Cache design for high performance computing requires the realization of two seemingly disjoint goals of higher hit ratios at reduced access times. Recent research advocates the use of ”resizable” caches to exploit cache requirement variability in programs. Existing schemes for resizable caches effectively employ either of the two fundamentally different methods: by changing the cache organization itself or by using a proper resizing strategy, that is, either static or dynamic resizing. Our paper looks at a new dynamic resizing strategy that aims at run time manipulation of the cache parameters to improve its performance. Two algorithms for dynamic reconfiguration are proposed and the results explained.
P. Srivatsan, P. B. Sudarshan, P. P. Bhaskaran