A Simple Linear Model of Demand Paging Performance

Predicting the performance of a proposed automatically
managed multilevel memory system requires 
a model of the patterns by which programs refer to the
information stored in the memory.  Some recent 
experimental measurements on the Multics virtual memory
suggest that, for rough approximations, a remarkably 
simple program reference model will suffice.  The simple
model combines the effect of the information 
reference pattern with the effect of the automatic management
algorithm to produce a single, composite 
statement: the mean number of memory references between
paging exceptions increases linearly with the 
size of the paging memory.  The resulting model is easy
to manipulate, and is applicable to such diverse 
problems as choosing an optimum size for a paging memory,
arranging for reproducible memory usage charges, 
and estimating the amount of core memory sharing.

CACM April, 1974

Saltzer, J. H.

paging, demand paging, memory models, program models,
performance measurement, multilevel memory 
systems, virtual memory, associative memory, memory usage accounting, Multics

3.79 4.30 4.32

CA740401 JB January 18, 1978  10:41 AM

1892	4	2669
1901	4	2669
2085	4	2669
2095	4	2669
2218	4	2669
2277	4	2669
2319	4	2669
2319	4	2669
2329	4	2669
2358	4	2669
2377	4	2669
2373	4	2669
2374	4	2669
2380	4	2669
2434	4	2669
2499	4	2669
2481	4	2669
2501	4	2669
2582	4	2669
2582	4	2669
2625	4	2669
2626	4	2669
2669	4	2669
2669	4	2669
2669	4	2669
2669	4	2669
2828	4	2669
2863	4	2669
2863	4	2669
2881	4	2669
2928	4	2669
2996	4	2669
3006	4	2669
3016	4	2669
3052	4	2669
3054	4	2669
3067	4	2669
1751	5	2669
1901	5	2669
2017	5	2669
2358	5	2669
2669	5	2669
2669	5	2669
2669	5	2669