z/VM 6.3 Resource Overcommitment

Abstract

This paper provides information about the performance of virtualized environments when operated with constrained resources. A z/VM based system installation runs a set of virtual systems of different types of workloads under Linux for System z. Another aspect analyzed was the efficiency of the z/VM paging subsystem when using EDEV-SCSI based paging devices, as opposed to ECKD based paging devices. It highlights the behavior of the new memory and CPU management introduced with z/VM 6.3.

View
Creation date
November 2014

IBM Cognos Business Intelligence 10.2.1 for Linux on System z - Performance and z/VM Resource Management

Abstract

This white paper is about setup and performance of a Cognos Business Intelligence (BI) server for Linux on System z running under z/VM. The tuning of some important Cognos BI server components is described and how to use the z/VM resource manager (VMRM) to manage constrained CPU resources between competing workloads.

View
Creation date
November 2014

z/VM 6.2 Live Guest Relocation with Linux Middleware - Various Loads

Abstract

This paper discusses the live guest relocation feature of z/VM 6.2 and focuses on how the relocation characteristics change at the workload level and how different workloads (Java™, File system, OLTP Database workloads) are affected by the relocation process.

In addition, it analyzes how a live guest relocation of one workload can remedy a CPU constraint situation.

View
Creation date
July 2013

z/VM 6.2 Live Guest Relocation with Linux Middleware

Abstract

Version 6.2 of the IBM z/VM operating system introduced the concepts of a z/VM cluster and Live Guest Relocation (LGR).

This paper shows the impact of live guest relocation on a Java application and gives recommendations for configuring z/VM systems in order to optimize it.

View
Creation date
February 2013

Optimizing Resource Utilization for Linux under z/VM - Part 1 (Webcast)

Live Virtual Class abstract

This session is the first of two sessions about resource utilization of Linux z/VM guests and means to reduce or control it. This is important especially in regard to resource overcommitment.

The presentation shows the performance impact of memory overcommitment and provides guidelines how to handle it.

It also shows how to set up Large Discontiguous Saved Segments (DCSS), and how memory requirements can be reduced by sharing memory. The current Linux distributions support DCSS larger than 2GB, which significantly improves the administration and handling. The impact of these options on performance and system behavior is analysed in environments with WebSphere Application Servers.

Replay
View
Creation date
May 2012

Optimizing Resource Utilization for Linux under z/VM - Part 2 (Webcast)

Live Virtual Class abstract

This session discusses the resource utilization of Linux z/VM guests, and means to reduce or control it. This is important especially with regard to resource overcommitment in virtualized environments.

The presentation shows how the resource usage of idling guests changes when the guest is paused, and compares Linux and z/VM methods to pause guests.

The presentation also compares the performance impact of multiple WebSphere Application Server Java Virtual Machines running on one guest or running on multiple guests (horizontal versus vertical stacking).

Replay
View
Creation date
May 2012

Methods to pause a z/VM guest: Optimize the resource utilization of idling servers

Abstract

This paper analyzes the behavior of Linux z/VM guests running a WebSphere workload or a relational database (RDB) workload when they are paused and restarted in an idle phase.

The absence of workload might not necessarily imply that a server is inactive from the Hypervisor perspective and its resource are free. The Linux suspend and resume function is compared to the z/VM CP STOP and BEGIN commands.

View
Creation date
September 2011

Large Discontiguous Saved Segments (DCSS) Under Linux

Abstract

A Discontiguous Saved Segment (DCSS) is a z/VM technology used to share memory between z/VM guests. It is a powerful tool that gives z/VM administrators a way to provide high levels of resource sharing between guests, combined with excellent performance.

This paper focuses on three areas of application for a large DCSS: sharing code, sharing read-only data, and using a DCSS as a swap device.

View
Creation date
April 2011

z/VM large memory

Abstract

This report analyzes the results observed with Linux guests running a database server under online transaction processing (OLTP) workload in a z/VM environment using a relatively large amount of main memory (80 GB) and then also overcommitting that memory.

To test the z/VM memory management behavior, the memory was overcommitted up to a factor of two and the impact of the following z/VM memory management features were tested:

These features were expected to improve the overall system performance in cases where the overall z/VM system is constrained for real storage.

View
Creation date
December 2007

z/VM and Xen virtualization performance

Abstract

The paper analyzes the performance of the hypervisors z/VM on IBM System z and Xen on IBM System x when scaling the level of CPU and memory overcommitment. WebSphere Application Server 6.1 environments were scaled on an 8-way server, where each environment was a server triplet consisting of an IBM HTTP server, the application server and a database server. Each server resided on an individual Linux guest. Additionally the paper analyzes the impact of encryption.

The z/VM hypervisor shows significant advantages when scaling into CPU and memory overcommitment in throughput and response times.

View
Creation date
January 2009

z/VM virtualization performance

Abstract

The paper analyzes the performance of the z/VM virtualization features when scaling WebSphere Application Server 6.0.2 environments on an 8-way server.

Each environment is a server triplet consisting of an IBM HTTP server, a WebSphere application server and a DB2 UDB database server, where each server resides on an individual Linux guest. Up to 15 systems are defined running on virtual hardware provided by one System z. The work load and the number of server triplets were scaled, which leads finally to an environment having 2.5 times more virtual CPUs than physical CPUs.

View
Creation date
September 2007

CMM

Abstract

This presentation focuses on Cooperative Memory Management (CMM1) and Collaborative Memory Management (CMM2). They both regulate Linux memory requirements under z/VM. The presentation shows how to activate CMM1 and CMM2, and that both methods improve performance when z/VM hits a system memory constraint.

View
Creation date
March 2009

Team

Please address any comments to the performance team: linux390@de.ibm.com