From: Andy Glew (glew_at_[hidden])
Date: 1999-08-29 17:11:58
> to explain why some requirements are still floating around in
> the discussion, since I sort of started it off by asking for such a thing.
Actually, Reid, you're not the only one looking for good cache classes.
My requirements are probably for even higher performance, more volatile,
messier, and more dynamically tunable classes than yours are: my application
is simulating CPU architectures, i.e. simulating real CPU hardware caches.
Not just the caches caches that y'all are using today, but also the caches
that I would like to put into the next chip I design. Varying the priority and
management policies is one of the typical things I do.
Since I cannot afford the overhead of virtual functions in many places
(measured: 20% slowdown) I explicitly design data structures for use
in inheritance hierarchies, statically overriding methods.
Hence my interest in designing primitives that can be combined by the
implementor of a caching class, rather than attempting to provide a cache class
that is everything to all users.
The reference Valentin gave us frankly does not look very promising to
me.