Browse Prior Art Database

A dispatcher in a queueing theory based workload simulator

IP.com Disclosure Number: IPCOM000029866D
Original Publication Date: 2004-Jul-15
Included in the Prior Art Database: 2004-Jul-15
Document File: 6 page(s) / 37K

Publishing Venue

IBM

Abstract

Disclosed is the implementation of a dispatcher in a workload simulator which is based on queueing theory. This dispatcher links several waiting lines to a set of servers. This enables the workload simulator to perform a simulation on virtualized resources like CPU, network devices or I/O.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 37% of the total text.

Page 1 of 6

A dispatcher in a queueing theory based workload simulator

1. Introduction

In modern IT environments, the element of virtualization becomes more and more important as customers start to exploit the virtualization capabilities of IBM eServer platforms (iSeries, pSeries or zSeries). Depending on the server type, it is possible to create servers that are completely based on virtualized resources like CPU, memory, disk I/O or network.

    Performance analysis in these environments is not straightforward, because a mix of different types of workload is competing for the same hardware resources . In this case, simple models are not helpful, due to the fact that they are simply matching only one type of workload on the hardware resources. Thus they don't allow a sophisticated modelling of various workload mixes. The workload simulator with the dispatcher described in this article, provides a tool to analyze performance problems in these complex, virtualized environments, helping to save time and money.

    The next section outlines the basic concept of a queueing theory based workload simulator. In section 3, a description of the disclosed dispatcher is provided.

2. Introduction to workload simulation

    On a queueing theory based simulation, a computer, or a network of computers is represented by a network of queues. Each queue contains a waiting line and one or more servers. The common graphical representation of these types of queues is shown in Fig.1.

waiting line

servers

waiting line

server

single server queue multi server queue

Fig. 1 In standard queueing theory, a queue consists of one or more servers serving a single waiting line.

    A waiting line is implemented by various one-dimensional array structures (like standard arrays or pointer chains). The Elements in such an array are pointers to data structures called "messages" or "items", or are null-pointers. Fig. 2 displays the relationship between a waiting line and the messages it contains .

[This page contains 2 pictures or other non-text objects]

Page 2 of 6

^nil

^nil

0

waiting mline essages (or item s)

Fig.2 The workload is represented by a set of data structures called 'messages'. The elements in the waiting line are either pointers to these data structures, or they are null-pointers.

    The "messages" or "items" represent parts of the workload, and have properties attached according to the workload they represent, as shown in Fig.4. For example, different workloads can be distinguished by the space they use (size), by the average time they spend in a server (service time) or by their priority.

Fig. 4 Examples of properties that can be associated with a message .

    Servers in this context are active elements that store exactly one message (that is, the pointer to a selected data structure ) for a certain period of time, the service time. This simulates the time the part of the workload represented by the message spends in a CPU of the computer. When this service time has passed, the message...