Figure 1(b) shows how the components of the average read cost respond to different degrees of replication () in a mirrored system under a random read workload. (In this case, .) The read overhead is lower than the write overhead, because it takes longer for the disk head to settle when servicing a write request. Note that mirroring reduces both the seek and rotational delays of read requests. These components, however, remain significant if mirroring is the only technique employed.