Storage hierarchies have existed for as long as stored program computers themselves. The Ferranti Mark 1 installed at Manchester University in February 1951 had a main store made up of eight Williams Tubes each containing 32 40-bit words and a drum backing store having a capacity of 3.75 Kwords. Initially users were required to organise their own store transfers between a selected drum track and a selected pair of Williams Tubes, but the Mark 1 Autocode system introduced in 1954 carried out these tasks automatically on behalf of users and so made the two levels of storage appear to them as a one-level store. This arrangement was possible because the Mark 1 was a single user machine, and its performance was acceptable because the time required to transfer one drum track was roughly the same as the time for a programmed floating-point addition. Later developments in both architecture and technology led to the need for more sophisticated systems and in this chapter we shall consider the virtual memory and paging systems used in Atlas, the cache stores used in some models in the IBM System/360 and System/370 ranges, and the MU5 storage hierarchy. Before discussing these systems in detail we introduce the technique of store interleaving, a technique used to obtain increased performance in these and many other computer systems.
|Title of host publication||Architecture of High Performance Computers|
|Subtitle of host publication||Volume I: Uniprocessors and vector processors|
|Place of Publication||London|
|Publisher||Macmillan Education UK|
|Number of pages||23|
|Publication status||Published - 1989|