Experimental Evaluation

  The architectural discussion is largely qualitative, and consequences for system performance are useful to understand. The best evaluation would be application-oriented benchmarks for applications in distributed environments, however this is always difficult for a new prototype system. Instead, we use micro-benchmarks and macro-benchmarks to obtain first-order quantification of performance, as well as identification of overhead introduced by the access control mechanism.

Our test hosts are 1 GHz Intel PIII machines with 256 MB of memory and 10 GB Western Digital Protege IDE hard drives. In the two-host, client-server tests that explore the network performance of our system, we connect our machines with 100 Mbps Ethernet. We did not use IPsec for the measurements, because results would vary considerably depending on the specific IPsec configuration parameters. For an evaluation of the performance impact of using IPsec in various configurations we refer the reader to [22]. In the following tables, FFS means measurements taken on the local file system. NFSv2 and NFSv3 denote measurements over NFS protocol versions 2 and 3 respectively using the UDP protocol. We did not measure our performance when running over TCP.