17-811 Self-Healing Systems: Class Discussion Summary

Summary of Class Discussion for April 2 by Vahe Poladian

1.  Rainbow demo and discussion:

- Q: what if the system has its own adaptation:

        - we want to separate concerns, and see if we can keep the strategy of repair separate from the system itself,

- Jung Soo: how to handle conflicting adaptations, e.g. Rainbow wants to do A and the system itself wants to do B.

        - Bradley: this is a reasearch question,
 
- Q: how does Taylor repair stragey work?

        - currently, it is coded as Java objects, but Owen working on abstracting this away,
 
        - also, possible to plug-in other repair mechanisms,

- Q: is there a user in the loop:

        - yes, to some extent.  The constraints are specified by the user.  See Joao's work,
 

2.  Owen's discussion:

   i. Active Events paper:
 
                - monitoring mechanism: probes to get raw info, gauges to combine that info at a higher level,
 
                - detection: smartevent to show bad conditions,
 
                - reslution: ?
 
                - adaptation: deploy worklets,
 
        - have a separation of concern between identifying repair plan and actulaly fixing,
 
                - at the higher level you might want to have a high level plan as to how to repair,
 
                - at the lower level, you want to translate into simple fixes that can be executed against the running system,
 
                ( similar to probe - gauge symbiosis in rainbow, but at the repair level, not at the monitoring level),
 
        - Q: what does a worklet exactly do?
 
        - at one extreme, the higher level adaptation can be very no-detailed, such as: "fix-yourself" command to the lower level,
 
   ii.  Adaptive mirroring:
 
        - Q: what is the model?
 
                - workflow? where does it come from? is adaptation done at that level?
 
        - Q: who does discovery of services and replacement of failed services?