Motivation and Overview Towards Defining and Exploiting 12.3.40.65 GET index.jsp Behaviorally related Similarities in Web Application Use 12.3.40.65 GET login.jsp View user sessions as use cases sequence of events 12.3.40.65 GET reg.jsp? performed by the user login=xxx&password=hello& Cases through User Session Analysis through interaction 12.3.40.65 GET myinfo.jsp Learn about dynamic behavior with the system User session analysis Test case generation Sreedevi Sampath, University of Delaware Amie Souter, Drexel University • Clustering via concept analysis Software Monitoring Lori Pollock, University of Delaware • Common subsequences analysis development/ load of maintenance traffic tools Content Workshop on Dynamic Analysis (WODA), May 25, 2004 Test suite personalization reduction co-located with International Conference on Software Engineering (ICSE 2004) GD GR GL RELATION TABLE attributes (URLs) Step 1 GD GR GL PL GS GB GM us1 us5 Clustering via Concept Analysis o us1 x x x GS b x x x x x us2 j • Mathematical technique for clustering objects e us3 x x x x x GB PL that have common discrete attributes c us4 x x x x x t GM us5 x x x us4 s us3 • Set of objects, O: user sessions, us us6 x x x x x x us2 us6 • Set of attributes, A: URLs, u ( {us1, us2, us3, us4, us5, us6} ,{GD,GR,GL} ) SPARSE • Relation, R: us requests u ( {us2, us3, us4, us6} ,{GD,GR,GL,GS} ) • Concept analysis identifies all concepts (O i , A j ) ( {us3} ,{GD,GR,GL,PL,GS} ) ( {us4} ,{GD,GR,GL,GS,GB} ) for a given tuple (O, A, R) ( {us2} ,{GD,GR,GL,GS,GM} ) ( {us6} ,{GD,GR,GL,PL,GS,GB} ) CONCEPT LATTICE: FULL ( null, {GD,GR,GL,PL,GS,GB,GM} ) Step 2 Hypothesis Motivating the Approach Heuristic for Test Suite Reduction • Common Subsequences Hypothesis : • Smallest set of user sessions The set of user sessions clustered • Covers all the URLs together into the same concept node will • Represents common URL subsequences GD GR GL have a high commonality in the of different use cases subsequences of URLs in their sessions us1 us5 GS Identify next-to-bottom nodes GB Pick one user session from each of these PL next-to-bottom nodes GM us4 us3 Resulting reduced test suite: {us2, us6} us2 us6 1
Metric for Common Subsequences Finding Common Subsequences of URLs Hypothesis • attr-size[n] set: level of node in lattice Subsequences of URLs are us3 us6 attr-size[5]: level 5 Common us2 representative of partial us1 us3 Subsequences a GD GD use cases of the user sessions a a b b b GR GR [GD, GR, GL] # user c # user c c [PL, GS] d sessions= 4 sessions = 3 GL GL e d NODE 003 a # URLs = 5 [GR, GL] # URLs = 5 a a PL GB b {a, b, c, d, e} b objects b e e e GS PL { us3, us6 } a d PL GS attributes Sub Common Percent GR GR seq subsequence attrs { GD, GL, GR, GS, PL } size GL GL covered Metric Percent of attributes covered PL GB 1 a, b, c, d, e 100 % by common subsequences GS PL of URLs of various sizes 2 ab, bc, be 80% GS 3 abc, abe 80% Results for Common Experiment: Applications Used Subsequences Hypothesis Percent of Attributes Covered Bookstore • Bookstore web application 52 � 9,748 LOC, 385 methods, 11 classes 50 48 � Front end: JSP, Backend: MySql 46 44 42 � 123 user sessions 40 38 36 34 • uPortal application 32 30 � 38,589 LOC, 4233 methods, 508 classes 28 26 24 � Java, JSP, XML, J2EE 1 4 710 13161922252831343740 � 2083 user sessions 8 Attr-Size Set Subsequence Size Results for Common Conclusions for Common Subsequence Hypothesis Subsequences Hypothesis Bookstore • Between user sessions of a node there exists commonality in subsequences of URLs • These common subsequences cover a reasonable percent of URLs (attributes) of the node • Clustering based on single URLs � clusters similar use cases � can choose one object from each node Result: subsequences of various sizes cover reasonable percent of attributes 2
Next-to-bottom Coverage of Use Cases Results for Next-to-bottom Coverage Hypothesis of Use Cases Hypothesis In addition to covering all the URLs of the original test suite, Metric: loss of coverage of use cases in remaining the user sessions in next-to-bottom nodes execute a set by the reduced set high percentage of the subsequences of URLs of the rest of the original test suite Bookstore GD GR GL remaining set us1 us5 reduced set GS {us1, us3, us4 {us2, us6} us5} GB PL user sessions all other user sessions GM belonging except sessions us4 us3 to next-to-bottom belonging to nodes next-to-bottom nodes Result: short sequences present but long us2 us6 sequences are missing Conclusion for Next-to-bottom Pros and Cons of Our Approach Coverage of Use Cases Hypothesis + Results from common subsequences • Long sequences absent but smaller hypothesis support using concept analysis sequences are present in reduced set for clustering user sessions • reduced set contains more URLs hence + Experiments show little coverage loss (tech may contain other URL sequences report) by reduced test suite absent in remaining set - Results from next-to-bottom coverage of use • Moderately supports picking next-to- cases hypothesis indicate further work bottom nodes for reduced test suite needed on heuristic Future Work • Explore additional heuristics • Additional user session analysis � Useful for other software engineering tasks 3
Recommend
More recommend
