15
CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building [email protected]

CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building [email protected]

Embed Size (px)

Citation preview

Page 1: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

CSE 691: Energy-Efficient ComputingLecture 3

SLEEP: full-systemAnshul Gandhi

1307, CS [email protected]

Page 2: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

Number of servers worldwide:

Google: 1-2%

2013:Google: > 1M serversMicrosoft: ~1M serversAmazon: <1M servers

Tencent: +700K servers!

Page 3: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

power_nap paper

Page 4: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu
Page 5: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

US data centers: 100 billion kWh by 2011?? $$

Server utilization: <30%

Idle server power: 60% of peak

Idle periods: ~secondswhy important?

Page 6: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

What does 30% utilization mean?

Page 7: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

Utilization data

Page 8: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

Existing techniques1. Consolidation

2. Sleep states

3. Throttling (DVFS)

Page 9: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

PowerNap1. Simple idea (only 2 states)• Minimize power draw in sleep• Fast transitions

2. Model (power and response time)

3. PowerNap vs DVFS

4. RAILS

Page 10: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile paper

Page 11: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile• PowerNap was NOT implemented

agile took first REAL step towards that

• Static consolidation vs Dynamic consolidation

• How to minimize latency penalties of dynamic consolidation? 3 ideas.

• agile: dynamic virtualization +

PowerNap implementation

Page 12: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile: main problem

Page 13: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile: low-power statesTurbo

C0 P states T states

C1

C1E

C2

C3

S0 C6

S1

S2

S3

G0 S4

G2 S5

G3

Page 14: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile: power vs. latency

Page 15: CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building anshul@cs.stonybrook.edu

agile: dynamic consolidation1. Host power-up2. VM migration3. Host power-down