We need to define a miss cost--what should this be based on? This is especially important if we do consider a multi-tier cache.