Virtual Melting Temperature: Managing Server Load to Minimize Cooling Overhead with Phase Change Materials

Transcription

1 Virtual Melting Temperature: Managing Server Load to Minimize Cooling Overhead with Phase Change Materials Matt Skach1, Manish Arora2,3, Dean Tullsen3, Lingjia Tang1, Jason Mars1 University of Michigan1 -- Advanced Micro Devices, Inc.2 -- UC San Diego3 ISCA 18

2 Datacenters Huge warehouses full of servers that host the internet and the cloud Facebook Ireland Datacenter Facebook datacenter 2

3 Datacenters Cooling Heat must be removed to prevent: Overheating Thermal downclocking Component failure 3

4 Global Energy Consumption (CIA World Factbook) Energy Consumption Electricity Consumption (TWh/year) 1 China 6,100 2 United States 4,100 3 European Union 3,100 4 India 1,300 5 Russia 1,000 6 Japan Canada 640 4

5 Datacenter Energy Consumption (Avgerinou, 2017) Energy Consumption Electricity Consumption (TWh/year) 1 China 6,100 2 United States 4,100 3 European Union 3,100 Datacenters (global, est.) 1,600 4 India 1,300 5 Russia 1,000 6 Japan Canada 640 5

6 Datacenter Energy Consumption (Avgerinou, 2017) Energy Consumption Electricity Consumption (TWh/year) 1 China 6,100 2 United States 4,100 3 European Union 3,100 Datacenters (global, est.) 1,600 4 India 1,300 5 Russia 1,000 6 Japan 980 Datacenter Cooling (global, est.) 650 Canada

7 Datacenter Cooling Datacenter cooling is very expensive Infrastructure can cost 10s of millions of dollars for large DCs (Kontorinis, 2014) Generally, more power efficient systems are more expensive up front Open Compute cooling system 7

8 Datacenter Workloads Google Search: US Load Diurnal load is problematic Work is uneven Work is distributed Heat is produced when work is done 8

9 Datacenter Cooling Build a big cooling system for peak load Underutilized most of the time Expensive 100% coverage, low utilization 9

10 Datacenter Cooling ctd. Build a big cooling system for peak load Underutilized most of the time Expensive 100% coverage, low utilization 10

11 Datacenter Cooling ctd. Build a big cooling system for peak load Underutilized most of the time Expensive 100% coverage, low utilization Best 50% coverage, maximum utilization 11

12 Thermal Time Shifting (TTS) [ISCA 15] Cooling Load Release heat during off hours Coupled Decoupled Store heat to flatten peak 3am 7am 7pm 12am Time 12

13 Cooling Load Metric of heat that must be removed Datacenter is primarily concerned with IT & support equipment 13

14 A Phase Change Material (PCM) Store energy in a Solid->Liquid phase change Commercial paraffin wax offers the best properties of currently available PCMs (Skach, 2015) 14

15 The problem with passive TTS Thermal Time Shifting: Paraffin has a limited range of melting temperatures Melting temperature cannot be changed Power and temperature profiles vary over lifetime of servers Wikimedia Commons 15

16 Virtual Melting Temperature Datacenters need more flexibility Create a virtual melting temperature separate from the actual melting temperature Microsoft, Wikimedia Commons 16

17 Test Infrastructure 2U High Throughput Server 2-day Google Workload trace divided between 5 datacenter workloads 17

18 Test Methodology 5 common datacenter workloads Web Search Data Caching Video Encoding Virus Scan Clustering Consider datacenter where all are colocated Contention mitigation techniques applied (eg. Bubble Up (Mars, 2011) and Protean Code (Laurenzano, 2014)) 18

19 Baseline: Load Balancing Schedulers Round Robin and Coolest First 19

20 Baseline: Load Balancing Schedulers Round Robin and Coolest First Problem: Average cluster temperature is too low to melt wax

21 Thermal Aware VMT Categorize jobs based upon thermal characteristics Binary classification: Would they melt significant wax in isolation? 21

22 Thermal Aware VMT Grouping Value (GV): Controllable ratio of group size Proportional to hot group size Locate hot jobs together in hot group to melt wax 22

23 Thermal Aware VMT Results Hot Group sized to melt wax during peak hours 23

24 Thermal Aware VMT Results Balance between melting wax too soon and not melting enough wax GV=24: Hot group is too big GV=22: Hot group is just right GV=20: Hot Group is too small 24

25 Thermal Aware VMT Results Balance between melting wax too soon and not melting enough wax GV=24: Hot group is too big GV=22: Hot group is just right GV=20: Hot Group is too small 25

26 Wax Aware VMT Begin with same setup as VMT-TA When wax in hot group is fully melted, expand hot group 26

27 Wax Aware VMT Results Hot Group slightly too small: automatically expands during peak load 27

28 Wax Aware VMT Results Wax expansion preserves significant cooling load reduction GV=24: Hot group is too big GV=22: Hot group is just right GV=20: Hot Group is too small 28

29 Wax Aware VMT Results Wax expansion preserves significant cooling load reduction GV=24: Hot group is too big GV=22: Hot group is just right GV=20: Hot Group is too small 29

30 VMT-TA vs. VMT-WA Both work well at ideal GV VMT-WA offers much more flexibility for unpredictable load Smaller Hot Group Bigger Hot Group 30

31 Summary VMT stores thermal energy when passive TTS alone cannot Reduces maximum cooling load of a diurnal workload Configurable for varying datacenter power and load levels VMT-enabled thermal energy storage can: Reduce cooling system size 12% Or allow up to 14% more servers under the same cooling budget 31

32 Thank you! 32

33 Questions? 33