LinuxCon + CloudOpen + ContainerCon NA 2015

Multi-core cache hierarchies benefit applications with several orders of magnitude in performance. However, when many user processes are running in parallel and/or concurrently, cache and memory bandwidth are the most contended resources and must be monitored and used efficiently to get the ultimate performance. In this presentation we look at how to monitor two shared resources, LLC and memory bandwidth consumption, in Linux running on top of Intel Xeon multi-core processors. Intel Xeon microarchitecture provides performance counters which Linux kernel can read on a timely basis and measure the bandwidth from one level of the cache hierarchy to the next – in this case focusing on the L3 cache, which is backed directly by system memory. As a result of this, memory bandwidth consumed by a process is monitored. This framework can be incorporated into Cloud and Data Center applications.