A Market Approach for Handling Power Emergencies in Multi-Tenant - - PowerPoint PPT Presentation

▶
a market approach for handling power
SMART_READER_LITE
LIVE PREVIEW

A Market Approach for Handling Power Emergencies in Multi-Tenant - - PowerPoint PPT Presentation

Jul 04, 2023 31 likes •362 views

A Market Approach for Handling Power Emergencies in Multi-Tenant Data Center Mohammad A. Islam, Xiaoqi Ren, Shaolei Ren, Adam Wierman, and Xiaorui Wang What makes up the costs in data centers? 2 Source: A. Greenberg, J. Hamilton, D. A. Maltz,

slide-1
SLIDE 1

A Market Approach for Handling Power Emergencies in Multi-Tenant Data Center

Mohammad A. Islam, Xiaoqi Ren, Shaolei Ren, Adam Wierman, and Xiaorui Wang

slide-2
SLIDE 2

What makes up the costs in data centers?

2

Source: A. Greenberg, J. Hamilton, D. A. Maltz, and P. Patel. 2008. The cost of a cloud: research problems in data center networks. SIGCOMM Comput. Commun. Rev.

slide-3
SLIDE 3

What makes up the costs in data centers?

3

Source: A. Greenberg, J. Hamilton, D. A. Maltz, and P. Patel. 2008. The cost of a cloud: research problems in data center networks. SIGCOMM Comput. Commun. Rev.

Capital Expenditure (CapEx) Operational Expenditure (OpEx)

slide-4
SLIDE 4

4

Grid Generator UPS AC/DC DC/AC

ATS

Cooling PDU PDU

slide-5
SLIDE 5

Infrastructure is really expensive

5

Tenant Tenant Tenant Tenant Grid Generator UPS AC/DC DC/AC

ATS

Cooling PDU PDU

especially for multi-tenant data centers

Owned by tenants Owned by

  • perators
slide-6
SLIDE 6

Infrastructure is really expensive

6

Tenant Grid Generator UPS AC/DC DC/AC

ATS

Cooling PDU PDU Tenant Tenant Tenant

especially for multi-tenant data centers

Owned by tenants Owned by

  • perators

Hyper-scale (e.g. google): 7.8% Enterprise: 53% Multi-tenant: 37%

Percentage of total data center industry electricity usage

Pie Chart from CoreSite’s “One Wilshire” (Photo: CoreSite)

slide-7
SLIDE 7

7

Time Power

Power budget

slide-8
SLIDE 8

8

We need to maximize the utilization!

Time Power

Power budget Unused capacity!

slide-9
SLIDE 9

9

We need to maximize the utilization!

Time Power

Power budget

Power oversubscription moves the line upward!

slide-10
SLIDE 10

Benefits of power oversubscription

10

180 270 360 450

50 100 150 200 250 300 350 400 450 500

10% 15% 20% 30%

Oversubscription Extra Revenue ($/kW/year)

slide-11
SLIDE 11

11

Challenges for power oversubscription

Time Power

Power budget

Power emergency!

25% tenants ≥ 1

downtime (several hours) in 2014.

slide-12
SLIDE 12

12

Challenges for power oversubscription

0% 5% 10% 15% 20% 25% 30% 35% UPS failure/overloading Cyber crime (DDoS) Accidential/human error 2010 2013 2016

Data from report “Cost of Data Center Outages” by Ponemon Institute, Jan 2016.

slide-13
SLIDE 13

How data center operators currently handle emergencies?

13

Before an outage occurs:

Operator Tenants

slide-14
SLIDE 14

How data center operators currently handle emergencies?

14

Operator Tenants

After an outage occurs:

Small rebate (approx. $3/kW/h)

slide-15
SLIDE 15

Consequences of power outage

15

On average, each incident is a million dollar loss

$0 $100,000 $200,000 $300,000 $400,000 $500,000 $600,000 $700,000 $800,000 $900,000 $1,000,000 Patial unplanned outage Total unplanned outage Overall average cost

2010 2013 2016

Data from report “Cost of Data Center Outages” by Ponemon Institute, Jan 2016.

slide-16
SLIDE 16

Consequences of power outage

16

On average, each incident is nearly a million dollar loss

$0 $100,000 $200,000 $300,000 $400,000 $500,000 $600,000 $700,000 $800,000 $900,000 $1,000,000 Patial unplanned outage Total unplanned outage Overall average cost

2010 2013 2016

slide-17
SLIDE 17

We need to handle power emergencies better!

slide-18
SLIDE 18

Natural ideas

  • Lower the IT power usage
  • There’re many power capping solutions
  • DVFS, admission control, load migration, etc. [X. Wang, 2009][H. Lim 2011][X. Fu,

2011][A. Bhattacharya, 2012][D. Wang, 2013]

  • But, operator does NOT control tenants’ servers
  • Even assuming it does, which tenants should reduce power and by how much?
  • Static power reduction contracts
  • Cannot predict power reduction from tenants during an emergency

18

slide-19
SLIDE 19

Natural ideas

  • Lower the IT power usage
  • There’re many power capping solutions
  • DVFS, admission control, load migration, etc. [X. Wang, 2009][H. Lim 2011][X. Fu,

2011][A. Bhattacharya, 2012][D. Wang, 2013]

  • But, operator does NOT control tenants’ servers
  • Even assuming it does, which tenants should reduce power and by how much?
  • Static power reduction contracts
  • Cannot predict power reduction from tenants during an emergency

19

Not applicable to multi-tenant data centers!

slide-20
SLIDE 20

Goal: provide a runtime design to extract power reduction from tenants at minimum performance loss!

slide-21
SLIDE 21

COOP: CO-Ordinated Power management

21

Cut power signal

Operator Tenants

Response Price (Reward) Cut power

slide-22
SLIDE 22

When a power emergency occurs…

  • Two-level capping: high-level UPS and low-level PDU
  • UPS capacity exceeded by 𝐸0
  • PDU capacity exceeded by 𝐸𝑗
  • 𝑂 tenants: each cut power 𝑡𝑗 and has a “performance cost” of 𝑑𝑗(𝑡𝑗)

22

slide-23
SLIDE 23

How to solve it?

  • Centralized control doesn’t work…
  • Market approach

23

slide-24
SLIDE 24

24

Pricing Auction

Operator predicts tenants’ responses; Tenants report nothing to the operator Tenants report all information, i.e., “performance cost” 𝒅𝒋(𝒕𝒋); Operator sets prices accordingly.

Supply function bidding

Tenants report some, but not all, information via supply functions

slide-25
SLIDE 25
  • If you offer me 𝑠, I will reduce power 𝑡𝑠…
  • Extensively studied in the context of electricity markets
  • We choose a parameterized supply function as follows
  • Efficiency [R. Johari, 2011][N. Chen, 2015]

𝒕𝒋(𝒄𝒋, 𝒔) = 𝜺𝒋 − 𝒄𝒋 𝒔

+

Supply function 𝑡 𝑠

25

slide-26
SLIDE 26

Parameterized supply function bidding

26

Cut power 𝑬𝒌 for j=0,1,…M

Operator Tenants

Supply bid 𝒄𝒋 Price 𝒔 Cut power by 𝒕𝒋 = 𝜺𝒋 − 𝒄𝒋

𝒔 +

#1: Operator announces supply function 𝒕𝒋(𝒄𝒋, 𝒔) = 𝜺𝒋 − 𝒄𝒋

𝒔 +

#2: Tenant 𝑗 submits bid 𝑐𝑗 #3: Operator clears market price 𝑠 to satisfy multi-level power capping #4: Power reduction is exercised

slide-27
SLIDE 27

How to bid?

  • Bid based on tenant’s own performance cost, but no need to disclose it

27

slide-28
SLIDE 28

How to set price?

  • Tenants reduce more power when offered higher price
  • Just sufficiently large to make sure that tenants are reducing

enough power

  • If no price is within the expected range (to ensure no profit loss for
  • perator), then enter “failover” mode

28

slide-29
SLIDE 29

Implementation

29

slide-30
SLIDE 30

Evaluation Methodology

  • 5 tenants running different workloads

housed on two clusters

  • DVFS for power reduction

30

slide-31
SLIDE 31

COOP is close to Optimal

  • COOP almost minimizes the performance costs as OPT
  • OPT is an idealized case where the operator dictates tenants’ power reduction

as in an owner-operated data center

  • Settling time: just <1 second

31

1 2 5% 10% 15% OPT COOP

Oversubscription rate Performance cost ($)

slide-32
SLIDE 32

COOP is win-win

32

0% 20% 40% 5% 10% 15%

Saving/Extra Profit Oversubscription

T#1 T#2 T#3 T#4 T#5 Operator

  • Tenants reduce power cost with minimum (temporary) performance impact
  • Operator increases profit by selling capacity to more tenants
slide-33
SLIDE 33

COOP: CO-Ordinated Power management

A market-based approach for handling power emergencies and helping operator better oversubscribe data center capacity

33

Simple, Scalable & Efficient