>>>>> "Sanjay" == Sanjay Bhandari <sbhandari@xxxxxxxxxxxx> writes: Sanjay> (If this is the wrong forum for this question, a pointer to Sanjay> the right one will be appreciated) Sanjay> What can I expect in terms of TLB and cache sharing between Sanjay> different processes when they are all non-exec'ed children of Sanjay> the same parent (i.e. fork but no exec)? There is no TLB sharing in standard Linux. Matt Chapman had a patch that enabled TLB sharing for memory regions with the same protections and virtual address in different address spaces; benchmarking showed a very large reduction in ITLB misses, but lesser reduction in DTLB misses (COW means that data areas are not actually shared). Overall, ITLB misses weren't a big enough factor in the workloads we had to make a real difference. If you had a workload that had a large shared memory region mapped into the same address in multiple processes, the tradeoffs may differ. Itanium caches are physically mapped, so they are always shared. -- Dr Peter Chubb http://www.gelato.unsw.edu.au peterc@xxxxxxxxxxxxxxxxxx http://www.ertos.nicta.com.au ERTOS within National ICT Australia - : send the line "unsubscribe linux-ia64" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
|