Thursday, May 27, 2010

Bias of the week : So what computer should I buy

I visited a customer this morning. I have done a lot with them over the years, they provide a good service to their users and are very nice people. They wanted help in deciding which product to buy as part of an upgrade programme. I don't give advice on what to buy. I see my role is helping them understand and remove risks, in addition to technical performance analysis.

Lets take an example of a case deciding how suitable a computer systems is for a particular purpose. If you are running a web server, the right choice might be different for modeling the start of the Universe from running a low latency equities trading system. I have heard that in a different world people use computers for things like writing documents, read email and download porn, but that is not the world of computing I live in.

The risk in question is that a particular architecture won't run a particular workload well, so the customer I saw this morning quite rightly reduced that risk by calling in an expert in computer systems performance. In the end they could not find one, so they had to make do with me. There is still some of this risk left on the table as some days may have higher or different load which I did not see in my monitoring for 30 minutes at a time of normal load. They could manage this risk by setting up a test suite and load testing or doing some long term monitoring. Both of which would probably be greater than the cost of the systems they are considering buying to replace the existing systems.

I think this is a case of a bias that there is a correct choice and I don't think there is. There are just different sets of risks which may manifest themselves depending on sets of events which we can't know the probability of them happening without great effort and cost.

I think this is a case of Information Bias mixed with Need for Closure by trying to get answers to questions that are expensive or too complex to answer within their environment.

In this case they did the right thing, looked around and tried to find as impartial an expert as they could (as mentioned they did not find one, so ended up with me) and sought to ask the question "what do you recommend we buy". I very much hope I did not favor either of the 3 options, but helped them both in process and content to weight up the process and cons of each options, this risks of each option and how tolerant to each risk they are given the cost of reducing the probability or impact of that risk. At the end of the day, apart from running and interpreting commands like prstat, mpstat (No DTrace, Solaris 9), most of what I did was Rational Process and that is the bit that really added value.

No comments:

Post a Comment