Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
A Spot Capacity Market to Increase Power Infrastructure Utilization in
Multi-Tenant Data Centers
Mohammad A. Islam, Xiaoqi Ren, Shaolei Ren, and Adam Wierman
This work was supported in part by the U.S. NSF under grants CNS-1551661, CNS-1565474, CNS-1518941, CPS154471, ECCS-1610471, and AitF-1637598.
Multi-tenant data centers
Utility
(Primary)
Generator
UPS
ATS
P
D
U
P
D
U
2
Managed by
operator
Multi-tenant data centers
Utility
(Primary)
Generator
UPS
ATS
P
D
U
P
D
U
2
Managed by
individual tenants Managed by
operator
Multi-tenant data centers
Utility
(Primary)
Generator
UPS
ATS
P
D
U
P
D
U
2
3
Multi-tenant data centers are everywhere
3
Google, Amazon, MS, Fb…
:7.8%
Enterprise:
53%
Multi-tenant:
37%
Multi-tenant data centers are everywhere
4
Who are using multi-tenant data centers?
4
25% of Apple’s servers ate in multi-tenant data centers
Who are using multi-tenant data centers?
4
25% of Apple’s servers ate in multi-tenant data centers
Who are using multi-tenant data centers?
4
25% of Apple’s servers ate in multi-tenant data centers
Who are using multi-tenant data centers?
Data center costs breakdown
5
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.
Data center costs breakdown
5
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.
Data center costs breakdown
5
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)
Data center costs breakdown
5
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)
Data center costs breakdown
5
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)
Cost of infrastructure
Utility
(Primary)
Generator
UPS
ATS
P
D
U
P
D
U
6
Underutilization in data centers
7
Underutilization in data centers
7
Underutilization in data centers
7
Increase infrastructure utilization
8
Exploit the “spot capacity”
Increase infrastructure utilization
8
Some inspirations
• “Power routing” in ASPLOS’10 and “soft fuse” in EuroSys’09
9
Some inspirations
• “Power routing” in ASPLOS’10 and “soft fuse” in EuroSys’09
• “Spot instances” from Amazon
9
Some inspirations
• “Power routing” in ASPLOS’10 and “soft fuse” in EuroSys’09
• “Spot instances” from Amazon
• “Preemptible VM” from Google Cloud
9
Spot capacity in multi-tenant data centers
10
No centralized control
Spot capacity in multi-tenant data centers
10
No centralized control Power routing,…
Spot capacity in multi-tenant data centers
10
A market for spot capacity
No centralized control Power routing,…
Spot capacity in multi-tenant data centers
10
A market for spot capacity
No centralized control Power routing,…
Tenants buy spot capacity from
the data center operator
Spot capacity in multi-tenant data centers
10
Spot capacity in multi-tenant data centers
• Flexibility for cost conscious tenants
11
Spot capacity in multi-tenant data centers
• Flexibility for cost conscious tenants
Peak-based
subscription
11
Spot capacity in multi-tenant data centers
• Flexibility for cost conscious tenants
Peak-based
subscription Conservative
subscription
11
Spot capacity in multi-tenant data centers
• Flexibility for cost conscious tenants
Peak-based
subscription Conservative
subscription Spot capacity
11
• Tenants: • tenants with insufficient capacity reservations can temporarily
process its workloads without power capping (or cap power less frequently/aggressively than it would otherwise).
Spot capacity in multi-tenant data centers
12
• Tenants: • tenants with insufficient capacity reservations can temporarily
process its workloads without power capping (or cap power less frequently/aggressively than it would otherwise).
• Operator: • Revenue from guaranteed capacity: not affected
• Extra revenue from spot capacity
Spot capacity in multi-tenant data centers
12
• Tenants: • tenants with insufficient capacity reservations can temporarily
process its workloads without power capping (or cap power less frequently/aggressively than it would otherwise).
• Operator: • Revenue from guaranteed capacity: not affected
• Extra revenue from spot capacity
Spot capacity market is a win-win for both
tenants and operator
Spot capacity in multi-tenant data centers
12
Challenges
• Spot capacity is limited and intermittent
13
Challenges
• Spot capacity is limited and intermittent
• Tenants’ spot capacity need is dynamic and invisible to the data center operator
13
Challenges
• Spot capacity is limited and intermittent
• Tenants’ spot capacity need is dynamic and invisible to the data center operator
• Infrastructure constraints require fine granularity in spot capacity allocation (e.g., rack level)
13
Goal: A scalable and runtime design
for spot capacity allocation
14
Problem formulation
• Goal: operator profit maximization
15
Rack level demand
Problem formulation
• Goal: operator profit maximization
15
Price of spot
capacity
Rack level demand
Problem formulation
• Goal: operator profit maximization
15
Infrastructure constraints
Price of spot
capacity
Rack level demand
Problem formulation
• Goal: operator profit maximization
15
How to solve it?
16
Unknown How to solve it?
16
Unknown How to solve it?
• Soliciting the demand curve privacy and overhead
16
Unknown How to solve it?
• Soliciting the demand curve privacy and overhead
• Pre-set price low level demand prediction
16
Unknown How to solve it?
• Soliciting the demand curve privacy and overhead
• Pre-set price low level demand prediction
• Market approach an in-between solution
16
SpotDC: spot capacity management
17
Operator Tenants
SpotDC: spot capacity management
17
Spot capacity
predictions
Operator Tenants
SpotDC: spot capacity management
17
Spot capacity
predictions
Operator Tenants
Response (bids)
SpotDC: spot capacity management
17
Spot capacity
predictions
Operator Tenants
Response (bids)
Price and actual spot
power allocation
SpotDC: spot capacity management
17
Spot capacity
predictions
Operator Tenants
Response (bids)
Price and actual spot
power allocation
Gain spot
power
Timings in SpotDC
18
Demand bidding
• A piece-wise-linear bid
19
Demand bidding
• A piece-wise-linear bid
• Tenants only submit four parameters
19
Demand bidding
• A piece-wise-linear bid
• Tenants only submit four parameters
• Captures tenants’ demand elasticity
19
Spot capacity prediction
• Available spot capacity prediction: max - predicted • UPS and PDU level predictions: Use previous time slot usage
as references.
20
Spot capacity prediction
• Available spot capacity prediction: max - predicted • UPS and PDU level predictions: Use previous time slot usage
as references.
20
Spot capacity prediction
• Available spot capacity prediction: max - predicted • UPS and PDU level predictions: Use previous time slot usage
as references.
20
Less than ± 2.5% change
99% of the time
Evaluation methodology
• 10 tenants with sprinting (delay sensitive) and opportunistic (delay tolerance) workloads
• Using Dynamic voltage and frequency scaling (DVFS) for power scaling.
21
Evaluation methodology
• 10 tenants with sprinting (delay sensitive) and opportunistic (delay tolerance) workloads
• Using Dynamic voltage and frequency scaling (DVFS) for power scaling.
21
Evaluation methodology
• 10 tenants with sprinting (delay sensitive) and opportunistic (delay tolerance) workloads
• Using Dynamic voltage and frequency scaling (DVFS) for power scaling.
21
Performance evaluation
22
Performance evaluation
22
Sprinting tenants drive up the price
Performance evaluation
23
Performance evaluation
23
Sprinting tenants avoid
SLO violations
Performance evaluation
23
Sprinting tenants avoid
SLO violations Opportunistic tenants
gain throughput boost
Tenants’ benefit from SpotDC
24
Tenants’ benefit from SpotDC
24
Performance boosts
with SpotDC
Operator’s extra profit
25
Operator’s extra profit
25
SpotDC is close to optimal
allocation with full information
SpotDC: Spot capacity management
A market-based approach for providing spot capacity to tenants and helping operator further increase data center utilization
26
SpotDC: Spot capacity management
A market-based approach for providing spot capacity to tenants and helping operator further increase data center utilization
26
Simple, Scalable & Efficient