Browse Prior Art Database

Cognitive Assistant for Workload Distribution

IP.com Disclosure Number: IPCOM000250041D
Publication Date: 2017-May-19
Document File: 3 page(s) / 201K

Publishing Venue

The IP.com Prior Art Database

Abstract

The invention is applicable for IT environments where the individual software applications are distributed across multiple servers that are organized in a computing cluster. Examples of such clusters are a z/OS SYSPLEX or a cluster of Linux or AIX servers. The invention claims a method to reconfigure application distribution over multiple servers by searching through a history of application placement configurations ( maps ) and associated measured quality of such placements ( health state ), in order to find the most suitable alternate configuration than present.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

1

Cognitive Assistant for Workload Distribution

The invention assumes that you have applications that are not fixed to one server of a computing cluster but can be started on different servers. It also assumes that the cluster is not constantly undergoing significant changes from both, the cluster setup and the application side.

The decision of where the applications are in a computing cluster is usually made by:  An automation software / cluster manager, according to some kind of automation policy  Script-based  Manual operator interaction

This distribution of applications might lead to the following problems:  Too many applications started on one server because automation software or the

operator has no knowledge of optimal distribution. This leaves some servers busy, others idling.

 Applications interact with each other and lead to resource (e.g. disk / memory) shortages on certain servers while others have idle capacity

 Response times for users suffer

IT environments usually have monitoring mechanisms in place which can detect these problems:

 performance measurement tools for the operating system  application-specific monitors  systems management and workload balancing tools

The invention describes a method and apparatus which can propose a distribution of applications in a cluster of computers which is better than a currently running distribution.

How the apparatus works:

1. It periodically collects the status of the application distribution across servers. Such a distribution is called a 'map'. Existing tools, like System Automation for z/OS or Tivoli System Automation for Multiplatforms can provide the location information within a cluster for each software (OS and application). This tools also know about relations between software such as start or stop dependencies. They know if software can be restarted, moved to other systems, stopped, duplicated, and so on. With this knowledge a map is created which includes the software application name and the location information (application A runs on system/LPAR X).

2. It periodically collects performance and monitor data that the user can pre-select and it aggregates this data into a numerical value that is called a 'health score'. Performance Monitoring tools like z/OS Resource Measurement Facility provide system and cluster performance parameters at a (consolidated) high level. For example CPU usage, I/O rate, memory delays. Application performance monitors provide response time values (e.g. SAP dialog response time). All monitor parameters are normalized to a common scale (e.g. 1 to 10, where 1 is considered ‘worst’ and 10 ‘best’). Each single monitor parameter is rated by the operator/administrator for its importance to the overall ‘health score’ condition of the system. For example, if I/O rate is most critical it would be...