Check out the new USENIX Web site. next up previous
Next: Write Latencies Up: Simulation Results and Performance Previous: Segment Utilization Distribution

Read/Write Latency

In previous discussion, we used overall write cost as the performance metric. Overall write cost is a direct measurement of system efficiency. We have shown that WOLF performs encouragingly better than LFS, as the former has much smaller overall write cost than the latter.

However, end-users would be more interested in user-measurable metrics such as the access latencies [3]. Overall write cost quantifies the additional I/O overhead when LFS does the garbage cleaning. The LFS performance is very sensitive to this overhead. To see whether the low overall write cost in WOLF can be translated to low access latencies, we also measured the average file read/write response times in the file system level. We collected the total file read/write latencies and divided the total number of file reads/writes requests. All these results include the cleaning overhead. The results are presented in this subsection.



Subsections

Jun Wang 2001-10-31