Abstract
In mobile-enabled service-oriented computing, resource restrictions and connection intermittences often lead to transactional services being confronted with handovers and weak consistency support. To ensure transaction validity in these environments, this paper proposes a mobile composition transaction model based on Markov Chains. By analyzing service uncertainty and transaction properties, formal methods are applied to construct a mobile atomic service model where uncertainties are formulated as state probabilities. These uncertainties are then propagated through a Markov Chain-based composition model. The research introduces execution semantics for typical logic structures and demonstrates through an application case that service availability, reliability, and transaction patterns significantly impact the success probability of composite transactions.
Introduction
In the cloud computing paradigm, stable connections and abundant resources are often assumed. However, mobile computing environments are characterized by inherent instability: devices move between cells (handovers), battery life is limited, and network signals fluctuate. When multiple services are composed—such as a travel application coordinating flights, hotels, and car rentals—these uncertainties pose a significant challenge. Traditional transaction models, like those adhering to strict ACID properties, often fail because they do not account for services that may suddenly become unreachable or weakly consistent. This paper addresses the need for a formal model that characterizes this uncertainty to ensure composite transactions either complete successfully or fail gracefully.
Methodology
Atomic Service and Uncertainty Modeling
The authors define "atomic" services using a 7-tuple model that integrates hosting device hardware (CPU, memory, power) with service logic. Uncertainty is characterized across two dimensions:
- Availability: The probability that a service is reachable.
- Reliability: The probability that a service performs correctly once reached. These values are calculated using probability distributions, such as Poisson for service arrivals and Log-normal for resource demands.
Markov Chain Framework and Transaction Patterns
The core of the model is a Markov-based framework for service composition. Each step in a business process is treated as a state, with transition probabilities derived from the service's uncertainty profile. The authors analyze four specific architectural patterns for handling failures:
- Non-failure recovery (NFR): Basic execution without a safety net.
- Two-phase submission (TPS): A "prepare-then-commit" style protocol.
- Compensation recovery (CR): Executing "undo" tasks for previous steps if a failure occurs.
- Reserved protocol (RP): Pre-allocating resources to ensure higher success probability.
Experimental Evaluation
The model was evaluated using a simulated Travel Reservation Process (TRP) involving a Customer, Travel Agent, Airline, Hotel, and Car Rental company. The simulation scaled up to 50 entity tasks (expanding to 100 for Reserved Protocol tests) across a hybrid 3G/Wi-Fi network with heterogeneous hardware.
Key Results
- Success Rate Decay: As the number of services in a composition increases, success probability drops exponentially. In the NFR model, the success rate approached zero once the chain exceeded 60-80 nodes.
- Pattern Performance: The Reserved Protocol (RP) significantly outperformed others, maintaining a success rate above 60% even with 200 service nodes, compared to the rapid decline of CR and TPS.
- Threshold Sensitivity: Transactions are highly sensitive to reliability thresholds; a minor dip in a single node's reliability can cause a sharp drop in the overall composite success rate.
Discussion and Conclusion
A primary insight of this work is that uncertainty is propagative: a single flaky service affects the transition probabilities of all subsequent states in the Markov Chain. The Reserved Protocol's advantage lies in its ability to "lock" resources early, mitigating risks associated with device mobility and power loss.
Future work will focus on integrating more nuanced semantic and context-aware descriptions (such as user intent) and developing dynamic adaptation mechanisms that allow systems to switch transaction patterns in real-time based on environmental changes.
Summary updated