< previous page page_643 next page >

Page 643
0643-01.gif
Figure 9.34
A cluster of independent disks.
9.5.5 Clusters of Independent Disks
In this configuration (Figure 9.34), we read complete files from disk to buffer, then (without overlap) transfer them from buffer to processor. (Disk read rate = buffer (B) to processor (P) transfer rate = 2.6 ms per 4-KB block.)e.g., if file A consists of blocks A1,A2,An, then all blocks are assigned to a single drive and access to A would be:
0643-02.gif
Suppose we have the preceding workload, f [244], where 70% of the accesses are to files of one block and the remaining files are uniformly distributed from 2 to 16 blocks. Now the expected file length E(f) is:
0643-03.gif
where Max is the maximum number of blocks in f and f(k) is the probability that there will be exactly k blocks in f.
For our case:
0643-04.gif
If the interarrival time for requests was 40 ms and requests were uniformly

 
< previous page page_643 next page >