Vmware, Inc.

Seongbeom Kim

Rajesh Venkatasubramanian

Alexander T Garthwaite

Kiran Tati

Haoqiang Zheng

Yury Baskakov

Puneet Zaroo

Irene Zhang

Nikhil Bhatia

Rishi Mehta

Fei Guo

Amitabha Banerjee

Xiaochuan Shen

Jagadish Kotra

In a computer system having multiple memory proximity domains including a first memory proximity domain with a first processor and a first memory and a second memory proximity domain with a second processor and a second memory, latencies of memory access from each memory proximity domain to its local memory as well as to memory at other memory proximity domains are probed. When there is no contention, the local latency will be lower than remote latency. If the contention at the local memory proximity domain increases and the local latency becomes large enough, memory pages associated with a process running on the first processor are placed in the second memory proximity domain, so that after the placement, the process is accessing the memory pages from the memory of the second memory proximity domain during execution.

Memory congestion aware NUMA management

Systems and methods for performing selection of non-uniform memory access (NUMA) nodes for mapping of virtual central processing unit (vCPU) operations to physical processors are provided. A CPU scheduler evaluates the latency between various candidate processors and the memory associated with the vCPU, and the size of the working set of the associated memory, and the vCPU scheduler selects an optimal processor for execution of a vCPU based on the expected memory access latency and the characteristics of the vCPU and the processors. The systems and methods further provide for monitoring system characteristics and rescheduling the vCPUs when other placements provide improved performance and efficiency.

Adaptive CPU NUMA scheduling

Systems and methods described herein align various types of hypervisor threads with a non-uniform memory access (NUMA) client of a virtual machine (VM) that is driving I/O transactions from an application so that no remote memory access is required and the I/O transactions can be completed with local accesses to CPUs, caches, and the I/O devices of a same NUMA node of a hardware NUMA system. First, hypervisor of the VM detects whether the VM runs on a single or multiple NUMA nodes. If the VM runs on multiple NUMA nodes, a NUMA client on which the application is executing the I/O transactions is identified and knowledge of resource sharing between the NUMA client and its related hypervisor threads is established. Such knowledge is then utilized to schedule the NUMA client and its related hypervisor threads to the same NUMA node of the NUMA system.

Systems and methods for non-uniform memory access aligned I/O for virtual machines

Memory performance in a computer system that implements large page mapping is improved even when memory is scarce by identifying page sharing opportunities within the large pages at the granularity of small pages and breaking up the large pages so that small pages within the large page can be freed up through page sharing. In addition, the number of small page sharing opportunities within the large pages can be used to estimate the total amount of memory that could be reclaimed through page sharing.

Identification of page sharing opportunities within large pages

Growing community of inventors

Sunnyvale, CA, United States of America

Seongbeom Kim