resource management in cloud computing

10
Resource Management in Cloud Computing Cristian Klein Researcher Department of Computing Science Umeå University Research and Leadership 2016-05-24 1

Upload: cristian-klein

Post on 17-Feb-2017

154 views

Category:

Science


2 download

TRANSCRIPT

Page 1: Resource Management in Cloud Computing

1

Resource Management in Cloud Computing

Cristian KleinResearcher

Department of Computing ScienceUmeå University

Research and Leadership2016-05-24

Page 2: Resource Management in Cloud Computing

2

What is Cloud Computing?

• Internet-scale Software– Requires huge computing and storage capacity

• Cloud computing= Rent computing and storage capacity– I.e., 1,000s of computer servers

Page 3: Resource Management in Cloud Computing

3

Why Cloud Computing?

• Buying cheaply is complicated• Find a nice geographic location– Cheap electricity– Cheap cooling– Fast Internet– Stable political climate

• Build a large building: datacenter• Add cooling• Add power distribution• Add servers

Page 4: Resource Management in Cloud Computing

4

Facebook Datacenter in Luleå

Page 5: Resource Management in Cloud Computing

5

Page 6: Resource Management in Cloud Computing

6

Why Cloud Research?

• “Amazon found every 100ms of latency cost them 1% in sales.”

• “Bing found that a 2 second slowdown changed queries/user by -1.8% and revenue/user by -4.3%.”

• “Google Search found that a 400 millisecond delay resulted in a -0.59% change in searches/user.”

Datacenters must offer performance

Page 7: Resource Management in Cloud Computing

7

Why Cloud Research?

• Datacenters are energy hungry– 1.5% of worldwide electricity– Fastest increase in power usage among all sectors

• Lower energy Lower carbon footprint Lower cost

Datacenters must be energy efficient

Page 8: Resource Management in Cloud Computing

8

Energy Performance Trade-off

✔ Good performance✖ Poor energy efficiency

✖ Poor performance✔ Good energy efficiency

Power off

Page 9: Resource Management in Cloud Computing

10

Challenges

• Large scale of problems– 10s of datacenters– 1 datacenter = up to 100,000 servers– 1 software = up to 1,000 of software components

• Unpredictability– Sudden popularity (flash-crowds)– Hardware failures

Page 10: Resource Management in Cloud Computing

11

Brownout: dealing with capacity shortage

Disable optional computationsE.g., recommender engine

Questions– When to disable such code per software?– How to coordinate multiple software?