One-Pagers‎ > ‎

OP 7: Virtualization

ESX Server provides mechanisms for deduplication in the form of content-based page sharing, effectively leveraging the level of indirection between physical and machine memory.

Design an extension of the mechanism that optimizes a workload for deployment on current computers with NUMA (non-uniform memory architectures) where some of the memory is closer (lower latency, higher bandwidth) to some processors and further from all others. Intel and AMD both have these architectures these days. In your design, try to identify the leverage points from the existing mechanism in ESX Server (as described in the paper), and separate the new ones that might required. You are allowed to use additional (even propose new) hardware mechanisms if this makes sense in your design. 

You should focus on memory management, and assume that the CPU scheduler will be optimized for NUMA (i.e., will try to keep the same VM on the same NUMA node whenever possible).

Favorites: