Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

how to improve cache performance on OS-level?


as we know ,the shared L2 cache is the important resource to improve cache performance.

most of performance improvement is depended on hardware ways.

there are some methods to improve cache performance to distribution shared L2 cache on OS-level ?

which improvements are feasible on shared L2 cache through OS-level?

could you give me some advice ?

thank you for reading.

0 Kudos
1 Reply
Black Belt
The sharing of L2 cache is dependent on processor archetecture. On some of the newer processor designs, each core has its own L2 cache.If the processor has HT then the HT siblings will share L2 but the performance may/may not benefit on sharing depending on mix of integer verses floating point operations.

For these processors, they tend to have an L3 cache. So some performance gains can be had by organizing work within an L3 cache. L3 cache is generally the socket cache, but in some cases this could be a die cache, when two dies are co-joined into one socket.

Scheduling to cache level is difficult to do with most threading tools, QuickThread being the exception see: (by the way QuickThread is not QT).

Jim Dempsey
0 Kudos