Elastic Tree: Saving Energy in Data Center Networks
2 likes2,420 views
This document proposes a method called Elastic Tree to reduce network power consumption in data centers. Elastic Tree dynamically adjusts the set of active network elements like links and switches based on changing traffic loads to satisfy demand with minimum power. It consists of three modules: an optimizer to find the minimum power network subset, a routing module to choose paths, and a power control module to toggle network elements. Different optimization methods are explored, including a formal model, greedy bin-packing, and topology-aware heuristic to reduce network power consumption while meeting traffic needs. When applied to an e-commerce data center, Elastic Tree was able to significantly reduce power consumption in the network.
![Introduction
• Networks basically are shared resources
connecting critical IT infrastructure, general
practice is to always leave them ON.
• Most efforts to reduce energy consumption in
Data Centers is focused on servers and cooling,
which account for about 70% of a data center’s
total power budget.
• This presentation focuses on reducing network
power consumption, which consumes 10-20% of
the total power.
-3 billion kWh in 2006[US] by networking elements.](https://image.slidesharecdn.com/elastictreemcclory-120218173025-phpapp01/85/Elastic-Tree-Saving-Energy-in-Data-Center-Networks-2-320.jpg)
![Difference between the topologies
• In a typical DCN[2N tree]. One failure can cut the
effective bisection BW in half. While two failures can
disconnect servers.
• Richer mesh topologies like the fat-tree handle failures
more gracefully; with more components and more
paths, the effect of an individual component failure
becomes manageable.
• This can be used in improving energy efficiency, by
dynamically varying the no. of active network
elements.
• It can be thought as a control knob to tune between
energy efficiency, performance and fault tolerance.](https://image.slidesharecdn.com/elastictreemcclory-120218173025-phpapp01/85/Elastic-Tree-Saving-Energy-in-Data-Center-Networks-4-320.jpg)
![Conclusion
• 3 Algorithms – Model, Greedy and Heuristic
have been examined.
• Applied the above algorithms on E-commerce
data center[Google data center], and found
that power consumption can be reduced.](https://image.slidesharecdn.com/elastictreemcclory-120218173025-phpapp01/85/Elastic-Tree-Saving-Energy-in-Data-Center-Networks-30-320.jpg)
Elastic Tree: Saving Energy in Data Center Networks
- 1. Elastic Tree: Saving Energy in Data Center Networks By Abhishek Sutrave-107907204 Kishen Machamada-107916576
- 2. Introduction • Networks basically are shared resources connecting critical IT infrastructure, general practice is to always leave them ON. • Most efforts to reduce energy consumption in Data Centers is focused on servers and cooling, which account for about 70% of a data center’s total power budget. • This presentation focuses on reducing network power consumption, which consumes 10-20% of the total power. -3 billion kWh in 2006[US] by networking elements.
- 4. Difference between the topologies • In a typical DCN[2N tree]. One failure can cut the effective bisection BW in half. While two failures can disconnect servers. • Richer mesh topologies like the fat-tree handle failures more gracefully; with more components and more paths, the effect of an individual component failure becomes manageable. • This can be used in improving energy efficiency, by dynamically varying the no. of active network elements. • It can be thought as a control knob to tune between energy efficiency, performance and fault tolerance.
- 5. Traffic collected from 292 servers hosting an E-com application over a 5 days period .The traffic peaks during the day and falls at night. Even though traffic varies significantly with time, the associated switches draw a constant power.
- 8. Energy Proportionality • Today’s network elements are not energy proportional – Fixed overheads such as fans, switch chips, and transceivers waste power at low loads. • Maximum efficiency can be realized by a combination of improved components and improved management. • Our strategy is simple: – Turn off the links and switches that we don’t need to keep available only as much networking capacity as required.
- 9. ELASTIC TREE • It is a network-wide power manager, which dynamically adjusts the set of active network elements-links and switches- to satisfy changing data center traffic loads. • It consists of three logical modules – Optimizer – Routing – Power Control
- 10. Elastic Tree Is to find minimum power network subset which Chooses path for all flows satisfies current traffic conditions. Toggles the states of -topology -traffic matrix -power model of each switch -desired fault tolerance properties
- 11. Optimizers • Role-Is to find minimum power network subset which satisfies current traffic conditions. • There are three different methods for computing a minimum power network subset: – Formal Model – Greedy-Bin Packing – Topology-aware Heuristic • Each method achieves different tradeoffs between scalability and optimality. • Methods can be further improved by considering a data center’s traffic history
- 12. Formal Model • Extension of the standard multi-commodity flow (MCF) problem with additional constraints which force flows to be assigned to only active links and switches. • The constraints include link capacity, flow conservation and demand satisfaction. • minimize Σ (Link + Switch Power) • Optimization goal is to minimize the total network power, while satisfying all constraints.
- 13. Formal Model • MCF problem is NP-complete • An instance of the MCF problem can easily be reduced to the Formal Model problem (just set the costs for each link and switch to be 0). • So the Formal Model problem is also NP- complete. • Still scales well for networks with less than 1000 nodes, and supports arbitrary topologies.
- 14. Greedy Bin-Packing • Evaluates possible flow paths from left to right. The flow is assigned to the first path with sufficient capacity. • Repeated for all flows. • Solutions within a bound of optimal are not guaranteed, but in practice, high quality subsets result.
- 30. Conclusion • 3 Algorithms – Model, Greedy and Heuristic have been examined. • Applied the above algorithms on E-commerce data center[Google data center], and found that power consumption can be reduced.
- 31. Questions ?
- 32. THANK YOU