next up previous contents
Next: MPI Up: Background Previous: Background

Apparatus

Both merge-sort and quadtree compression were implemented and tested on a two-node Compaq ServerNet system. Both nodes are dual Pentium Pro Compaq workstations, with 64 MB RAM, and connected by both the campus 10baseT Ethernet network and by a ServerNet gigabit router. The ServerNet NICs are connected to the workstations' PCI busses. They are capable of full duplex communication with a bandwidth of 50MB/s in each direction, for a total 100MB/s or 800Mb/s. The NICs are connected to two of the six ports of a ServerNet Interface Switch. The router uses wormhole routing to connect any pair of ports simultaneously, for a total bandwidth of 300MB/s. Data transmission, packet switching, flow control, and routing are all implemented in the the switch's hardware.
Both workstations were running Windows NT 4.0. The parallel software was developed using Microsoft Visual C++ 6.0, and the MPI Pro implementation of the Message Passing Interface (MPI) standard library for interprocess communication.


Garth Brown
2001-01-26