3.5 MFA Performance Management
Last update PvD

3.5.1 SLA creation

The creation of a Service Level Agreement requires the following steps:

• Define the areas of responsibility (management domains).
  Obviously it defines the responsibilities of the provider (and clear domains favours one-stop-shopping).  But it also defines the responsibilities of the customer (e.g. contact persons, permission to access a site, etc.).
  Define the operating conditions, environment and constrains for the SLA to apply (i.e. under what conditions does the SLA not apply).
• Determine the major (Quality Of) Service parameters for that customer.
  Each service has its characteristics, and therefore specific QoS parameters.  Than, not all parameters are important for the customer's application; do restrict yourselves to the elements which really matter — the rest is wasted effort.
  So, repeat the following steps for each service type:
  1. List the (Quality Of) Service parameters that are to be measured, with the service window (i.e. the time the parameters are active).
     Remark:  excellent service (management) organisations understand their customer's perception of service, and have identified the actions necessary to reinforce good customer service.
  2. Define how each of the parameters will be measured, appreciated and reported.
     Note that above formulation is not abstract at all:  it is not 'what each parameter means' but the mechanism, leaving less room for misunderstanding and debate.  Presentation of abstract values is detrimental to their usefulness.
     Also, the presentation form is important for perception and appreciation; the more it addresses the customer's purposes, the better.  A particular parameter can be presented as a single value, or in a more elaborate form.  For example, the parameter 'dependability' (availability) can be presented as a single percentage, or more elaborate like 'total outage', 'number of outages', 'average outage' (=Mean time to Restore), 'longest outage', etc.  Also, thresholds can be relative (e.g. 'Errored Packets: <1% of Total Packets') or staged (e.g. 'Delay: >90% within 0.5 seconds, >95% within 1 second').
     Note that the length of an evaluation interval is important:  'unavailability' of 0.1% over 1 minute is probably unnoticed, whereas over a year it is a full working day which could have serious impact on the customer's business.  Use reasonable intervals.
     Specify the maximum number of customer transactions (or any other volume) which are assumed for the SLA.
• Define the process by which information will be communicated, reviewed, and issues resolved.
  This process is often much more important than (service level) thresholds, and when carried out well, hardly require thresholds and fines.  Often there are changes in the environment which require updating of the SLA (new standards, more sites, more traffic, etc.).
  It is also argued that for the SLA to be taken serious, there should be real penalties (but the level of the penalties should be reasonable).  Also, for multi-service SLAs there can be a service interdependency which should be taken into account.

Note that SLAs are not adversarial — there is no point in setting unrealistic targets.

For popular services, one commonly has 3 service levels ('Classes of Service'):

Standard

applicable for normal users in most cases;
a restricted set of parameters, some modest thresholds, hardly fines, standard price, probably no reporting:  the default.

Premium

when the service is essential for the customer's business;
elaborate set of parameters, critical thresholds, serious fines, significantly increased price, standard reporting, available off-the-shelve.

Custom

specially adapted to the customer's need;
extensive set of parameters, probably including some special values, critical thresholds, serious fines, special price;  probably requiring special measures in resources and/or management systems.

Further references:

HP (Operations and Performance Management) White Papers (formerly HP OpenView)

Various articles on setting up SLAs and corresponding problems.  Meant for the IT-area (i.e. performance of computer services), but of more general applicability and valuable.

Search the web for Writing a Service Level Agreement

Advises on writing SLAs.

=O=