In AgentScape, we have experimented with penalties including:

- simple notification (alerting the participants, but taking no other action)

- canceling service,

- canceling payment,

- reducing payment (by a percentage specified in the SLA) but continuing service, and 

- charging a fixed fine per incident (a set amount specified in the SLA) but continuing service.

We are also looking at the option of requesting renegotiation of the SLA, such as lower (more feasible) QoS for a lower payment. Another option we are looking at is integrating a reputation system that builds profiles of each participant in the system based on the number of successful or violated agreements. These are very interesting options, but we have not yet fully developed them.

I've also attached some papers on these topics.

Kind regards,

Kassidy