This document discusses parallel disk models and techniques for emulating a parallel disk head model on multiple disks that can each access one block per I/O. It presents several algorithms for writing and reading blocks in parallel across disks with low maximum load, including:
1) Queued writing that stores each block on two disks and achieves an expected write time of 1+exp(-D) for (1-ε)D blocks.
2) A majority method that stores each block on three disks and can schedule writes and reads together with maximum load O(log log n / log n).
3) Techniques for orienting allocation graphs to achieve maximum loads of 1, 2, and O(log log