Sarquol Ltd

It is standard wisdom that if you want a database to perform well then you carefully design a set of indices for the tables. Thus, by careful design of the database tables based on their contents and the common queries you can build a database that performs well. The index structures make such a difference that if they are not used on tables with large numbers of rows the performance is unusable. When this approach works then it works really well. There are problems, however, in environments that can lead to real issues, though I won’t go into detail here. Read the rest of this entry »

When planning to deliver a system based on commercial software it is common to start by using manufacturer’s data on the capability of the product to do initial sizing. (e.g. This will support 200 parallel connected users per CPU.) This data often comes from benchmarking and is valid as long as the test conditions are realistically comparable to your production environment. Read the rest of this entry »

A strategy that is commonly employed at present and which often seems strange on first sight is the use of storage servers. This is common, and is often justified for a host of reasons centring around the management of the data on the disks. It is strange, however, that it is also claimed to improve performance, rather than being implemented at the expense of performance. It common to assume that to improve performance you should remove client-server network hops, and network storage goes against this. Can it be true, or is this a myth? Read the rest of this entry »

Once the function, behaviour type and usage pattern data have been provided the model starts to produce useful information based on its estimations. The most important of this is the daily function usage estimation. This information is presented on the “Estimated Function Demand” sheet within the model, and provides as estimate, by year, of the average daily usage for each function. This is then used to provide an estimate of the average and peak operations per second for the system. Read the rest of this entry »

Generic Performance profile graph

When drawing a graph of the response time performance of a system with increasing load it will usually follow a standard profile, namely:

•  

  A very low load increasing the load will actually improve the performance of the system slightly. This is because most systems use various caching mechanisms and if there is too little load none of the caches will have data in.

• As the load increases a near flat region will be experienced, where all requests act as if they have near dedicated resources available to them. Read the rest of this entry »

I have been aware for a while that the IT infrastructure library (ITIL) contains a section on Capacity Management. There is, however, relatively little information that is publicly available about ITIL and the way that it is fitted together. (A summary of ITIL is available here.) As a member of the BCS I found a summary of the foundations of ITIL on the 24×7 books site that membership provides access to. Possibly more interesting, however, is the following article that provides some description of the ITIL Capacity Management, along with a simple example of an ITIL Capacity Plan: ITIL Capacity Management Deep Dive Read the rest of this entry »

 Last year (in this article) I asserted that monitoring of all user operations on an on-going basis was highly valuable. At the time I was not aware of any automated tools to do this task, meaning that solutions would need to be implemented on a per-application basis. In the worst case, this could mean laboriously adding logging statements into application code and then providing analysis mechanism to extract meaningful information from this logging. Read the rest of this entry »

I will skip the title sheet, which is merely descriptive, and start to work with the function definitions. In this case I highly recommend starting with a few key functions that are likely to provide the majority of the key performance behaviour. If this model turns out to be insufficient at a later stage then add more information as the model develops. This means that the level of detail being worked on remains manageable. Read the rest of this entry »

I have received feedback that the performance model would be more useful if it was better documented. The is a tour of the model available (here), but that doesn’t make it clear how to apply the model ina typical usage scenario. I have, therefore, decided to illustrate the use of the model using a fictional example. Read the rest of this entry »

During a recent engagement I needed to use queuing theory based modelling to examine the performance of a system. Searching the Internet I found the JMT open source tools:

http://jmt.sourceforge.net/

Read the rest of this entry »