Power Saving in the Data Center: Auto-Scaling Available Machine Pool Size

[Gerrit91]

In order to save energy in the data center, we should implement an algorithm that can reduce the amount of waiting machines in the available machine pool. The algorithm has to act on every partition individually.

On trigger events like machine allocation / deletion / time interval, the algorithm needs to decide whether to reduce or increase the amount of waiting machines as defined by rules provided by the operator. Machines can be reduced or increased by starting and stopping them through machine IPMI.

[Gerrit91]

Initial idea:

• Add field for partition (waitingPoolSize of type string, can be a number or percentage), unset means disabled
• In FSM
  • On enter waiting state: check whether machines that are waiting and exceed the pool threshold can be shutdown
  • On leave waiting state: check whether machines that are sleeping and undershoot the pool can be started up

Probably following questions still arise:

• When considering race conditions, can it happen that a partition is drowning? (machines are asleep and no machines waiting anymore)
• Maybe it has to be remembered which machines are shutdown by this algorithm (possible conflict with machine resurrection / machine liveliness --> can this be moved into the FSM, too?)