3.3. Deadline-driven, Random
Another way to distribute the system load across the clusters is to employ a random placement of data across the clusters ,  (Figure 3.6). With deadline-driven (DD) approach, each block request is tagged with a deadline and the disk services requests using an earliest-deadline-first (EDF) policy. Note that round-robin (RR) uses an object-based retrieval while random uses a block-based retrieval, thus, a request with RR is for an entire object while with random, it is for a block of an object. Thus, a request with RR is for an entire object while with DD, it is for a block of an object.
Figure 3.6. Random placement.
With random, the system may suffer from a statistical variation of the number of block requests ...