This dissertation has addressed some key issues in the design and implementation of Web Service messaging frameworks in mobile computing environments including the reduction of message size while preserving semantics and the proposition of new communication mechanisms to better utilize of the high latency and narrow bandwidth wireless network. We argue that the conventional SOAP-based Web Service messaging through HTTP fails to deliver a sufficient performance model in order to integrate mobile devices as Web Service participants. Particularly, applications on mobile devices which have a frequent message exchange ratio suffer more. This fact creates an increasing need for an architecture which supports more efficient data representation in message exchanges and a high performance communication channel.
In this dissertation, our primary goal is to design and develop an overall system framework, which supports an efficient Web Service message exchange communication
architecture, applications or Web Service participants exchange messages in an optimized streaming fashion, and a representation of those messages are flexibly chosen by participants through the negotiation at the beginning of the stream. This flexible representation is achieved by distinguishing between message semantics and message syntax.
We have developed the HHFR prototype and have presented the details of the implementation in this dissertation. We chose Java as the language platform for both mobile and conventional computing sides, but the architecture is not limited to any specific language platforms and can be applied to any message communication between heterogeneous platforms. The prototype implements three distinct things: a) “the interpret-style stubs” to encode and decode messages, b) the high-performance communication channel, and c) the Context-store to store meta-data of a message stream. We presented a description language, the Simple_DFDL, which is a small subset of the XML Schema Definition (XSD) but has a few additions. The Simple_DFDL is used to describe the data structure and types and we have demonstrated the successful automatic data conversion between a descriptive format (i.e. SOAP) and non-descriptive format (i.e. binary) using Simple_DFDL description and “interpret-style stubs”. We explained how this can be generalized to DFDL when this is deployed.
The message stream in HHFR framework is achieved in two levels. A semantic level message stream can be achieved by exchanging message in a flexible representation and storing redundant / unchanging message parts in the Context-store. Messages in the stream which shares meta-data (e.g., a message representation, message parts, and stream characteristics) will be related each other.
The primary purpose of using the Context-store in HHFR framework is to provide the Web Service Database semantics. However, since an application of a participating node can retrieve stream meta-data, the use of the Context-store also guarantees a semantically persistent recovery from the connection or node failure by building the semantically same message stream from the disconnection.
We showed that the negotiation using SOAP messages is interoperable with the conventional Web Service messaging paradigm and is a sufficient mechanism to set up an optimized high-performance communication channel. If a responding participant to a negotiation request SOAP message responds with a negative value on the HHFR capability. Or the participant responds with a SOAP fault message, the negotiation initiator will fall back to conventional messaging.
We have evaluated the performance of HHFR messaging through two benchmark applications. We presented our system models and examined the analytic cost models for both HHFR and conventional messaging. Our observed empirical results show that the breakeven point of the two benchmark applications is found after only a few messages are exchanged between participants. Applications in the HHFR significantly outperform the conventional messaging framework after the breakeven point. We found the only major overhead in the HHFR messaging is the negotiation overhead. Additionally, we evaluated the Context-store access performance through our mobile interfaces and presented the scalability analysis of this service, which can serve up to hundreds of mobile sessions concurrently.
We have argued that it is critical to address the performance issues in mobile Web Service messaging at the system level rather than in small pieces. Integrating small
solutions in pieces may produce additional problems in scalability, robustness, or performance. We claim that the HHFR architecture addresses issues at the overall system framework level and provides an alternative message serializing mechanism: a description language, communication channel options, and interfaces to the Information Services in order to store meta-data, all in a single design. Our flexible representation messaging approach is more efficient for mobile applications, which exchange a series of messages, even while it is interoperable to the conventional Web Service messaging.
Efficient Web Service messaging is essential for mobile applications despite recent improvements in device specifications and infrastructure since the gap between wireless and wired computing environments still exists and won’t be closed easily for technical and economic reasons. Our experience and evaluation demonstrate that the HHFR messaging approach is efficient in mobile Web Service messaging, which is exchanging a series of messages and helps to close the gap.