1. Service Reliability Engineering The Chinese University of Hong Kong
Zibin Zheng
The Chinese University of Hong Kong 1

2. Modern Web Applications
Web applications permeate modern software systems!
Composed by distributed Web services
Service reliability engineering posts new research challenges

3. Building Reliable Service-Oriented Systems
It is difficult to build reliable service-oriented systems
Reliability of the system is highly dependent on the invoked Web services
Web services are provided by other organizations
The Internet environment is unpredictable
Service reliability engineering becomes a major challenge
Approaches for building reliable service-oriented systems
Fault avoidance
Fault removal
Fault tolerance
Fault prediction

4. Reliability Prediction of Web Services
Target: determine the optimal Web service from a set of functionally equivalent candidates.
Method 1: evaluate all the candidates
Weak points of Method 1:
Expensive: Requiring a lot of Web service invocations
Time-consuming: A large number of candidates to evaluate
Inaccurate: Users are not experts on WS evaluation

5. Reliability Prediction of Web Services
Method 2: Predict reliability/QoS of Web services
The prediction should be personalized for a specify user
A user may invoked some or none of the service candidates
Advantages:
Low cost: no additional WS invocations for evaluation purpose
Efficient: no need to wait for the evaluation results
Research problem:
How to make personalized Web service reliability/QoS prediction?

6. Approach 1: Neighborhood-Based
[ICSE’10, ACM SIGSOFT Distinguished Paper Award]
Reliability is extended to Quality-of-Service (QoS)
Key idea: Using past usage experiences of similar users.
Issue: How to calculate user similarity?

7. Approach 1: Neighborhood-Based
Similarity Computation
User-item matrix: M×N, each entry is the failure probability of a Web service
0.5 ?
Pearson Correlation Coefficient (PCC)

8. Approach 1: Neighborhood-Based
Drawbacks of Neighborhood-based Approach
Computational complexity
Matrix sparsity problem
Not easy to find similar users (or similar items)

9. Approach 2: Model-Based
Approach 2: Model-based Approach
[IEEE TSC’13a]
A small number of factors influencing the QoS performance
A user’s Web service QoS values correspond to a linear combination of the factors
Each row of UT is a set of feature factors, and each column of V is a set of linear predictors  Matrix Factorization (MF)
The error between the actual
Value and the prediction
s1 s2 s3 s4 s5 s6
Regularization terms
UT V

10. Fault-Tolerant Web Services
Nature of service-oriented systems by Web services
Web services are hosted by other organizations
May contain faults
May become unavailable suddenly
Source codes of the Web services are unavailable
The Internet environment is unpredictable
Resources are abundant
How to employ the redundant Web services and their QoS values for building fault-tolerant service-oriented systems?

11. Adaptive Fault Tolerance
[ESE’10]
Internet environment is highly dynamic
Network condition changes frequently and abruptly
Continuous software/hardware updates of the Web services
Server workload changes without notice
Traditional fault tolerance strategies are too static
Fixed at design time
Cannot adapt to the dynamic environment

12. Adaptive Fault Tolerance
Idea: determine optimal fault tolerance strategy dynamically at runtime based on the Web service QoS values.

13. Fault Tolerance for Significant Components
[IEEE TSC’12]
Ranking
Build a component invocation graph
Identify the most significant components
Select optimal fault tolerance strategies for the significant components.
Idea: A component that is invoked frequently by other important components is considered as a significant component (PageRank).

14. Fault-Tolerant Web Services Framework
Fault-Tolerant Framework
[IEEE TC’13]
Target: Optimal fault tolerance strategy selection for each task under local and global constraints
Local constraint: Response time of t1 < 1000 ms
Global constraint: Success-rate of the whole service plan > 99%
Candidates: Basic fault tolerance strategies and their combinations

15. Dataset Publication Dataset Publication
[ICWS’10, Best Student Paper Award]
The evaluation results are released at:
Downloaded about 2000 times by more than 150 universities (or research institutes) from more than 30 counties.
The datasets can be used in research topics of:
Web service selection and composition
Web service recommendation
Web service QoS prediction
Fault-tolerant Web services
…………………
The largest-scale real-world Web service reliability evaluation
Related papers have been cited more than 1000 times