Check out the new USENIX Web site. next up previous
Next: Performance under WAN conditions Up: Performance Evaluation Previous: Trace-based experiments

Flash Performance Breakdown

The next experiment focuses on the Flash server and measures the contribution of its various optimizations on the achieved throughput. The configuration is identical to the single file test on FreeBSD, where clients repeatedly request a cached document of a given size. Figure 11 shows the throughput obtained by various versions of Flash with all combinations of the three main optimizations (pathname translation caching, mapped file caching, and response header caching).

Figure 11: Flash Performance Breakdown - Without optimizations, Flash's small-file performance would drop in half. The eight lines show the effect of various combinations of the caching optimizations.

The results show that each of the optimizations has a significant impact on server throughput for cached content, with pathname translation caching providing the largest benefit. Since each of the optimization avoids a per-request cost, the impact is strongest on requests for small documents.

Peter Druschel