Collaborative QoS Prediction in Cloud Computing Department of Computer Science & Engineering The Chinese University of Hong Kong Hong Kong, China Rocky Yilei Zhang Nov. 15, 2011

Outlines Introduction System Architecture Memory-Based QoS Prediction Time-Aware QoS Prediction Conclusion 2

Cloud Computing  Cloud computing provides a model for enabling convenient, on- demand network access to a shared pool of computing resources :  Networks  Servers  Databases  Services 3

Cloud Applications  Building on a number of distributed cloud components  Large-scale  Complicated  Time sensitive  High-quality  Case 1: New York Times  Used EC2 and S3 to convert 15 million scanned news articles to PDF (4TB data)  100 Linux computers 24 hours  Case 2: Nasdaq  Uses S3 to deliver historic stock and fund information  Millions of files showing price changes of entities over 10 minute segments 4

Non-Functional Performance of Cloud Components Non-functional performance of cloud components is essential for building cloud applications: – Cloud Component selection – Cloud Component composition – Cloud Component recommendation 5

Performance of Cloud Components  High-quality cloud applications rely on the high-quality of cloud components.  remote network access  Location independence  Personalized performance evaluation on cloud components is essential.  Method 1: evaluating all the components to obtain their QoS performance.  Impractical: time-consuming, expensive, thousands of components.  Method 2: collaborative filtering approach  Predicting component QoS by employing usage experiences from similar users. 6

System Architecture 7

Memory-Based QoS Prediction Yilei Zhang, Zibin Zheng, and Michael R. Lyu, “Exploring Latent Features for Memory-Based QoS Prediction in Cloud Computing”, in Proceedings of the 30th IEEE Symposium on Reliable Distributed Systems (SRDS 2011), Madrid, Spain, Oct. 4-7,

Example User-component matrix: m × n, each entry is a QoS value. – Sparse – Prediction accuracy is greatly influenced by similarity computation. 9

Latent Features Learning 10 Latent-component matrix HLatent-user matrix V u1 u2 u3 u4 c1 c2 c3 c4 c5 c6

Similarity Computation Pearson Correlation Coefficient (PCC) Similarity between users: Similarity between components: 11 Latent-component matrix H Latent-user matrix V u1 u2 u3 u4 c1 c2 c3 c4 c5 c6

Neighbors Selection For every entry w i,j in the matrix, a set of similar users towards user u i can be found by: A set of similar items towards component c j can be found by: 12

Missing Value Prediction Similar User-based: Similar Component-based: Hybrid: 13

Experiments  QoS Dataset  Metrices  : the expected QoS value.  : the predicted QoS value  N : the number of predicted values. 14

Performance Comparisons 15

Impact of Matrix Density 16

Impact of Top-K 17

Impact of Dimensionality 18

Conclusions and Future Work  Conclusions:  A collaborative approach for personalized cloud component QoS value prediction  A large-scale real-world experiment  A publicly released real-world QoS dataset  Future Work:  Investigation of more QoS properties  Experiments on different kinds of cloud components 19

Time-Aware QoS Prediction Yilei Zhang, Zibin Zheng, and Michael R. Lyu, “WSPred: A Time-Aware Personalized QoS Prediction Framework for Web Services”, in Proceedings of the 22th IEEE Symposium on Software Reliability Engineering (ISSRE 2011), Hiroshima, Japan, Nov. 29-Dec. 2,

Quality-of-Service Quality-of-Service (QoS): Non-functional performance. – User/Time-independent QoS properties. price, popularity. No need for evaluation – User/Time-dependent QoS properties. failure probability, response time, throughput. Different users receive different performance at different time. Impact factors: – Remote network access – Location – Invocation time 21

Time-Aware QoS Performance Time-aware personalized QoS evaluation on cloud components is essential for: – Automatically selection – Dynamically composition 22

Challenge: How to Evaluate? – Evaluating all the cloud components to obtain their QoS performance before building cloud application s. Time-consuming Expensive Thousands of cloud components – QoS prediction Predicting QoS values by employing usage experiences in the past. 23

Related Work Predicting average performance – Memory-based – Model-based Need to considering the difference in terms of time 24

Case Study 25

Tensor Factorization User 26 Component Time

Objective Function 27

Missing Value Prediction 28

Dataset Time-Aware Web Service QoS Dataset 29

Metrics Mean Absolute Error (MAE) Root Mean Squared Error (RMSE) – : the expected QoS value (ground truth). – : the predicted QoS value – N : the number of predicted values. 30

Comparison with Other Methods MF1 – This method considers the user-service-time tensor as a set of user- service matrix slices in terms of time. Then employ MF. MF2 – compresses the user-service-time tensor into a user-service matrix. Then apply MF. TF – tensor factorization-based prediction method WSPred – tensor factorization-based recommendation with average QoS value constraints 31

Experimental Results 32

Impact of Tensor Density 33

Impact of Dimensionality 34

Conclusions  A time-aware approach for Cloud Component QoS value prediction  A large-scale experiment  A publicly released Time-Aware QoS dataset 35

Thank you! 36