Increasing TLB Reach Using Superpages Backed by Shadow Memory.pdf

Increasing TLB Reach Using SuperpagesBacked by Shadow MemoryNovember 24, 1997AbstractThe amount of memory that can be accessed without causing a TLB fault, the so-called reach of a TLB, is failing to keep pace with the increasingly large working setsof applications. We propose to extend TLB reach via a novel Memory Controller TLB(MTLB) that lets us aggressively create superpages from non-contiguous, unaligned regionsof physical memory. This exibility increases the OSs ability to use superpages on arbitraryapplication data. The MTLB supports shadow pages, regions of physical address space forwhich the MTLB remaps accesses to \real physical pages. The MTLB preserves per-basereferenced and dirty bits, which enables the OS to swap shadow-backed superpagesmore eciently than conventional superpages. Simulation of nine applications, including theSPECint95 benchmarks, demonstrated that a modest-sized MTLB improves performanceof applications with moderate-to-high TLB miss rates by 5-20%. Simulation also showedthat this mechanism can more than double the eective reach of a processor TLB with nomodication to the processor MMU.1 IntroductionThe size of application program virtual address spaces has increased dramatically over the years,driven by the availability of large inexpensive DRAMs and the increasing complexity of appli-cations. On most modern machines, virtual memory is implemented by memory managementunits (MMU) that map virtual to physical addresses, usually on a page granularity. Typicalpage sizes are xed at 4-8 kilobytes, which is quite small compared to the total amount ofphysical memory in most machines. This organization eases management of the physical spaceand limits the amount of physical storage wasted due to internal fragmentation.To complete a load or store operation, a processor must rst convert the requestedvirtual address to a physical address. Given the size of typical process virtual address spaces,the total number of translations is huge, and th