Evaluation Benchmarks

For all tests, messages were passed through an unloaded switch. The following measurements were performed:

• Throughput: Average application to application throughput was measured by timing bulk data transfers over a sufficiently long period of time using test programs written from scratch for the different APIs available with STREAMS. These tests were conducted between different machines connected via the unloaded switch.

• Latency: Applications developed from scratch were written for each API studied which echo a message of a specified number of bytes between user processes (locally and over the network). The application waits to receive the M-byte message back. The interaction was repeated N times between the client and server after which timing stops. From the N readings obtained, an average round-trip time for exchanging an M-byte message between the two hosts is calculated.

• Processor Utilization: CPU usage was measured during the throughput and latency tests using TOP [15]