Architecture Seminar: Efficient Memory Virtualization: can we have higher dimensionality at the cost of none?

Tuesday, April 5, 2016 -
4:00pm to 5:00pm

Speaker Name: 

Jayneel Gandhi

Speaker Institution: 





Abstract: Two important trends in computing are evident. First, computing will become more data centric, where low-latency access to a very large amount of data is critical. Second, virtual machines will play a critical role in enabling server consolidation, security and fault tolerance as substantial computing migrates to shared resources in cloud-like services. Since the software stack accesses data using virtual addresses, fast address translation is a prerequisite for efficient data-centric computation and for providing the benefits of virtualization to a wide range of applications. We show that the overheads of virtual memory are mostly associated with long latency page walks which can be 1D (native) or 2D (virtualized). We address the above challenge in a comprehensive manner that allows fast address translation in both the virtualized and native system for wide variety of workloads. We aim to achieve near-zero overheads of virtual memory through various mechanisms that reduces the dimensionality of page walks while retaining benefits of paging.