Check out the new USENIX Web site. next up previous
Next: Implication of Different Disk Up: Read/Write Latency Previous: Write Latencies

Read Performance

Figure 9(a) shows the file read performance of LFS and WOLF under eight traces. Figure 9(b) plots the performance improvement of WOLF over LFS. The results show that, for most traces, the read performance of WOLF is at least comparable to that of LFS. This is expected, as WOLF does not directly affect the read operations of LFS. Although WOLF changes the physical layout on disk for LFS, WOLF's grouping algorithm includes the similar policy which is used in locality-grouping rules of regular LFS, such that files in same directory are put in same segment and etc., WOLF does not have much impact on the read performance when the load is light. When the load is heavy, we may see a little better read performance of WOLF than that of LFS because WOLF reduces the cleaning overhead so that WOLF ameliorates the competition of disk bandwidth. RES and TPC-D got little loss for their more random reads because random reads have poor spatial locality which results in much longer disk seeks and rotations during garbage collection.

Figure 9: Average File Read Response Time. Errorbar shows the standard deviation. Disk utilization is 90%.
\includegraphics[width=3.2in, height=1.8in]{readerrorbar.eps} \includegraphics[width=3.2in, height=1.8in]{pr.eps}


next up previous
Next: Implication of Different Disk Up: Read/Write Latency Previous: Write Latencies
Jun Wang 2001-10-31