Capacity and performance engineers often want to predict the response times of applications over ICT systems. Modelling techniques can help capacity and performance engineers achieve this. However, there are different modelling techniques that can be used. ITIL defines the main modelling approaches as trending, baselining, analytical and simulation. (A word of warning, the ITIL definitions are not definitive, for example baselining isnt a modelling technique as such.) This article lists some of the reasons that you might want to consider using discrete event simulation modelling to predict the performance of your ICT system.
Building simulation models generally takes longer than building analytical models; because of this the author recommends simulation modelling projects should only be carried out when absolutely necessary. Managers and capacity planners sometimes find it difficult to determine if simulation modelling is really needed and whether the additional time and cost can be justified when compared to other modelling techniques.
The author has put together a list of questions to ask when deciding whether the use of simulation modelling is required. This list acts as a basic guide and is not exhaustive.
1) Are there serious implications of not meeting SLA response times?
If the system is an internal IT system which isnt running critical applications then it is difficult to justify the use of modelling. If the system is large and the SLA penalty points are punitive then you will need to do some sort of modelling. However, this by itself isnt enough to justify the building of a simulation model. Questions 2. to 5. need to be addressed before you consider building a simulation model.
2) Is there any complex behaviour designed into the ICT system?
This describes the situation where aspects of the ICT system design that may become unpredictable or unstable under certain conditions. Some examples of this are retry mechanisms and control mechanisms (e.g. use of leaky buckets in Automatic Call Gapping for telephony systems). This type of complexity may not require detailed modelling within a simulation model (i.e. the complex behaviour can be modelled at a black-box level). This is a key question that needs to be addressed at the start of any major modelling project.
3) Are there any mixed workloads over the same infrastructure?
This is where time-sensitive workload and non time-sensitive workload is handled by the same infrastructure. You may want to determine the impact of non time-sensitive workload on time sensitive workload under different conditions.This type of behaviour can be observed using a discrete event simulation model of the system.
4) Are there a large number of source or destination systems?
There can be synchronization issues that occur from many sources sending transactions to a central system and vice versa. These are difficult to ascertain in a test environment because test environments tend to be substantially scaled down versions of the live environments.It is possible to use discrete-event simulation models to observe this type of behaviour, because modelling doesnt have the scalability limitations of test environments.
5) Do you need a finely-tuned system?
ICT systems often have a range of performance parameters (e.g. thread pools) that have a significant impact on the performance and stability of the system. These parameters are often configured using best guesses. This approach is fine when a finely tuned system is not needed. However, a more sophisticated approach is required for optimal performance tuning of the systems. This typically involves doing some level of modelling.If optimal performance is needed then one approach involves building a detailed simulation model of the ICT system that models the tuneable performance parameters. There is a large effort involved in building and maintaining this type of model. However, the advantage is that the model can be used to reduce the amount of guesswork (and time) that is normally carried out when performance tuning an ICT system.
If the answer is yes to both 1. and one or more of 2. to 5. then you should consider using a discrete event simulation tool.
Discover how capacity management can support business growth and maximise revenue potential - download here.
