Impact of Extra Space on Eager-writing

In order for eager-writing to be effective, one needs to reserve enough extra space so that a free block can always be located near the current disk head position. Let disk utilization be $U$; we define dilution to be $D_d = 1/U$. For example, when $D_d = 2$, we use twice as much capacity as is necessary.

Figure 1(a) shows how the components of the average write cost respond to different dilution factors ($D_d$) under a simple random write workload running on a 10,000 RPM Seagate disk (ST39133LWV). (In this case, $D_m = D_s = 1$. The block size is 4 KB, and there is no queueing.) In this figure (and the rest of this paper), overhead is defined to include various processing times and transfer costs. Under eager writing, when the closest free block is located in the current track, only rotational latency is incurred. When the closest free block is located in a neighboring track, a track switch or a small seek is also needed, and this time is counted as seek time.

As we increase the amount of extra space, both the rotational delay and seek time decrease as the disk head travels a shorter distance to locate the nearest free block. This improvement reaches diminishing return as the overhead dominates.