Determining Placement Fitness For Partitions Under A Hypervisor

Abstract

A technique for determining placement fitness for partitions under a hypervisor in a host computing system having non-uniform memory access (NUMA) nodes. In an embodiment, a partition resource specification is received from a partition score requester. The partition resource specification identifies a set of computing resources needed for a virtual machine partition to be created by a hypervisor in the host computing system. Resource availability within the NUMA nodes of the host computing system is assessed to determine possible partition placement options. A partition fitness score of a most suitable one of the partition placement options is calculated. The partition fitness score is reported to the partition score requester.

Description

BACKGROUND[0001]1. Field[0002]The present disclosure relates to virtually-partitioned computing systems. More particularly, the disclosure concerns partition placement in systems exhibiting Non-Uniform Memory Access (NUMA) characteristics.[0003]2. Description of the Prior Art[0004]By way of background, many current computer systems utilize NUMA-like hardware designs. In these designs, microprocessor chips have access to memory that is relatively “close” to them (low latency and high bandwidth) and also memory that is relatively “far” away (higher latency, sometimes lower bandwidth). System topologies may involve many levels of memory attachment, and as interconnections between the processors themselves.[0005]When a virtual partition is created to provision a virtual machine (VM), it is generally necessary to allocate a set of resources for the partition. In the case of some resources, such as processors, the resource mappings can be soft. For example, a “home processor” for a virtua...

AI Technical Summary

Benefits of technology

The patent text describes a system, method, and computer program for determining the best placement for virtual machines in a computer system with non-uniform memory access. The system assesses the availability of computing resources in the host system and calculates a fitness score for each possible placement option. The fitness score takes into account factors like the hypervisor's ability to provision resources and the resource characteristics needed by the virtual machine. The system can then make a reservation request for the necessary resources and proceed with the placement. The technical effect of this system is to optimize the allocation of computing resources for virtual machines in a hyperV Sysplex system, ensuring efficient and effective performance.

Problems solved by technology

The NUMA properties of many computer systems can create rather large performance differences between partitions with “good” placement versus “bad” placement.

In this case, a first partition that comprises CPU core 0 on processor chip A has been allocated memory from physical memory X that is locally attached to processor chip A. A second partition that comprises CPU core 1 on processor chip A has been allocated memory from physical memory Y. This memory is locally attached to processor chip B but is remote with respect to processor chip A. For many workloads, the need for remote memory accesses in the second partition will produce a performance penalty.

Remote accesses to high speed I / O can also introduce problems, particularly when large amounts of data are involved.

However, as far as known, the actual resource usage is not exposed in a granular enough way to determine if a partition will fit well upon a system or not.

Thus, existing management tools are not able to judge whether a particular partition placement will be “good” or “bad” from the standpoint of resource access performance.

Images