ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

45
ElasticTree: Saving Energy in Data Center Networks 許許許 2013/5/28

Transcript of ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Page 1: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

ElasticTree: Saving Energy in Data Center Networks 許倫愷2013/5/28

Page 2: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

About the paper

Brandon Heller, Srini Seetharaman, Priya Mahadevan, Yiannis Yiakoumis, Puneet Sharma, Sujata Banerjee, Nick McKeown

NSDI’10 (USENIX conference on Networked systems design and implementation)

Citation: 174

16 pages

Page 3: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 4: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 5: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

The motivation

Page 6: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

The motivation

Very inefficient!!

Desired

Page 7: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Why wasting power

Provisioning for peak

Time varying traffic demands

Low efficiency at low loads

Page 8: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

The goal of ElasticTree

Page 9: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

The approach…

Turn off unneeded links and switches

Page 10: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

The challenge

Performance

Fault tolerance

Scalability

Page 11: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 12: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Introduction

What is ElasticTree:

ElasticTree is a system for dynamically adapting the energy consumption of a data center network

• What does it do:

Finding minimum-power network subsets across a range of traffic patterns

Trade-off:

energy efficiency, performance and robustness

Page 13: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Introduction

Page 14: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Data center network

(Traditional) 2N Tree:

One failure can cut the effective bisection bandwidth in half; two failures can disconnect servers

Page 15: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Data center network

Fat tree: SIGCOMM 2008, A Scalable, Commodity Data Center Network Architecture

Page 16: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Data center network

provision for peak workload

Traffic varies daily, weekly, monthly, and yearly.

Page 17: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Energy Proportionality

The strategy: turn off the links and switches that we don’t need

Page 18: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 19: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

ElasticTree

ElasticTree is a system for dynamically adapting the energy consumption of a data center network

Page 20: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

ElasticTree

If 0.2 Gbps of traffic per host ,1 Gbps link…

Page 21: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

ElasticTree

13/20 switches and 28/48 links stay active

ElasticTree reduces network power by 38%

0.2

0.4

0.8

Page 22: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

ElasticTree

The optimizer: find the minimum- power network subset which satisfies current traffic conditions

Page 23: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Optimizer

As traffic conditions change, the optimizer continuously re-computes the optimal network subset

3 approaches:

Formal Model , Greedy Bin-Packing , Topology-aware Heuristic

Page 24: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Optimizer comparison

Page 25: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Formal model

Finding the optimal flow assignment alone is an NP-complete problem for integer flows.

Derived from standard multi-commodity flow (MCF) problem

The model outputs a subset of the original topology, plus the routes taken by each flow to satisfy the traffic matrix

O(n^3.5+)

Page 26: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Greedy Bin-Packing

Strategy: choose the leftmost one with sufficient capacity

O(n^2+)

1G link

Page 27: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Greedy Bin-Packing

1G link

Page 28: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Topo-aware Heuristic

1. does not compute the set of flow routes

2. assumes perfectly divisible flows

=> pack every link to full utilization and reduce TCP bandwidth

=> starter subset

Decoupling power optimization from routing :

=> can be applied alongside any fat tree routing algorithm

Page 29: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Topo-aware Heuristic

An edge switch doesn’t care which aggregation switches are active, but instead, how many are active

Page 30: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Topo-aware Heuristic

Decoupling power optimization from routing

Page 31: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Optimizer comparison

Page 32: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 33: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

How to test

K = 6, fat tree

OpenFlow

Page 34: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis

Traffic pattern:

Near: servers communicate only with other servers through their edge switch

Far: servers communicate only with servers in other pods

Page 35: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis

Random demand:

Individual aggregation/core switches turning on/off

Page 36: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis70% to outside, 30% inside DCN

Different traffic load

Page 37: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis: redundancy

If only the MST is on

=> no redundancy => no fault tolerance

Page 38: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis: redundancy

+MST: additive cost, multiplicative benefit

Page 39: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis: latency

0.250.33 0.5

Need safety margin!!

Ethernet overheads (preamble, inter-frame spacing, and the CRC) cause the egress buffer to fill up Packets either get dropped or significantly delayed

Page 40: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Analysis: latency

Safety margin is the amount of capacity reserved at every link by the optimizer

Traffic overload is the amount each host sends and receives beyond the original traffic matrix

Trade-off between Energy and Performance

Page 41: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Outline

The big picture

Introduction

ElasticTree system

Analysis

Conclusion

Page 42: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Summary

Page 43: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Reference

The paper

The slide (by the author)

A youtube video (by the author, too)

http://www.youtube.com/watch?v=G2_D-CH4tQk

Page 44: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Questions

Page 45: ElasticTree: Saving Energy in Data Center Networks 許倫愷 2013/5/28.

Thank you!