Information Technology Reference
In-Depth Information
19.3 Example KPI: Machine Allocation
Suppose we are developing a KPI to assess the quality of the process by which virtual ma-
chines (VMs) are created. This KPI may apply to a public cloud service provider, it may
be geared toward a team that creates VMs for an internal cloud service within a large com-
pany, or perhaps it is intended just to assess the process that a team uses for creating its
own VMs.
19.3.1 The First Pass
WebeginwithStep1,EnvisiontheIdeal.Inanidealworld,peoplewouldgettheVMsthey
want as soon as they request them. There would be no delay.
In Step 2, Quantify Distance to the Ideal, we simply measure the duration of time from
the request to when the VM is created.
In Step 3, Imagine How Behavior Will Change, we brainstorm all the ways that people
could behave but still match the incentive. We foresee many challenges.
One challenge is that people could get very creative about the definition of “start time.”
It could be the time when the request is received from the user, or it could be when the cre-
ation process actually begins. If requests are made by creating tickets in a helpdesk request
system, the delay before someone processes the ticket could be very large. If the requests
come in via a web portal or API, they may be queued up and processed sequentially. If that
wait time is not included in the metric, it would make the team look good, but would not
truly reflect the service users receive.
A more realistic indication comes from measuring the end-to-end result from the cus-
tomer's perspective. Doing so might inspire the team to move from a ticket-based request
system to a self-service portal or API. This would not only replace the human process of
re-entering data from a ticket, but also ensure that all the information needed to complete
theprocessiscollectedattheverybeginning.Thusitavoidstheback-and-forthcommunic-
ationthatmightberequiredtocollectinformationtheuserforgotordidn'tknowtoinclude
in the ticket.
People could also demonstrate creativity in how they define when the request is “com-
plete.” The incentive is to interpret completion time as soon as possible. Is the VM created
whentheVMManagerallocatesRAM,disk,andotherresourcestocreatetheemptyvirtual
machine, or when that virtual machine has loaded its operating system? If a different team
is responsible for the OS installation mechanism, can we be “complete” once the OS in-
stallation starts, whether or not it finishes successfully? To the user, a failed, half-installed
VM is worthless, but a mistated KPI might permit it.
Search WWH ::




Custom Search