140 likes | 261 Views
Analysis of Communication Mechanisms in WS C ompositions. Verification Problem. Check correctness of the behavior of WS composition with respect to the set of different properties of interest: Deadlock freeness , LTL properties
E N D
Verification Problem • Check correctness of the behavior of WS composition with respect to the set of different properties of interest: • Deadlock freeness, LTL properties • Verification relies on the existence of the formal model for: • composition representation, interactions representation • Communication mechanisms are complex and implementation-dependent: • queues, diversity of protocols, message overpasses • No appropriate formal model for the analysis that allows to analyze arbitrary composition scenario: • Complexity and decidability problems for general models • Specific models are restricted w.r.t. the set of scenarios WS-FM'05
Example 1 • Problems with synchronous communication model: • Each output happens synchronously with corresponding input • Concurrent emissions are possible, queues are needed !a a ?a ?b b !b !a !b !d d ?d ?c c !c WS-FM'05
Example 2 • Problems with asynchronous communication model: • Output is non-blocking • Messages are stored in FIFO queues • Message overpasses are possible, non-FIFO queues might be needed !a ?a ?b !a !b ?a ?a ?b ?b !b ?a !b ?a ?b GOOD BAD BAD GOOD WS-FM'05
Aim of the Approach • Adequateness Starting from the composition scenario, find an adequate communication model that completely describes the behavior of the composition • Composition boundedness Check that the system execution does lead to infinitely growing queues • Composition completeness Check that in any possible execution messages are not lost This kind of analysis allows for abstracting from the communication-level problems, thus enabling further analysis of business-level behavioral requirements WS-FM'05
Solution • Previous work (WSFM’05) • Based on the Extended Composition Model • Complex implementation • Restricted w.r.t. set of analyzable properties • Potentially incomplete results for an arbitrary scenario • Current work • Based on the STS with channels model • Simple implementation • Efficient analysis • Wider set of properties • Complete theoretical results WS-FM'05
Formalization • Assumptions: • Disjoint and perfect channels • Non-blocking operations • Arbitrary (up to non-FIFO) queue implementations • Arbitrary (but bounded) queue length • No internal loops in the service implementation • STS with Channels corresponding to composition of n STSs: • State is a pair <S,C> where S=<s1,…,sn> is global control state and C = <w1,…,wm> is a content of m FIFO queues • T = GS x (I υ O υ {τ}) x GS is a transition relation: • If action is input ?a, message is consumed: a.C’ = C • If action is output !a, message is added to a queue: C’ = C.a • If action is internal τ, queue content is not changed: C’ = C WS-FM'05
Formalization (cont’d) • Behavior: • Reachability tree (possibly infinite): nodes labeled with global states of CSTS root is labeled with initial state of CSTS edges labeled with actions • Reachability graph obtained from reachability tree by merging nodes with identical labels • Finiteness of the model: • THEOREM: reachability graph is bounded iff all the queues are bounded [that is there is a constant K s.t. length of the queue < K in any global state] • The problem of unboundedness is undecidable in general WS-FM'05
Adequateness Problem • Given a composition scenario (set of STSs), determine a minimal CSTS (channel configuration) such that the behavior is the maximal for the scenario • Minimal model: • The simplest interaction mechanisms, the most trivial implementation • The most efficient analysis techniques • Maximal behavior: • The behavior should be the same as in the worst case where everything is possible • Relies on the model simulation relation between models • Requires the definition of the worst case - most general model WS-FM'05
Adequateness Problem (cont’d) • Model simulation relation • Model Δ1 simulates model Δ2, Δ2 ≤ Δ1, if for any composition scenario set of behaviors of the composition under model Δ2 is included in the set of behaviors of the composition under model Δ1 • THEOREM: Let Δ1 Δ2 be two communication models with queue structures <M11*,…, M1m*> and <M21*,…, M2n*>. If for any alphabet M2i there exists an alphabet M1j such that M2i ≤ M1j , then Δ1 ≤ Δ2 • Most general communication model • Δ is the communication model with the following queue structure: For any message type a there is a separate queue with alphabet M = {a} • THEOREM: for any communication model Δ, Δ ≤ Δ • The boundedness problem for the Δ model is decidable WS-FM'05
Adequateness Analysis • Consider a sequence of communication models from the simplest to the most general: Δ1 ≤ Δ2 ≤ … ≤ Δ • Starting from the first model, compute the behavior of the composition under the current model and compare it with the behavior under the most general model • If different, take the next model and repeat the step • If the same, then Δi = Δand therefore the model is adequate • The implementing algorithm is based on the reachability graph construction • DFS algorithm • Boundedness is checked on the fly • Completeness is checked on the fly WS-FM'05
Analysis issues • The outcome of the analysis is represented as global transition system • Some states may be marked to be unbounded • Some states may be marked to be incomplete • This model is used for further analysis of properties [with presence of variables] in NuSMV Overall analysis approach is a combination of explicit search techniques (graph construction) and symbolic techniques (further model checking) WS-FM'05
Analysis issues (cont’d) • The explicit search algorithm allows for application of Partial Order Reduction and on-the-fly optimizations. • The experimental results demonstrate much higher performance w.r.t. previous NuSMV implementations • Future work: Knowledge-level verification • May be performed on-the-fly combined with the graph constructions algorithm WS-FM'05
? Anyquestion WS-FM'05