#Shared cache vs private cache code#

#Shared cache vs private cache code#

General-Purpose SystemsДЄs mentioned earlier, transparent caches can improve code even if that code has not been designed or optimized specifically for the given cache's organization.

Wang, in Memory Systems, 2008 Perspective: Embedded vs. Future Linux kernels supported on Intel Xeon Phi coprocessors may include transparent huge page support.Д«ruce Jacob. There is also the hugetlbfs Linux library package that can be added to the environment that enables huge page support for standard allocation calls. In C/C++, the mmap () call provides a means to request huge pages. However, to ensure your application uses huge pages, it may be necessary to use means other than standard library calls (malloc) to allocate memory. Using huge pages for large data sets and arrays can provide a significant performance boost by reducing TLB misses versus using the default of 4-KB pages depending on the application memory access patterns. Two-megabyte huge page support is available, but the application and/or environment may need modification to specifically request it. As of this writing, 64-KB page support is not prevalent in existing microprocessors and generally will not be available without specific support additions to the Linux kernel and therefore currently goes unused. Four-kilobyte pages have been the long-time primary standard supported by Linux operating systems. Of course, the use of smaller page size entries will result in less overall memory mapped and available for fastest access. In the L2 TLB, there are no restrictions on mixing page size entries. In a shared-cache organization, coherence can be part of the cache's existing interconnect infrastructure. In a private cache organization, coherence must be maintained after the last-level cache. Note that the choice of LLC cache organization impacts how and where the cache coherence can be implemented, at least to maintain coherence between the caches on-chip. However, the key application characteristics concerning the amount of data sharing and type of sharing are important for the entire design space. Of course, there is a large solution space between the two extremes, including replication of read-only blocks, migration of blocks, and selective exclusion, to name just a few. In essence (and at the risk of oversimplifying), the CMP design team, building a chip with C cores and room for N bytes of last-level cache, must decide between building C private caches of N/C bytes each or one large shared cache of N bytes. Furthermore, the shared cache naturally accommodates variations between the working-set sizes of the different threads. Cache blocks that are referenced by more than one core are not replicated in the shared cache. In this case, a single, large, last-level cache, which may be centralized or distributed, depending on bandwidth requirements, is shared by all the on-die cores. On the other hand, if the target multi-threaded applications exhibit a significant amount of sharing, or if the threads have varying working-set sizes, a shared-cache CMP is more attractive. On a cache miss, the hierarchies of all the other cores' caches must be snooped (depending on the specifics of the inclusion policy). Any memory block that is shared by more than one core is replicated in the hierarchies of the respective cores, thereby lowering the effective cache capacity. In such a case, each core has its own private cache hierarchy. For example, if the target multi-threaded applications exhibit little or no sharing, and the threads have similar working-set sizes, a simple вЂњSMP on a chipвЂќ strategy may be the best approach. The most important application characteristics that drive this cache hierarchy and design are the amount and type of sharing exhibited by important multi-threaded applications. As multi-core processors become pervasive and the number of cores increases, a key design issue facing processor architects will be the hierarchy and policies for the last-level cache (LLC). Wang, in Memory Systems, 2008 Shared or Private Last-Level Cache?ДЄnother cache-organization issue is the design choice of a shared versus private last-level cache.