1
Capping the Brown Energy Consumption of Internet
Services at Low Cost by K. Le, R. Bianchini, T.
D. Nguyen, O. Bilgir, and M. Martonosi An
Algorithmic Perspective

• David Shmoys
• Cornell

2
Goals of the Paper

• Provide optimization-based model to balance the
  usage of brown energy vs. green energy in
  assigning customers to data centers
• Key issues - cap-and-trade framework above
  annual baseline brown energy cap
• - amortized requirement on response time - user
  requests are of multiple types - capacity of
  centers - minimize total energy cost

3
One Epoch t1,,T
n(j,t)
U(i)
Centers i
Request Types j
4
One Epoch t1,,T
Brown cost
n(j,t)
U(i)
Centers i
Request Types j
5
One Epoch t1,,T
Green cost
n(j,t)
U(i)
Centers i
Request Types j
6
One Epoch t1,,T

• There is also a baseline energy cost for each
  center depending on
• Brown vs. Green
• which center is load Load(i)

n(j,t)
U(i)
And also an offset cost C(i,Load(i))
Centers i
Request Types j
7
Decisions, Decisions, Decisions

• Compute mix of green vs brown power available at
  each center (fixed for entire planning period
  1,,T)

8
One Epoch t1,,T
n(j,t)
U(i)
Centers i
Request Types j
9
Decisions, Decisions, Decisions

• Compute mix of green vs brown power available at
  each center (fixed for entire planning period
  1,,T)
• Compute assignment of requests to centers
  (assuming proportionate energy type transmission
  cost?)
• Service level requirement (at least p requests
  handled with delay at most D)?
• Total brown energy cap over all centers

10
Obvious Algorithmic Playground

• Online algorithms Do requests come one at a
  time? Or rather epoch by epoch? Apply viewpoint
  of competitive analysis?
• Two-stage stochastic model use distributional
  model on arrivals to determine brown/green
  mixture rates and then minimize expected cost
  over planning horizon (say willing to adjust
  mixture each epoch)
• Multi-stage stochastic models?