GCASR 2016‎ > ‎Presentations‎ > ‎

Efficient Virtual Address Translation for Modern Workloads

Efficient Virtual Address Translation for Modern Workloads
Michael Swift (University of Wisconsin -- Madison)


Page-based virtual memory improves programmer productivity, security, and memory utilization, but costs performance overhead due to costly page table walks after TLB misses. This overhead can reach 50% for modern workloads that access increasingly vast memory with stagnating TLB sizes.

I will discuss two approaches to reducing the overhead of virtual memory based on "ranges", which are sets of a process’s pages that are both virtually and physically contiguous First, I will describe mapping one range from a process’s linear virtual address space with a single direct segment, while page mapping the rest of the virtual address space. Direct segments use minimal hardware—base, limit and offset registers per core—to map contiguous virtual memory regions directly to contiguous physical memory. I will also show how they can be applied to virtualized settings, where address translation costs are high. Direct segments are very efficient, but place substantial demands on hardware and software.  Second, I will describe Redundant Memory Mappings (RMM) as a generalization of direct segments, which provide an efficient, alternative representation of many ranges. RMM translates each range with a single range table entry, often enabling modest number of entries to translate most of the process’s address space.

Mike Swift is an associate professor at the University of Wisconsin, Madison. His research focuses on the hardware/operating system boundary, including devices drivers, new processor/memory technologies, and transactional memory. He grew up in Amherst, Massachusetts and received a B.A. from Cornell University in 1992. After college, he worked at Microsoft in the Windows group, where he implemented authentication and access control functionality in Windows Cairo, Windows NT, and Windows 2000. He received a Ph.D. on operating system reliability from the University of Washington in 2005.