On Saturday, December 15, 2018 2:48:30 AM CET Dan Williams wrote: > Changes since v4: [1] > * Default the randomization to off and enable it dynamically based on > the detection of a memory side cache advertised by platform firmware. > In the case of x86 this enumeration comes from the ACPI HMAT. (Michal > and Mel) > * Improve the changelog of the patch that introduces the shuffling to > clarify the motivation and better explain the tradeoffs. (Michal and > Mel) > * Include the required HMAT enabling in the series. > > [1]: https://lkml.kernel.org/r/153922180166.838512.8260339805733812034.stgit@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx > > --- > > Quote patch 3: > > Randomization of the page allocator improves the average utilization of > a direct-mapped memory-side-cache. Memory side caching is a platform > capability that Linux has been previously exposed to in HPC > (high-performance computing) environments on specialty platforms. In > that instance it was a smaller pool of high-bandwidth-memory relative to > higher-capacity / lower-bandwidth DRAM. Now, this capability is going to > be found on general purpose server platforms where DRAM is a cache in > front of higher latency persistent memory [2]. > > Robert offered an explanation of the state of the art of Linux > interactions with memory-side-caches [3], and I copy it here: > > It's been a problem in the HPC space: > http://www.nersc.gov/research-and-development/knl-cache-mode-performance-coe/ > > A kernel module called zonesort is available to try to help: > https://software.intel.com/en-us/articles/xeon-phi-software > > and this abandoned patch series proposed that for the kernel: > https://lkml.org/lkml/2017/8/23/195 > > Dan's patch series doesn't attempt to ensure buffers won't conflict, but > also reduces the chance that the buffers will. This will make performance > more consistent, albeit slower than "optimal" (which is near impossible > to attain in a general-purpose kernel). That's better than forcing > users to deploy remedies like: > "To eliminate this gradual degradation, we have added a Stream > measurement to the Node Health Check that follows each job; > nodes are rebooted whenever their measured memory bandwidth > falls below 300 GB/s." > > A replacement for zonesort was merged upstream in commit cc9aec03e58f > "x86/numa_emulation: Introduce uniform split capability". With this > numa_emulation capability, memory can be split into cache sized > ("near-memory" sized) numa nodes. A bind operation to such a node, and > disabling workloads on other nodes, enables full cache performance. > However, once the workload exceeds the cache size then cache conflicts > are unavoidable. While HPC environments might be able to tolerate > time-scheduling of cache sized workloads, for general purpose server > platforms, the oversubscribed cache case will be the common case. > > The worst case scenario is that a server system owner benchmarks a > workload at boot with an un-contended cache only to see that performance > degrade over time, even below the average cache performance due to > excessive conflicts. Randomization clips the peaks and fills in the > valleys of cache utilization to yield steady average performance. > > See patch 3 for more details. > > [2]: https://itpeernetwork.intel.com/intel-optane-dc-persistent-memory-operating-modes/ > [3]: https://lkml.org/lkml/2018/9/22/54 Has this hibernation been tested with this series applied?
