Heat-aware Loadbalancing - Is it a thing? Lukas Ifflnder, Norbert - - PowerPoint PPT Presentation

heat aware loadbalancing is it a thing
SMART_READER_LITE
LIVE PREVIEW

Heat-aware Loadbalancing - Is it a thing? Lukas Ifflnder, Norbert - - PowerPoint PPT Presentation

Heat-aware Loadbalancing - Is it a thing? Lukas Ifflnder, Norbert Schmitt , Andreas Knapp, and Samuel Kounev Chair of Software Engineering Universitt Wrzburg November 13, 2020 11th Symposium on Software Performance 2020, 12.13.


slide-1
SLIDE 1

https://se.informatik.uni-wuerzburg.de/

Heat-aware Loadbalancing - Is it a thing?

Lukas Iffländer, Norbert Schmitt, Andreas Knapp, and Samuel Kounev Chair of Software Engineering – Universität Würzburg November 13, 2020 11th Symposium on Software Performance 2020, 12.–13. November 2020, Leipzig, Germany

slide-2
SLIDE 2

2

  • Increasing amount of cloud users and services

Motivation

[1]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-3
SLIDE 3

3

  • Increasing amount of cloud users and services
  • Increasing requirements on computing power by more traffic

 Increasing computing power needed

Motivation

[1] [2]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-4
SLIDE 4

4

  • Increasing amount of cloud users and services
  • Increasing requirements on computing power by more traffic

 Increasing computing power needed

  • Solution 1: More servers? ↯

Motivation

[1] [2]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-5
SLIDE 5

5

  • Increasing amount of cloud users and services
  • Increasing requirements on computing power by more traffic

 Increasing computing power needed

  • Solution 1: More servers? ↯
  • Solution 2: Better/Full usage of existing computing power

make use of unallocated CPU boost power

Motivation

[1] [2]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-6
SLIDE 6

6

Goals

1.

Is it possible to exceed permanently and feasibly the clock rate with host relay?

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-7
SLIDE 7

7

Goals

1.

Is it possible to exceed permanently and feasibly the clock rate with host relay?

2.

What is the effect on the computing power compared to normal behavior?

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-8
SLIDE 8

8

Goals

1.

Is it possible to exceed permanently and feasibly the clock rate with host relay?

2.

What is the effect on the computing power compared to normal behavior?

3.

What is the effect on the energy consumption compared to normal behavior?

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-9
SLIDE 9

9

Contributions

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

A first approach to heat-aware load-balancing

1.

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-10
SLIDE 10

10

Contributions

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

A first approach to heat-aware load-balancing A prototype implementation with SDN and an initial evaluation

1. 2.

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-11
SLIDE 11

11

Approach - Boosting

  • Intel Turbo Boost:
  • CPU-Boost ≠ Overclocking in general
  • CPU Boost respects specifications of hardware
  • Max. n° of boosted cores and max. frequency is limited by headroom
  • OS requests more computing power  Boost
  • Example Intel:

– Stepwise increasing clock with 133,33 MHz – Time in boost state depends on:

  • Temperature
  • Energy
  • Other common used boost: AMD Turbo Core / AMD Precision Boost
  • Boosts are usually disabled in related work and HPC environments due to unpredictability

[3]

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-12
SLIDE 12

12

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-13
SLIDE 13

13

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-14
SLIDE 14

14

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-15
SLIDE 15

15

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-16
SLIDE 16

16

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-17
SLIDE 17

17

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-18
SLIDE 18

18

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-19
SLIDE 19

19

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-20
SLIDE 20

20

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-21
SLIDE 21

21

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-22
SLIDE 22

22

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Definition: Not max. boosted anymore  time to switch

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-23
SLIDE 23

23

Approach

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Definition: Not max. boosted anymore  time to switch

  • Assumptions
  • The workload puts enough

stress on a server that it will boost

  • The server is under stress that

the migrated workload will only be adequately serviced when boosted

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-24
SLIDE 24

24

Prototype Implementation

  • Client: HTTP Load Generator

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

[4]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-25
SLIDE 25

25

Prototype Implementation

  • Client: HTTP Load Generator
  • Worker:
  • Workload: Bungee-LU-Servelet
  • Data collector: Telegraf

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

[4] [5] [6]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-26
SLIDE 26

26

Prototype Implementation

  • Client: HTTP Load Generator
  • Worker:
  • Workload: Bungee-LU-Servelet
  • Data collector: Telegraf
  • Monitor
  • Database: InfluxDB
  • Dashboard: Chronograf
  • SDN-Controller framework: Ryu

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

[4] [5] [6] [6] [6] [7]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-27
SLIDE 27

27

Prototype Implementation

  • Client: HTTP Load Generator
  • Worker:
  • Workload: Bungee-LU-Servelet
  • Data collector: Telegraf
  • Monitor
  • Database: InfluxDB
  • Dashboard: Chronograf
  • SDN-Controller framework: Ryu
  • Switch: Software-Switch with Open vSwitch

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

[4] [5] [6] [6] [6] [8] [7]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-28
SLIDE 28

28

Prototype Implementation

  • Client: HTTP Load Generator
  • Worker:
  • Workload: Bungee-LU-Servelet
  • Data collector: Telegraf
  • Monitor
  • Database: InfluxDB
  • Dashboard: Chronograf
  • SDN-Controller framework: Ryu
  • Switch: Software-Switch with Open vSwitch
  • Testbed with three servers, SDN, and controller

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

[4] [5] [6] [6] [6] [8] [7]

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-29
SLIDE 29

29

Evaluation

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

  • Three evaluated scenarios:
  • A. Highly overbooked
  • B. Medium overbooked
  • C. Not overbooked

A

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-30
SLIDE 30

30

Evaluation

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

A B

  • Three evaluated scenarios:
  • A. Highly overbooked
  • B. Medium overbooked
  • C. Not overbooked

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-31
SLIDE 31

31

Evaluation

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

A C

  • Three evaluated scenarios:
  • A. Highly overbooked
  • B. Medium overbooked
  • C. Not overbooked

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-32
SLIDE 32

32

Evaluation

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

B C

  • Three evaluated scenarios:
  • A. Highly overbooked
  • B. Medium overbooked
  • C. Not overbooked

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-33
SLIDE 33

33

Preliminary Measurements

Foundations > Reasearch Question > Approach > Evaluation > Conclusion

Switched Unswitched Average response time in seconds 0.85 1.83 Number of requests sent/reveiced/lost 1310 / 1310 / 0 1310 / 1304 / 6 Temperature max. r04/r05/r06 in °C 61.99 / 61.13 / 61.35 69.76 / * / * Temperature average r04/r05/r06 in °C 59.29 / 58.81 / 59.13 65.96 / * / * Average CPU frequency in MHz r04/r05/r06 2459.86 / 2147.86 / 2333.61 3518.44 / * / * Average CPU maximum in MHz 3670.24 3580.32 Average power consumption (idle: 98.19W) in Watt 135.54 147.98 Additional power consumption in total and percent 37.35 / 38.03% 49.79 / 50.71%

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-34
SLIDE 34

34

Achievements

  • Working heat-aware and boost-oriented load

rotation

  • Higher average of max. CPU frequency

Basics > Reasearch Question > Approach > Evaluation > Conclusion

Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-35
SLIDE 35

35

Achievements

  • Working heat-aware and boost-oriented load

rotation

  • Higher average of max. CPU frequency

Open Tasks

  • Include performance metrices
  • Throughput
  • Latency
  • Eliminate simplifications
  • Stateful migration
  • Tests with additional workloads
  • Extend to heterogenous systems
  • Power and temperature evaluation

Basics > Reasearch Question > Approach > Evaluation > Conclusion

Conclusion

Motivation Goals & Contributions Approach Evaluation Conclusion

slide-36
SLIDE 36

Thank You!

https://se.informatik.uni-wuerzburg.de/

slide-37
SLIDE 37

37

References

[1] https://3s81si1s5ygj3mzby34dq6qf-wpengine.netdna-ssl.com/wp-content/uploads/2018/02/aws-q4-2017-revenue-income.jpg [2] https://www.grandviewresearch.com/static/img/research/iot-market.png [3] https://www.intel.com/content/dam/doc/white-paper/intel-microarchitecture-white-paper.pdf [4] https://se.informatik.uni-wuerzburg.de/fileadmin/10030200/user_upload/logos/HTTPlogo.png [5] https://se.informatik.uni-wuerzburg.de/fileadmin/_processed_/9/9/csm_csm_BUNGEE_02_837dac3f07_dddb202d42.png [6] http://influxdata.github.io/design.influxdata.com/branding-docs/img/research/tick-stack-1.jpg [7] https://osrg.github.io/ryu/css/images/LogoSet02.png [8] https://raw.githubusercontent.com/openvswitch/openvswitch.github.io/master/assets/vswitch.png Motivation Goals & Contributions Approach Evaluation Conclusion