The OST cache

Current split data caches classify data as having either spatial locality or temporal locality. The cache usually comprises two subcaches with each subcache being used to cache one type of data. More precisely, however, data can exhibit not only temporal or spatial locality, but also both localities, or no locality. Current split data caches do not take advantages of this more precise characterization. In addition, runtime locality change is not supported completely by current split data caches, leading to lower system performance. This thesis presents a new data classification method which divides data into five types in terms of the locality they present. A new kind of split data cache, called the OST cache, which has three subcaches, and an innovative control mechanism, which can detect data locality accurately and supports flexible runtime locality change, are also proposed in this thesis. Experimental results show that the OST cache can improve system performance greatly compared to a conventional cache. Compared to other split caches (SS/NS and NTS), the OST cache also shows a better overall system performance. The OST cache can be used in embedded systems which have small L1cache and off-chip L2 cache, and narrow bus between L1 cache and L2 cache.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat