Introduction

Business intelligence (BI) and business analytics

Business intelligence (BI) and business analytics have become widely accepted solutions for more than just business analysts and chief executives. Large enterprises need to be able to access BI and analytics on a comprehensive and expanding scale. To support customers in taking advantage of business intelligence, this white paper includes information about the setup, performance, and z/VM resource management of a distributed Cognos® Business Intelligence (BI) server for Linux™ on System z®, running under z/VM.

Some organizations have various BI capabilities in place but want to add more. Others want to use their capabilities more effectively and efficiently. Imagine you can choose an analytics solution that works for your organization today, but can expand to meet your needs in the future. You could order several options, such as personal discovery tools, and expand to a fuller option later. No matter what you order, the later addition of other services would be seamless and easy.

From business intelligence to financial performance and from strategy management to analytics applications, Cognos software can provide whatever your organization needs to become top-performing and analytics-driven.

Cognos Enterprise combines methods of reporting, analysis, modeling, planning, and collaboration for smarter decision-making and better business outcomes. It equips users with what they need to explore information, analyze key facts, quickly collaborate to gain alignment with key stakeholders and act with confidence to drive better business outcomes.

These premium features are available on Linux on System z under z/VM for a reduced total cost of ownership and an easier centralized environment, which results in an increase of the return on investment.

Objectives

From the maintenance and administration view, BI-workloads have a certain drawback. They need a high amount of system resources in terms of memory and CPU to analyze a large amount of data with short response times. Probably these resources are only needed for a limited amount of time each day. The remaining time, these resources are possibly idling. When running in a virtualized environment, they could be used for other workloads running on the same hypervisor. However, this bears the risk, that the BI-workloads, when running in parallel, do not get sufficient resources and the transaction response times increase. The ideal solution would be a management tool which guarantees that a selected workload gets a certain amount of resources regardless of the load level of the system.

z/VM Resource Manager (VMRM)

With VMRM, z/VM offers a feature that provides a workload management feature for the resource CPU. This study analyzes how VMRM can manage the CPU resource for two different workload types that run in parallel:

• Cognos BI:

  A CPU intensive workload with relative long-running transactions was used. The Cognos BI server components were implemented with several z/VM guests (distributed installation).

• Online stock trading system:

  The DayTrader benchmark application was used, a transactional workload with response times in the millisecond range. DayTrader is a WebSphere Application Server (WAS) application, implemented from three components (IHS, WAS, DB2). Three instances of this workload were used in parallel.

z/VM was equipped with sufficient memory to ensure that there is no constraint for this resource.

To manage the available CPU resources for the workloads, two strategies were considered:

• One-workload-gets-all: VMRM was set up in a way that one workload gets all of the CPU resources that it requests. This ensures the best possible response times, but bears the risk that the other workloads suffer, if the remaining CPU resources run short.
• A-chance-for-others: The idea is that always some CPU resources are available for the less important workloads. They may perform slower, but time-outs should be avoided. This could be implemented by applying more CPU resources than the preferred workload uses at its maximum load. Or by choosing a balanced VMRM resource management, where the preferred workload gets the largest portion of the resources, but not all.

The system under test was implemented in a way that various load levels of the two different workloads were adjusted, and a Cognos BI workload level below the CPU resource-full condition was chosen. The DayTrader workloads then fill the gap until all CPU resources are almost completely used or even overused. Both strategies and their corresponding VMRM CPU resource goals were tested for various workload levels and combinations.

Notational conventions