SlideShare a Scribd company logo

Noha danms13 talk_final

N
Noha Elprince

This document presents an approach for autonomous resource provisioning in virtual data centers. It aims to optimize resource allocation to avoid under and overprovisioning while enabling service differentiation. The proposed solution uses machine learning models to predict resource needs and a fuzzy rule-based system to tune allocations based on errors. It was tested on a real workload trace achieving accurate predictions. The system can also adapt online or offline through retraining models with new data to maintain optimal resource allocation over time.

1 of 30
Downloaded 11 times
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Autonomous Resource Provision in Virtual Data Centers Presented by: Noha Elprince noha.elprince@uwaterloo.ca IFIP/IEEE DANMS, 31 May 2013
Unused resources 2 Demand Capacity Time Resources Demand Capacity Time Resources Static Data Centers vs. Dynamic Figure: RAD Lab, UC Berkeley •  Fixed pre-assigned Resources ( provision for peak ) •  Static Environment •  Manual change of configurations •  Cloud Elasticity ~ “Pay as you go” •  Virtualized Environment •  Automated “Self-service” change of configurations.
3 •  under-provisioning Heavy Penalty Lost revenue Lost users Resources Demand Capacity Time (days) 1 2 3Resources Demand Capacity Time (days) 1 2 3 Resources Demand Capacity Time (days) 1 2 3 3 Figures: RAD Lab, UC Berkeley Cloud Elasticity problems…
Cloud Elasticity … 4 •  over-provisioning overutilization Demand Capacity Time Resources Unused resources Figure: RAD Lab, UC Berkeley
5 Virtualization (Cloud Foundation) •  Virtualization allows a computational resource to be partitioned on multiple isolated execution environments (VMs) •  Turning the machine into a “virtual image” ~ Self-immunity from: Ø  Hardware breakdowns. Ø  Running out of the resources Challenge: Service Differentiation
Problem q  Over and under provisioning in spite of: difficulty of estimating the actual needs due to time- varying and diverse workload. q  Enabling service differentiation in a virtualized environment. 6
Methodology •  Develop and implement an autonomic resource management controller that: Ø  Effectively optimize the resource by predicting current resource needs. Ø  Continuous resource self-tuning to accommodate load variations and enforce service differentiation during resource allocation. •  Test the proposed prototype on real traces. 7
Motivation •  Help datacenters to manage resources effectively. •  Propagate Cloud Computing (increase cloud users => less expensive) •  Optimize resources (green I.T !) 8
Related Work v Approaches for autonomic resource management: –  Utility based self-optimizing approach. –  Model-based approach based on perf. Modeling. –  Machine Learning Approach –  Fuzzy logic approach. 9
Proposed Solution Architecture: Sys. Modeling 10 r(t+1)
I. System Modeling : Data set 11 •  Idea: Learning from successful jobs: (normal termination, fulfill client’s anticipated perf.) •  A real computing center trace of Los Alamos National Lab (LANL) •  LANL is a United States Department of Energy (DOE) national laboratory. •  LANL conducts multidisciplinary research in fields such as national security , space exploration, renewable energy, medicine, nanotechnology and supercomputing. •  System: 1024-node Connection Machine CM-5 from Thinking Machines •  Jobs: 201,387 , Duration: 2 Yrs.
12 v  Feature Selection •  Use stepwise regression to: Sort variables out & leave more influential ones in the model. •  Results: Out of 18 features, 5 features were selected: => run_time, wait_time, Avg_cpu_time, used_mem, status v  Filter •  Remove jobs with status =unsuccessful ( failed/ aborted ) •  Discard records that have average_cpu_time_used <=0 and used_mem <=0 v  Data Cleaning •  Normalize data to remove noise I. System Modeling : Data Preprocessing
13 I. System Modeling : Statistical Analysis
14 ! Cascaded classifiers (MISO model) I. System Modeling : Model I/O
15 §  Linear Regression §  Sugeno Fuzzy Inference System (FCM, SUB) §  Regression Tree (REP-Tree) §  Model Tree (M5P) §  Boosting (Rep-Tree, M5P) §  Bagging (Rep-Tree, M5P) I. System Modeling : ML approaches Why ML ? - Due to the Non-linear nature of the data -  Ability to deal with complex nature of data. -  Detect dependency between i/ps and o/ps efficiently.
16 Bagging vs. Boosting Classifiers •  Bagging (Bootstrap aggregating) uses bootstrap sampling. •  Trains k classifier on each bootstrap sample. •  A weighted majority (voting) of the k learned classifiers (using equal weights). •  Boosting: weak classifiers form a final strong classifier. •  After a weak learner is added, the data is reweighted: Ø  misclassified examples=> gain weight Ø  examples classified correctly => lose weight. •  Thus future learners focus more on the data that previous weak learners misclassified.
II. Res. Predictor 17 v The client requests hosting a specific type of application with a pre-specified response time. v  An initial estimate is generated. v Rate of prediction is accompanied by the coming of the client to the data center.
18 Classifier Type RMSE MAE RAE CC Linear Reg. C1 0.0024 0.0008 50.33% 0.70 C2 0.0023 0.0001 57.29% 0.71 C3 0.0026 0.0003 58.15% 0.98 Sugeno FIS (SUB) C1 0.0021 0.0009 44.89% 0.66 C2 0.0012 0.0002 51.06% 0.66 C3 0.0011 0.0002 53.93% 0.85 Boosting (M5P) C1 0.0020 0.0006 34.59% 0.80 C2 0.0018 0.0007 39.20% 0.84 C3 0.0003 0.0001 10.99% 0.99 Bagging Tree (M5P) C1 0.0018 0.0005 32.57% 0.84 C2 0.0017 0.0007 36.38% 0.84 C3 0.0003 0.0001 11.82% 0.99 Validation: Perf. Measures for different prediction models
19 Resource Predictor: Learning Time Comparison
III. Resource Allocator 20 1.  Res Allocator initially allocates resources ( based on the prediction model). 2.  Check the error resulting from the tuner. 3.  The tuner Calculates the normalized error in resource allocation 4. Takes the feedback from the tuner (ResAdjustment) and sends a command to the VC in the VM with the appropriate decision. RespTimeError(k)= RespTime(k)ref !RespTime(k)obs RespTime(k)ref
21 IV. Resource Tuner : Rule-Based Fuzzy System i/ps o/p RespTimeError ClientClass Status ResDirection ResController ( mamdani)
22 IV. Resource Tuner: Rule-Based Fuzzy System !!!! "#$%&'()($%! !!!!!!!#$%*+',$-((.(! /.0! 1$2'3,! 4'56! ! ! ! 78'$9:!! ;8<%%! =.82! !!!!!!!! ! ! >?/! !!!!!!!!>&1! ! >?1! !!!!!!>&4! ! >?4! >'8@$(! !!!!!!>&/! ! ! !! !!!!!!!>&1! ! >?1! !!!!!>&4! ! >?4! A(.9B$! !!!!!!>&/! ! !! !!!!!!!>&1!!!!!!!!! ! ! !!!!!!!>&4! >?4! ! Over provision / Under provision -  Total # rules : 18 -  The grades of membership of each attribute (high, medium , low) are adjusted by experts in the datacenter. -  ResDir : •  reflects a percentage of the resource that should be utilized in the VC. (ResAdjust = ResDir x ResWt x VCres) •  ranges [-1 +1] with MFs (low, med, high) for: Ø  speed up (+ve side) Ø  step down (-ve side)
V. Adaptive Learning 23 New incoming data will be fed into the prediction model by different ways (depending on the prediction model used): -  Directly via clustering (if clustering is used as in TS-FIS) => online learning -  OR it will be stored in the database until a certain threshold reached, then an ECA rule is fired , initiating re- modeling => offline learning
V. Adaptive Learning: update Rules in Fuzzy Tuner FIS 24 Rule Editor
25 Resource Tuner validation - Example !!!! "#$%&'()($%! !!!!!!!#$%*+',$-((.(! /.0! 1$2'3,! 4'56! ! ! ! 78'$9:!! ;8<%%! =.82! !!!!!!!! ! ! >?/! !!!!!!!!>&1! ! >?1! !!!!!!>&4! ! >?4! >'8@$(! !!!!!!>&/! ! ! !! !!!!!!!>&1! ! >?1! !!!!!>&4! ! >?4! A(.9B$! !!!!!!>&/! ! !! !!!!!!!>&1!!!!!!!!! ! ! !!!!!!!>&4! >?4! ! Over provision / Under provision Method: Testing cases using the fuzzy rule viewer.
26 I/ps: RespTimeError : medium , client class : Gold and status: underprovision O/p: ResDirection : SUM (speed up medium) Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.9 status= 0.2 ResDirection = 0.5
27 I/ps: RespTimeError: medium, Client class: Silver, Status : underprovision o/p: ResDirection is SUM (speed up medium) Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.5 status= 0.2 ResDirection = 0.5
28 I/ps: RespTimeError : medium , client class : bronze , status: underprovision o/p: ResDirection : noAction Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.19 status= 0.2 ResDirection = 0.01
Conclusions •  Proposed ML model predicts the right amount of resources (Bagging/Boosting is promising). •  The Fuzzy tuner - Accommodates any deviation in workload c/cs. - Enforces service differentiation. •  Adaptive Learning guarantee having an up-to-date model that lowers future SLA violations. 29
Questions ? 30

More Related Content

PDF
Fuzzy Self-Learning Controllers for Elasticity Management in Dynamic Cloud Ar...
Pooyan Jamshidi
 
PDF
Using Simple PID Controllers to Prevent and Mitigate Faults in Scientific Wor...
Rafael Ferreira da Silva
 
PPT
Clustering
Anjan Goswami
 
PDF
Alex Smola, Professor in the Machine Learning Department, Carnegie Mellon Uni...
MLconf
 
PDF
Fast Perceptron Decision Tree Learning from Evolving Data Streams
Albert Bifet
 
PPTX
A Scaleable Implementation of Deep Learning on Spark -Alexander Ulanov
Spark Summit
 
PDF
Automatic Features Generation And Model Training On Spark: A Bayesian Approach
Spark Summit
 
PDF
RUCK 2017 MxNet과 R을 연동한 딥러닝 소개
r-kor
 
Fuzzy Self-Learning Controllers for Elasticity Management in Dynamic Cloud Ar...
Pooyan Jamshidi
 
Using Simple PID Controllers to Prevent and Mitigate Faults in Scientific Wor...
Rafael Ferreira da Silva
 
Clustering
Anjan Goswami
 
Alex Smola, Professor in the Machine Learning Department, Carnegie Mellon Uni...
MLconf
 
Fast Perceptron Decision Tree Learning from Evolving Data Streams
Albert Bifet
 
A Scaleable Implementation of Deep Learning on Spark -Alexander Ulanov
Spark Summit
 
Automatic Features Generation And Model Training On Spark: A Bayesian Approach
Spark Summit
 
RUCK 2017 MxNet과 R을 연동한 딥러닝 소개
r-kor
 

What's hot (12)

PPTX
MLconf NYC Xiangrui Meng
MLconf
 
PDF
Master's Thesis - climateprediction.net: A Cloudy Approach
kabute
 
PDF
Sergei Vassilvitskii, Research Scientist, Google at MLconf NYC - 4/15/16
MLconf
 
PDF
Alex Smola, Director of Machine Learning, AWS/Amazon, at MLconf SF 2016
MLconf
 
PDF
RUCK 2017 김성환 R 패키지 메타주성분분석(MetaPCA)
r-kor
 
PDF
Efficient Online Evaluation of Big Data Stream Classifiers
Albert Bifet
 
PDF
Automatic Resource Elasticity for HPC Applications
rrrighi
 
PDF
Implementation of the fully adaptive radar framework: Practical limitations
Luis Úbeda Medina
 
PDF
Mining big data streams with APACHE SAMOA by Albert Bifet
J On The Beach
 
PDF
Albert Bifet – Apache Samoa: Mining Big Data Streams with Apache Flink
Flink Forward
 
PDF
QCon Rio - Machine Learning for Everyone
Dhiana Deva
 
PDF
Machine Learning - Introduction
Empatika
 
MLconf NYC Xiangrui Meng
MLconf
 
Master's Thesis - climateprediction.net: A Cloudy Approach
kabute
 
Sergei Vassilvitskii, Research Scientist, Google at MLconf NYC - 4/15/16
MLconf
 
Alex Smola, Director of Machine Learning, AWS/Amazon, at MLconf SF 2016
MLconf
 
RUCK 2017 김성환 R 패키지 메타주성분분석(MetaPCA)
r-kor
 
Efficient Online Evaluation of Big Data Stream Classifiers
Albert Bifet
 
Automatic Resource Elasticity for HPC Applications
rrrighi
 
Implementation of the fully adaptive radar framework: Practical limitations
Luis Úbeda Medina
 
Mining big data streams with APACHE SAMOA by Albert Bifet
J On The Beach
 
Albert Bifet – Apache Samoa: Mining Big Data Streams with Apache Flink
Flink Forward
 
QCon Rio - Machine Learning for Everyone
Dhiana Deva
 
Machine Learning - Introduction
Empatika
 
Ad

Viewers also liked (20)

PDF
Debug me
Noha Elprince
 
PDF
Intro to Machine Learning by Microsoft Ventures
microsoftventures
 
PPTX
Deep Learning - What's the buzz all about
Debdoot Sheet
 
PDF
T2 fs talk
Noha Elprince
 
PDF
Introduction to Deep Learning
Oleg Mygryn
 
PDF
Web UI, Algorithms, and Feature Engineering
BigML, Inc
 
PDF
Detecting fraud with Python and machine learning
wgyn
 
PDF
Using Deep Learning to do Real-Time Scoring in Practical Applications
Greg Makowski
 
PDF
Enterprise Knowledge Graph
Lukas Masuch
 
PDF
Introduction to Deep learning
Massimiliano Ruocco
 
PDF
Machine Learning for Fraud Detection
Nitesh Kumar
 
PPTX
Introduction to machinel learning and deep learning
Ibrahim Amer
 
PDF
Deeplearning NLP
Francesco Gadaleta
 
PPTX
Deep Learning in Robotics
Sungjoon Choi
 
PDF
Slides Bank England
Arthur Charpentier
 
PPTX
Fraud Detection Architecture
Gwen (Chen) Shapira
 
PDF
Deep learning - A Visual Introduction
Lukas Masuch
 
PPTX
Machine Learning with Applications in Categorization, Popularity and Sequence...
Nicolas Nicolov
 
PDF
COCOA: Communication-Efficient Coordinate Ascent
jeykottalam
 
PDF
Machine Learning Pipelines
jeykottalam
 
Debug me
Noha Elprince
 
Intro to Machine Learning by Microsoft Ventures
microsoftventures
 
Deep Learning - What's the buzz all about
Debdoot Sheet
 
T2 fs talk
Noha Elprince
 
Introduction to Deep Learning
Oleg Mygryn
 
Web UI, Algorithms, and Feature Engineering
BigML, Inc
 
Detecting fraud with Python and machine learning
wgyn
 
Using Deep Learning to do Real-Time Scoring in Practical Applications
Greg Makowski
 
Enterprise Knowledge Graph
Lukas Masuch
 
Introduction to Deep learning
Massimiliano Ruocco
 
Machine Learning for Fraud Detection
Nitesh Kumar
 
Introduction to machinel learning and deep learning
Ibrahim Amer
 
Deeplearning NLP
Francesco Gadaleta
 
Deep Learning in Robotics
Sungjoon Choi
 
Slides Bank England
Arthur Charpentier
 
Fraud Detection Architecture
Gwen (Chen) Shapira
 
Deep learning - A Visual Introduction
Lukas Masuch
 
Machine Learning with Applications in Categorization, Popularity and Sequence...
Nicolas Nicolov
 
COCOA: Communication-Efficient Coordinate Ascent
jeykottalam
 
Machine Learning Pipelines
jeykottalam
 
Ad

Similar to Noha danms13 talk_final (20)

PPTX
Ajila (1)
akanksha kunwar
 
PPTX
Data-driven AI for Self-adaptive Information Systems
Andreas Metzger
 
PDF
Machine learning and big data
Poo Kuan Hoong
 
PDF
لموعد الإثنين 03 يناير 2022 143 مبادرة #تواصل_تطوير المحاضرة ال 143 من المباد...
Egyptian Engineers Association
 
PDF
Intro to machine learning for web folks @ BlendWebMix
Louis Dorard
 
PDF
Thesis Presentation on Energy Efficiency Improvement in Data Centers
Monica Vitali
 
PPTX
Machine Learning & Predictive Maintenance
Arnab Biswas
 
PDF
Web Service QoS Prediction Based on Adaptive Dynamic Programming Using Fuzzy ...
redpel dot com
 
PPTX
AI-900 - Fundamental Principles of ML.pptx
kprasad8
 
PPTX
Using Machine Learning to Optimize DevOps Practices
Peter Varhol
 
PPTX
Thesis presentation: Applications of machine learning in predicting supply risks
TuanNguyen1697
 
PDF
A study of Machine Learning approach for Predictive Maintenance in Industry 4.0
Mohsen Sadok
 
PPTX
Intelligent Career Guidance System.pptx
Anonymous366406
 
PPTX
kdd2015
Deepak Agarwal
 
PDF
SP Big Data Meetup - March/16
Flavio Clesio
 
PDF
rerngvit_phd_seminar
rerngvit yanggratoke
 
PDF
Towards a Unified Data Analytics Optimizer with Yanlei Diao
Databricks
 
PDF
A business level introduction to Artificial Intelligence - Louis Dorard @ PAP...
PAPIs.io
 
PDF
Transfer Learning for Improving Model Predictions in Highly Configurable Soft...
Pooyan Jamshidi
 
PDF
Witekio introducing-predictive-maintenance
Witekio
 
Ajila (1)
akanksha kunwar
 
Data-driven AI for Self-adaptive Information Systems
Andreas Metzger
 
Machine learning and big data
Poo Kuan Hoong
 
لموعد الإثنين 03 يناير 2022 143 مبادرة #تواصل_تطوير المحاضرة ال 143 من المباد...
Egyptian Engineers Association
 
Intro to machine learning for web folks @ BlendWebMix
Louis Dorard
 
Thesis Presentation on Energy Efficiency Improvement in Data Centers
Monica Vitali
 
Machine Learning & Predictive Maintenance
Arnab Biswas
 
Web Service QoS Prediction Based on Adaptive Dynamic Programming Using Fuzzy ...
redpel dot com
 
AI-900 - Fundamental Principles of ML.pptx
kprasad8
 
Using Machine Learning to Optimize DevOps Practices
Peter Varhol
 
Thesis presentation: Applications of machine learning in predicting supply risks
TuanNguyen1697
 
A study of Machine Learning approach for Predictive Maintenance in Industry 4.0
Mohsen Sadok
 
Intelligent Career Guidance System.pptx
Anonymous366406
 
kdd2015
Deepak Agarwal
 
SP Big Data Meetup - March/16
Flavio Clesio
 
rerngvit_phd_seminar
rerngvit yanggratoke
 
Towards a Unified Data Analytics Optimizer with Yanlei Diao
Databricks
 
A business level introduction to Artificial Intelligence - Louis Dorard @ PAP...
PAPIs.io
 
Transfer Learning for Improving Model Predictions in Highly Configurable Soft...
Pooyan Jamshidi
 
Witekio introducing-predictive-maintenance
Witekio
 

Noha danms13 talk_final

  • 1. Autonomous Resource Provision in Virtual Data Centers Presented by: Noha Elprince noha.elprince@uwaterloo.ca IFIP/IEEE DANMS, 31 May 2013
  • 2. Unused resources 2 Demand Capacity Time Resources Demand Capacity Time Resources Static Data Centers vs. Dynamic Figure: RAD Lab, UC Berkeley •  Fixed pre-assigned Resources ( provision for peak ) •  Static Environment •  Manual change of configurations •  Cloud Elasticity ~ “Pay as you go” •  Virtualized Environment •  Automated “Self-service” change of configurations.
  • 3. 3 •  under-provisioning Heavy Penalty Lost revenue Lost users Resources Demand Capacity Time (days) 1 2 3Resources Demand Capacity Time (days) 1 2 3 Resources Demand Capacity Time (days) 1 2 3 3 Figures: RAD Lab, UC Berkeley Cloud Elasticity problems…
  • 4. Cloud Elasticity … 4 •  over-provisioning overutilization Demand Capacity Time Resources Unused resources Figure: RAD Lab, UC Berkeley
  • 5. 5 Virtualization (Cloud Foundation) •  Virtualization allows a computational resource to be partitioned on multiple isolated execution environments (VMs) •  Turning the machine into a “virtual image” ~ Self-immunity from: Ø  Hardware breakdowns. Ø  Running out of the resources Challenge: Service Differentiation
  • 6. Problem q  Over and under provisioning in spite of: difficulty of estimating the actual needs due to time- varying and diverse workload. q  Enabling service differentiation in a virtualized environment. 6
  • 7. Methodology •  Develop and implement an autonomic resource management controller that: Ø  Effectively optimize the resource by predicting current resource needs. Ø  Continuous resource self-tuning to accommodate load variations and enforce service differentiation during resource allocation. •  Test the proposed prototype on real traces. 7
  • 8. Motivation •  Help datacenters to manage resources effectively. •  Propagate Cloud Computing (increase cloud users => less expensive) •  Optimize resources (green I.T !) 8
  • 9. Related Work v Approaches for autonomic resource management: –  Utility based self-optimizing approach. –  Model-based approach based on perf. Modeling. –  Machine Learning Approach –  Fuzzy logic approach. 9
  • 10. Proposed Solution Architecture: Sys. Modeling 10 r(t+1)
  • 11. I. System Modeling : Data set 11 •  Idea: Learning from successful jobs: (normal termination, fulfill client’s anticipated perf.) •  A real computing center trace of Los Alamos National Lab (LANL) •  LANL is a United States Department of Energy (DOE) national laboratory. •  LANL conducts multidisciplinary research in fields such as national security , space exploration, renewable energy, medicine, nanotechnology and supercomputing. •  System: 1024-node Connection Machine CM-5 from Thinking Machines •  Jobs: 201,387 , Duration: 2 Yrs.
  • 12. 12 v  Feature Selection •  Use stepwise regression to: Sort variables out & leave more influential ones in the model. •  Results: Out of 18 features, 5 features were selected: => run_time, wait_time, Avg_cpu_time, used_mem, status v  Filter •  Remove jobs with status =unsuccessful ( failed/ aborted ) •  Discard records that have average_cpu_time_used <=0 and used_mem <=0 v  Data Cleaning •  Normalize data to remove noise I. System Modeling : Data Preprocessing
  • 13. 13 I. System Modeling : Statistical Analysis
  • 14. 14 ! Cascaded classifiers (MISO model) I. System Modeling : Model I/O
  • 15. 15 §  Linear Regression §  Sugeno Fuzzy Inference System (FCM, SUB) §  Regression Tree (REP-Tree) §  Model Tree (M5P) §  Boosting (Rep-Tree, M5P) §  Bagging (Rep-Tree, M5P) I. System Modeling : ML approaches Why ML ? - Due to the Non-linear nature of the data -  Ability to deal with complex nature of data. -  Detect dependency between i/ps and o/ps efficiently.
  • 16. 16 Bagging vs. Boosting Classifiers •  Bagging (Bootstrap aggregating) uses bootstrap sampling. •  Trains k classifier on each bootstrap sample. •  A weighted majority (voting) of the k learned classifiers (using equal weights). •  Boosting: weak classifiers form a final strong classifier. •  After a weak learner is added, the data is reweighted: Ø  misclassified examples=> gain weight Ø  examples classified correctly => lose weight. •  Thus future learners focus more on the data that previous weak learners misclassified.
  • 17. II. Res. Predictor 17 v The client requests hosting a specific type of application with a pre-specified response time. v  An initial estimate is generated. v Rate of prediction is accompanied by the coming of the client to the data center.
  • 18. 18 Classifier Type RMSE MAE RAE CC Linear Reg. C1 0.0024 0.0008 50.33% 0.70 C2 0.0023 0.0001 57.29% 0.71 C3 0.0026 0.0003 58.15% 0.98 Sugeno FIS (SUB) C1 0.0021 0.0009 44.89% 0.66 C2 0.0012 0.0002 51.06% 0.66 C3 0.0011 0.0002 53.93% 0.85 Boosting (M5P) C1 0.0020 0.0006 34.59% 0.80 C2 0.0018 0.0007 39.20% 0.84 C3 0.0003 0.0001 10.99% 0.99 Bagging Tree (M5P) C1 0.0018 0.0005 32.57% 0.84 C2 0.0017 0.0007 36.38% 0.84 C3 0.0003 0.0001 11.82% 0.99 Validation: Perf. Measures for different prediction models
  • 20. III. Resource Allocator 20 1.  Res Allocator initially allocates resources ( based on the prediction model). 2.  Check the error resulting from the tuner. 3.  The tuner Calculates the normalized error in resource allocation 4. Takes the feedback from the tuner (ResAdjustment) and sends a command to the VC in the VM with the appropriate decision. RespTimeError(k)= RespTime(k)ref !RespTime(k)obs RespTime(k)ref
  • 21. 21 IV. Resource Tuner : Rule-Based Fuzzy System i/ps o/p RespTimeError ClientClass Status ResDirection ResController ( mamdani)
  • 22. 22 IV. Resource Tuner: Rule-Based Fuzzy System !!!! "#$%&'()($%! !!!!!!!#$%*+',$-((.(! /.0! 1$2'3,! 4'56! ! ! ! 78'$9:!! ;8<%%! =.82! !!!!!!!! ! ! >?/! !!!!!!!!>&1! ! >?1! !!!!!!>&4! ! >?4! >'8@$(! !!!!!!>&/! ! ! !! !!!!!!!>&1! ! >?1! !!!!!>&4! ! >?4! A(.9B$! !!!!!!>&/! ! !! !!!!!!!>&1!!!!!!!!! ! ! !!!!!!!>&4! >?4! ! Over provision / Under provision -  Total # rules : 18 -  The grades of membership of each attribute (high, medium , low) are adjusted by experts in the datacenter. -  ResDir : •  reflects a percentage of the resource that should be utilized in the VC. (ResAdjust = ResDir x ResWt x VCres) •  ranges [-1 +1] with MFs (low, med, high) for: Ø  speed up (+ve side) Ø  step down (-ve side)
  • 23. V. Adaptive Learning 23 New incoming data will be fed into the prediction model by different ways (depending on the prediction model used): -  Directly via clustering (if clustering is used as in TS-FIS) => online learning -  OR it will be stored in the database until a certain threshold reached, then an ECA rule is fired , initiating re- modeling => offline learning
  • 24. V. Adaptive Learning: update Rules in Fuzzy Tuner FIS 24 Rule Editor
  • 25. 25 Resource Tuner validation - Example !!!! "#$%&'()($%! !!!!!!!#$%*+',$-((.(! /.0! 1$2'3,! 4'56! ! ! ! 78'$9:!! ;8<%%! =.82! !!!!!!!! ! ! >?/! !!!!!!!!>&1! ! >?1! !!!!!!>&4! ! >?4! >'8@$(! !!!!!!>&/! ! ! !! !!!!!!!>&1! ! >?1! !!!!!>&4! ! >?4! A(.9B$! !!!!!!>&/! ! !! !!!!!!!>&1!!!!!!!!! ! ! !!!!!!!>&4! >?4! ! Over provision / Under provision Method: Testing cases using the fuzzy rule viewer.
  • 26. 26 I/ps: RespTimeError : medium , client class : Gold and status: underprovision O/p: ResDirection : SUM (speed up medium) Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.9 status= 0.2 ResDirection = 0.5
  • 27. 27 I/ps: RespTimeError: medium, Client class: Silver, Status : underprovision o/p: ResDirection is SUM (speed up medium) Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.5 status= 0.2 ResDirection = 0.5
  • 28. 28 I/ps: RespTimeError : medium , client class : bronze , status: underprovision o/p: ResDirection : noAction Resource Tuner Validation - Example RespTimeError= 0.5 ClientClass= 0.19 status= 0.2 ResDirection = 0.01
  • 29. Conclusions •  Proposed ML model predicts the right amount of resources (Bagging/Boosting is promising). •  The Fuzzy tuner - Accommodates any deviation in workload c/cs. - Enforces service differentiation. •  Adaptive Learning guarantee having an up-to-date model that lowers future SLA violations. 29