The design tradeoff between miss rate and access time in set-associative caches
versus direct mapped caches has led several researchers to suggest ways to
achieve miss rates similar to two-way set-associative caches with fast access
times by using modified direct-mapped caches that use a serialized or
sequential search to compare tags in set-associative caches. We simulate
two such caches - the Predictive Sequential Associative (PSA)
[1] Cache and the Column Associative (CA) [2]
Cache and compare their performances. The CA cache minimizes the
conflicts that arise in direct-mapped accesses by
allowing conflicting addresses to dynamically choose alternate hashing
functions,so that most of the conflicting data can reside in the cache. At the
same time, the critical hit access path is unchanged. The PSA cache
is designed to overcome the fixed order probe and the delay in
access path that occurs in sequential access set-associative caches. It uses
prediction sources to guide the cache examination, reducing the amount of
searching and thus the average access latency. The CA cache and
the PSA cache with 2 predictive mechanisms - the ``Eff''
(described as the one with best performance) and the ``XOR-5-5''
(described as the one which is easy to implement and has less access time) are
to be simulated with ATOM on the DEC ALPHA and the performance compared.
The simulation is to be performed on two benchmarks - compress and eqntott.
Although the PSA cache does not offer a significant performance over CA cache in reducing the miss rate, it considerably decreases the average cache occupancy which is
important in avoiding delay due to conflicts between outstanding and newly issued
references.Our simulation results show a trend similar to that obtained by Calder et al. [1]