![Page 1: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/1.jpg)
© 2009 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice
Data Mining for Sustainable Data Centers
Manish Marwah Senior Research Scientist Sustainable Ecosystem Research Group Hewlett Packard Laboratories [email protected]
![Page 2: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/2.jpg)
2 3 May 2012
Motivation Industry challenge: Create technologies, IT infrastructure and business models for the low-carbon economy
2%
Aviation Total carbon emissions 2%
IT industry
The footprint of IT will need to be reduced quite significantly in a low-carbon economy.
![Page 3: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/3.jpg)
3 3 May 2012
Motivation Industry challenge: Create technologies, IT infrastructure and business models for the low-carbon economy
98% The rest of the
global economy Total carbon emissions
2% IT industry
IT must play a central role in addressing the global sustainability challenge.
![Page 4: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/4.jpg)
Sustainability
“sustainable development is development that meets the needs of the present without compromising the ability of future generations to meet their own needs” the Brundtland Commission of the United Nations, 1987
4 3 May 2012
![Page 5: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/5.jpg)
5 3 May 2012
Sustainability What do I mean by “sustainability”?
Social (“People”)
Economic (“Profit”)
Environmental (“Planet”)
Risk: Ecological Damage
Sustainable
Risk: Limited Adoption
Risk: Commercially unfeasibility
Figure Credit: A. Agogino, UC Berkeley
![Page 6: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/6.jpg)
6 3 May 2012
Environmental Sustainability
• Life Cycle View
![Page 7: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/7.jpg)
7 3 May 2012
Sustainable Data Centers Lifecycle Assessment
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Han
dhel
d
Not
eboo
k
Des
kto
p
Blad
eSe
rver
Dat
aCe
nter
Frac
tion
of L
ifec
ycle
Ene
rgy
OperationalEmbedded
Results are illustrative only. Actual footprint may differ.
Ref: IEEE Computer 2009
![Page 8: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/8.jpg)
8 © Copyright 2011 Hewlett-Packard Company Chandrakant D. Patel; [email protected]
Cloud Data Center Supply and Demand Side
Chilled Water loop
Cooling Tower loop
CHILLER
Warm Water
Air Mixture In
QCond Return Water
QEvap
Makeup Water
Wp
Wp
UPS
PDU
Qdata center
Switch Gear
Data Center
Power
Computing
Cooling
![Page 9: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/9.jpg)
Sustainable Ecosystem Research Group HP Labs
• Sustainable Data Center − Integrated management of IT, power and cooling
towards a net-zero data center
• Resource Management as a Service − Improve sustainability of urban infrastructure, e.g.
power, water.
9 3 May 2012
![Page 10: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/10.jpg)
10
Sustainable Operation and Management of Chillers using Temporal Data Mining (KDD ‘09)
•Data Centers −Cooling Infrastructure
• Problem Statement • Prior Work •Our Approach −Symbolic representation −Event encoding −Motif mining −Sustainability characterization
• Experimental Results • Summary
![Page 11: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/11.jpg)
11
Data Center Cooling Infrastructure
Computer room air-conditioner (CRAC)
Chiller Unit
Cooling Towers
Water Return (Tin)
Water Supply (Tout)
Consumes from 1/3 up to 1/2 of total power consumption
![Page 12: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/12.jpg)
12
Ensemble of Chillers •Challenging to operate efficiently −Complex physical system
• Dynamic • Heterogeneous • Inter-dependencies • Many constraints
−Accurate models not available −Rapid cycles undesirable – reduce
lifespan
•Domain experts determine settings based on heuristics
•Can it be automated through a data-driven approach?
• Which unit to turn ON/OFF?
• At what utilization?
• How to handle increase/decrease in cooling load?
Chiller Ensemble
![Page 13: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/13.jpg)
13
Problem Statement •Given the following chiller time series −utilization levels −power consumption −cooling loads
• Is it possible to determine which operational settings are more energy efficient?
•And then use this information to advise data center facility operators
![Page 14: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/14.jpg)
14
Some Terminology • IT cooling load
• Chiller utilization
• Chiller power consumption
• Coefficient of performance (COP)
Cooling Load Power consumption
![Page 15: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/15.jpg)
15
Prior Work •Classical approaches to model time series data −Principal component analysis −Discrete Fourier transforms
•Discrete representations: SAX [Keogh et al.] •Motifs: Repeating subsequences [Yankov et al.]
![Page 16: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/16.jpg)
16
Our approach • Goal: Sustainability
characterization of multi- variate time series data − Chiller utilization data
• Four Main Steps − Symbolic representation − Event encoding − Motif mining − Sustainability
Characterization
Cluster Analysis
Multivariate Time Series Data
Event Encoding
Frequent Motif Mining
Symbolic representation
Transition-event sequence
Frequent motifs
Sustainability characterization
of frequent motifs
Other discrete data sources can be integrated
![Page 17: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/17.jpg)
17
Clustering • Individual vector:
Utilization across all chiller units
• Raw Data: Sequence of such vectors
• Perform k-means clustering • Use cluster labels to
encode multi-variate time series
![Page 18: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/18.jpg)
18
Event Encoding and Motif Mining
•Event sequences •Motif mining −Episode Framework −Non-overlapped occurrences −Inter-event gap constraint
![Page 19: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/19.jpg)
19
• Event Sequence Ei = Event type ti = Time of occurrence
• Episode − Ordered collection of events occurring together
• Episode occurrence
− Events same ordering as episode in the data.
• Motifs − Frequently occurring episodes
),(),...,,(),,( 2211 NN tEtEtE
)21,(),20,(),17,(),15,(),14,(),12,(),6,(),4,(),3,(),1,( ACDBEACDBA
( )CBA →→
<(A,1), (B,3), (D,4), (C,6), (E,12), (A,14), (B,15), (C,17)>
Some Definitions
![Page 20: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/20.jpg)
20
Redescribing time series data
• Perform run-length encoding: − Note transitions from
one symbol to another
• Higher level of abstraction − Transition events
![Page 21: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/21.jpg)
21
Motif mining • Frequency counting: Non-overlapped
occurrences
• Level-wise (Apriori-style) episode mining
A B A C B A C B A C
Non-overlapped
![Page 22: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/22.jpg)
Itemset Mining/Association rule mining • Example: Market Basket Analysis • Items frequently purchased together:
Bread ⇒PeanutButter •Uses: −Placement −Advertising −Sales −Coupons
![Page 23: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/23.jpg)
Apriori Algorithm • Frequen t I t em set P roper ty : Any subset of a frequent itemset is frequent. •Contrapositive:
If an item set is not f requen t , none of it s superset s are f requen t .
![Page 24: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/24.jpg)
Level-wise (Apriori-based) motif mining
24 3 May 2012
Candidate generation followed by counting
![Page 25: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/25.jpg)
25
Episode Counting • Finite state automata based counting algorithm • Support = |largest set of non-overlapped
occurrences of transition-event episodes| •Count allows gaps or intervening junk symbols
![Page 26: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/26.jpg)
26
aabbbbbaaaxaaacccccaaaaabbbbbbaaeaaaaaacccccbggaaa
Discrete representation of chiller ensemble time-series Clustering
aabbbbbaaaxaaacccccaaaaabbbbbbaaeaaaaaacccccbggaaa
Occurrence #1 Occurrence #2 ab->ba->ac Motif
Transition
Encoding
Frequent
Episode
Mining
Methodology Summary M
ulti-
varia
te ti
me-
serie
s Vector representation
![Page 27: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/27.jpg)
27
Advantages of our approach
• We model transitions from one state to another − States correspond to clusters
• We allow don't cares between state transitions in a more expressive way − Provides robustness to clustering
• Result of mining is a set of occurrences of a motif −Motifs must repeat at least N times to be considered
frequent − Lowers the likelihood of finding false positives
![Page 28: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/28.jpg)
28
Robustness of motif occurrences
Matches approximately similar patterns
Don’t care transition events in the encoded sequence
![Page 29: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/29.jpg)
29
Sustainability characterization of Motifs
•Average motif COP (coefficient of performance) − Indicates cooling efficiency of a chiller unit
• COP = IT Cooling Load Power consumed
• Frequency of oscillations of a motif − Impacts chiller lifespan −Normalized number of mean-crossings
![Page 30: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/30.jpg)
30
Experimental Results •Data −From HP R&D data center in Bangalore
• 70,000 sq ft • 2000 racks of IT equipments
−Ensemble of five chiller units • 3 air cooled chillers • 2 water cooled chillers
−480 hours of data • July 2 – 7, Nov 27 – 30, Dec 16 – 26, 2008
•22 motifs found in the data
![Page 31: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/31.jpg)
31
11,11,11,8,8,8,8,10,10,10,10,10,10,10,10,10,10,10,10,10,10,10,13,14,12,12,11,11,11
11-8,6637 8-10,6641 10-13,6656 13-14,6657 14-12,6658 12-11,6660
[ 11-8 , 8-10 , 10-13 , 13-14 , 14-12, 12-11 ]
4 15 1 1 2
Symbol seq:
Encoded seq:
Time Series
Transition Motif:
Inter-transition gap constraint = 20 min
A Motif – Detailed Example (1/3)
![Page 32: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/32.jpg)
32
11,11,11,8,8,8,10,10,10,10,10,10,10,10,10,10,10,10,10,13,13,14,12,12,11,11,11
11-8,6698 8-10,6701 10-13,6714 13-14,6716 14-12,6717 12-11,6719
3 13 2 1 2
[ 11-8 , 8-10 , 10-13 , 13-14 , 14-12, 12-11 ]
Symbol seq:
Encoded seq:
Time Series
Transition Motif:
Inter-transition gap constraint = 20 min
A Motif – Detailed Example (2/3)
![Page 33: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/33.jpg)
33
11,11,11,8,8,8,10,10,10,10,10,10,10,10,10,10,10,10,10,10,13,14,12,12,12,11,11,11
11-8,6758 8-10,6761 10-13,6775 13-14,6776 14-12,6777 12-11,6780
3 14 1 1 3
[ 11-8 , 8-10 , 10-13 , 13-14 , 14-12, 12-11 ]
Symbol seq:
Encoded seq:
Time Series
Transition Motif:
Inter-transition gap constraint = 20 min
A Motif – Detailed Example (3/3)
![Page 34: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/34.jpg)
34
Motif 5
Two Interesting Motifs
C1, C2, C3 → Air cooled
C4, C5 → Water cooled Motif 8
Time (min) →
Chiller
C1
C2
C3
C4
C5
18%
49%
44%
0%
0%
34%
11%
0%
66%
0%
34
Motif 8 Motif 5
COP 4.87 5.40 Units operating 3 air-cooled 2 air-cooled, 1
water cooled
![Page 35: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/35.jpg)
35
Potential Savings
•Annual saving from operating in Motif 5 instead of Motif 8 −Cost savings = $40,000 (~10%) −Carbon footprint savings = 287,328 kg of CO2
![Page 36: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/36.jpg)
36
Summary • Data centers chillers consume substantial power −Ensemble of chillers – part of data center cooling
infrastructure – are challenging to operate energy efficiently
•Mine and characterize motifs −Symbolic representation −Event encoding −Motif mining −Sustainability characterization
•Demonstrated our approach on data from a real data center – indicates significant potential energy savings
![Page 37: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/37.jpg)
Some other projects • Anomaly detection (SensorKDD 2010) • Energy Disaggregation (SDM 2011) • Automating Life Cycle Assessment (IEEE Computer 2011) • Fine-grained PV output prediction (AAAI 2012) • Building Energy Management (BuildSys 2011)
37 3 May 2012
![Page 38: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/38.jpg)
©2009 38
Energy Disaggregation
![Page 39: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/39.jpg)
©2009 39
Proposed Variant of Factorial HMM’s (SDM 2011)
![Page 40: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/40.jpg)
Data Analytics for Urban Infrastructure
Waste
Power
Transport
Water
…
Streaming
Data
Analytics Engine
Probabilistic Models
Optimization
Data Fusion Simulation
Policy
Sustainability Metrics
Actions
![Page 41: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/41.jpg)
References • P. Chakraborty, M. Marwah, M. Arlitt, and N. Ramakrishnan. Fine-grained Photovoltaic Output Prediction using a Bayesian Ensemble, in Proceedings of the 26th Conference on Artificial Intelligence (AAAI'12), Toronto, Canada, 7 pages, July 2012, To appear.
• Z. Liu, Y. Chen, C. Bash, A. Wierman, D. Gmach, Z. Wang, M. Marwah, C. Hyser, "Renewable and Cooling Aware Workload Management for Sustainable Data Centers", ACM SIGMETRICS/Performance, June 11-15 2012, London, UK, To appear.
• Manish Marwah, Amip Shah, Cullen Bash, Chandrakant Patel, Naren Ramakrishnan, "Using Data Mining to Help Design Sustainable Products," IEEE Computer, August 2011
• Hyungsul Kim, Manish Marwah, Martin Arlitt, Geoff Lyon and Jiawei Han, "Unsupervised Disaggregation of Low Frequency Power Measurements", SIAM International Conference on Data Mining (SDM 11), Mesa, Arizona, April 28-30, 2011.
• Gowtham Bellala, Manish Marwah, Martin Arlitt, Geoff Lyon, Cullen Bash, "Towards an understanding of campus-scale power consumption." In ACM BuildSys, November 1, 2011, Seattle, WA.
• Manish Marwah, Ratnesh Sharma, Wilfredo Lugo, Lola Bautista, "Anomalous Thermal Behavior Detection in Data Centers using Hierarchical PCA," in SensorKDD in conjunction with KDD 2010.
• D. Patnaik, M. Marwah, Sharma, Ramakrishna, "Sustainable Operation and Management of Data Center Chillers using Temporal Data Mining," In ACM KDD, June 27 - July 1, 2009, Paris, France.
• Amip Shah, Tom Christian, Chandrakant D. Patel, Cullen Bash, Ratnesh K. Sharma: Assessing ICT's Environmental Impact. IEEE Computer 42(7): 91-93, July 2009.
![Page 42: Data Mining for Sustainable Data Centersweb.stanford.edu/.../lecture/may1/mm_hp.pdf · • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT](https://reader035.vdocument.in/reader035/viewer/2022062603/5f0314bb7e708231d4077075/html5/thumbnails/42.jpg)
2nd KDD Workshop on Data Mining Applications In Sustainability Date: August 12, 2012 Location: Beijing, China Objective The goals of this KDD workshop are: • to bring together researchers working on
applications of KDD to sustainability in diverse areas, especially in infrastructures such as IT, Smart Grids, water, and transportation.
• to familiarize the mainstream KDD community with diverse application areas within sustainability.
• to serve as a meeting ground and launchpad to galvanize and foster the development of this budding sub-community.
Organizing Committee Chairs • Naren Ramakrishnan, Virginia Tech (co-chair) • Manish Marwah, HP Labs (co-chair) • Mario Berges, CMU (co-chair) • Zico Kolter, MIT (co-chair)
Paper Submission Two types of papers in ACM SIGKDD format are encouraged: long papers with a maximum of 8 pages describing completed work on data mining problems in sustainability and short papers of 4-6 pages describing ongoing research or preliminary results. We also invite a 1-2 pages extended abstract for early-stage work to be presented as posters. Important Dates: Submission: May 23, 2012 Notification: June 4, 2012 Camera-ready Versions: June 8, 2012 Workshop: August 12, 2012 For More Information: http://marioberges.com/SustKDD12