IBM PowerVM technology has been enhanced to include memory virtualization through the Active Memory Sharing (AMS) feature. With this addition, PowerVM empowers IBM customers to optimize memory utilization similar to Shared Processor Partition’s (SPLPAR) CPU optimization. The PowerVM hypervisor manages real memory across multiple AMS enabled partitions, distributing memory to partitions based on their workload demand. Analogous to low CPU utilization in IT datacenters, actual memory utilization in datacenters can be also low. PowerVM AMS is intended to enable customers to manage the memory utilization at a higher level potentially reducing the need for additional memory investment.
In today’s IT environments there are multiple factors that play into low memory utilization. Workloads running in a partition have a working set size in memory which, if the partition’s activity level goes down, the working set can go down either through releasing memory or through OS page stealing. In either case the freed or aged pages do not get used for other partitions. Second, customer workloads have high peaks infrequently, so memory utilization levels are low frequently. AMS optimizes memory utilization by sharing physical memory across multiple partitions so memory pages that are not actively used by one partition are given to another partition. Therefore through consolidation of workloads that do not peak concurrently, overall memory utilization levels can be increased in a system.
AMS also enhances a customer’s ability to optimize CPU, and I/O resources along with memory as it allows them to consolidate a larger number of partitions on a system. Until now, physical memory has placed a limitation on how many partitions a customer can configure on IBM Power Systems. AMS reduces this requirement thereby may allow customers to create additional partitions.
This paper provides a high-level overview of Active Memory Sharing including an overview of its design and capabilities, a description of its benefits, and usage considerations. The focus of this paper will be the performance aspects of AMS, both where and why it will be beneficial as well as covering the environments where AMS may not be suitable.