's network manager.
With Linux now running on each PC, edit the file /etc/hosts
on each of the four computers so that it contains these lines:
192.168.1.1 node1 server
192.168.1.2 node2
192.168.1.3 node3
192.168.1.4 node4
Edit the file /etc/hosts.equiv on each of the four computers
so that it contains these lines:
node1
node2
node3
node4
This is to configure the computers so that MPICH's p4 device
may be used to execute a distributed parallel application.
On the server node, make a directory /mirror. Configure the
server to be an NFS server, and in /etc/exports add this
line:
/mirror node1(rw) node2(rw) node3(rw) node4(rw)
On the other (non-server) nodes, make a directory /mirror.
Add this line to /etc/fstab:
server:/mirror /mirror nfs rw,bg,soft 0 0
This exports the directory /mirror from the server and mounts
it on each of the clients for easy distribution of software between
the nodes.
On the server node, install MPICH. Documentation on MPICH is available at http://www.mcs.anl.gov/mpi/mpich/docs.html.
For each user that you create on the cluster (you will have to create
the user on the each of the nodes), it is advised that you create a
subdirectory owned by that user in the /mirror directory,
such as /mirror/username, where the user can put MPI
programs and shared data files. This way, the user just needs to
compile the MPI programs on the server node, copy the compiled
programs to his or her /mirror subdirectory if necessary, and
run the MPI programs from his or her /mirror subdirectory
using the p4 MPI device.
Cost
The total cost of this small cluster is approximately:
| Equipment | Cost |
|---|---|
| 4 Pentium II's with 64MB RAM, 2GB disks, EIDE CD-ROM | $8000.00 |
| 5 SMC EtherPro 10/100 Network Cards | 400.00 |
| 1 Bay 350T Fast Ethernet Switch | 2500.00 |
| 5 Category 5 UTP Cables | 50.00 |
| Linux CD | 40.00 |
| MPICH | 0.00 |
| Total: | $10,990.00 |