Online algorithms for geographical load balancing
read more
Citations
A Survey on Mobile Edge Computing: The Communication Perspective
A Survey on Mobile Edge Computing: The Communication Perspective
Data Center Energy Consumption Modeling: A Survey
Dynamic right-sizing for power-proportional data centers
Greening geographical load balancing
References
Live migration of virtual machines
Power provisioning for a warehouse-sized computer
Managing energy and server resources in hosting centers
Cutting the electric bill for internet-scale systems
Power and performance management of virtualized computing environments via lookahead control
Related Papers (5)
Frequently Asked Questions (13)
Q2. How many traces are used to capture the availability of renewable energy?
To capture the availability of solar and wind energy, the authors use traces with 10 minute granularity from [35], [36] for Global Horizontal Irradiance (GHI) scaled to average 1, and power output of a 30kW wind turbine.
Q3. What is the effect of geographical load balancing?
2) The impact of geographical load balancing: Geographical load balancing is much more efficient at using renewable supply than LOCAL because it can route traffic to the data center with higher renewable generation.
Q4. What is the cost function in the proof of theorem 2?
Note that the cost function in the proof of Theorem 2 is applicable to data centers that impose a maximum load on each server (to meet QoS or SLA requirements).
Q5. Why does RHC perform poorly in the heterogeneous setting?
The reason that RHC may perform poorly in the heterogeneous setting is that it may change provisioning due to (wrongly) assuming that the switching cost would get paid off within the prediction window.
Q6. How many servers are in the queue?
The authors model the queueing delays using parallel M/GI/1/Processor Sharing queues with the total load ∑j λt,j,s divided equally among the xt,s active servers, each having service rate µs = 0.2(ms)−1.
Q7. What is the competitive ratio in the heterogeneous setting?
In particular, for any w > 0 the competitive ratio in the heterogeneous setting is at least as large as the competitive ratio in the homogeneous setting with no predictions (w = 0).
Q8. What is the value of the first bracketed term in (14)?
0.But since (ũA, ũB) optimizes (6), the authors havegτ,τ+w((uA, ũB)) − gτ,τ+w((ũA, ũB)) ≥ 0.Thus the first bracketed term in (14) is non-positive, whenceg1,T (ξ τ+1) − g1,T (ξ τ ))≤ gτ,τ+w((ũA, uB)) − gτ,τ+w((uA, uB))≤
Q9. What is the cost of switching a server in a data center?
this operating cost function means that servers in data center s consume a little bit more energy when s is smaller, and they are very inefficient at processing workload of types higher than s.
Q10. What grants were used to support this work?
This work was supported by NSF grants CCF 0830511, CNS 0911041, and CNS 0846025 MURI grant W911NF-08-1-0233, Microsoft Research, the Lee Center for Advanced Networking, and ARC grant FT0991594.
Q11. What is the energy cost of a data center?
One important property of ft,s for their results is e0,s, the minimum cost per timeslot for an active server of type s. i.e., ft,s(xt,s, ·) ≥ e0,sxt,s.The total energy cost of data center s during timeslot t isEt,s = ft,s(xt,s, ∑j λt,j,s). (2)2) Switching cost:
Q12. What is the cheapest way to avoid the excessive cost of processing type t jobs?
Then RHC would start with Λ servers in data center 1 (the cheapest to turn on) at timeslot 1, and then at each t ∈ [2, S] would switch off servers in data center (t− 1) and turn on Λ servers in data center t (the cheapest way to avoid the excessive cost of processing type t jobs using servers in data center s with s < t).
Q13. What is the drawback of solar during the night?
the fact that solar is not available during the night is a significant drawback, which makes wind necessary to power the data centers during night.