Reducing User Perceived Latency with a Proactive Prefetching Middleware for Mobile SOA Access

Abstract

Network latency is one of the most critical factors for the usability of mobile SOA applications. This paper introduces prefetching and caching enhancements for an existing SOA framework for mobile applications to reduce the user perceived latency. Latency reduction is achieved by proactively sending data to the mobile device that could most likely be requested at a later time. This additional data is piggybacked onto responses to actual requests and injected into a client side cache, so that it can be used without an additional connection. The prefetching is done automatically using a sequence prediction algorithm. The benefit of prefetching and caching enhancements were evaluated for different network settings and a reduction of user perceived latency of up to 31% was found in a typical scenario. In contrast to other prefetching solutions, our piggybacking approach also allows to significantly increase battery lifetime of the mobile device.

A Reference Architecture for Mobile SOA

“…Hence prefetching improves availability and performance of a system, prefetching, together with service advertisement, can eliminate service selection time from the user's point of view improving its efficiency and usability. This pattern group (prefetching) consists of the following patterns: peer‐to‐peer prefetching based on statistical data [Chang, Ling, and Krishnaswamy, ], consumer side prefetching [Liu and Deters, ] based on workflows defined by Business Process Execution Language (BPEL) files, and provider side prefetching based on statistical data [Natchetoi, Kaufman, and Shapiro, ; Liu and Deters, ; Schreiber et al., ], data mining [Tseng and Lin, ], and proximity relations [Eikerling, Benesch, and Berger, ]. In provider side prefetching, prefetched service results and computed hit ratio can be sent during a connection's idle time [Natchetoi, Kaufman, and Shapiro, ] or by piggybacking [Schreiber et al., ]; the latter makes a trade‐off between energy consumption and response time [Schreiber et al., ].…”

“…Preprocessing of service requests (filling default parameters by proxy [Schreiber et al., ]) and filtering service invocation results [Schreiber et al., ; Hamdi, Wu, and Dagtas, ] are processes dealt with by pattern d .…”

“…As mentioned, this layer has no corresponding layer in the SOA RAF; some responsibilities of this layer may be realized as a middleware [Schreiber et al., ]. As patterns associated to this layer (see Table ) imply two main responsibilities the first is to optimize resource consumption based on predictable sequences and the second is to adapt service interactions to enhance a user's experience and to provide service continuity.…”

“…(4) Data achieved from public services such as regulations [Sefid‐Dashti and Habibi, ]. This layer uses data from various sources to make predictions for service requests and to prefetch it into a consumer's mobile device [Chang, Ling, and Krishnaswamy, ; Eikerling, Benesch, and Berger, ; Liu and Deters, ; Natchetoi, Kaufman, and Shapiro, ; Schreiber et al., ; Tseng and Lin, ]. It also uses data to provide or generate UI [Broll et al., ; Sefid‐Dashti and Habibi, ] appropriate for the facilities of the mobile device, a user's personality and the environment [Sefid‐Dashti and Habibi, ].…”

“…Furthermore, the consumer layer is coordinated with the Prediction Methods and Statistical Data Layer , to provide logs of cache access and logs of chosen options of the UI to improve accuracy of prefetching and accuracy of generating UI, respectively. These feedback data can be sent by piggybacking [Schreiber et al., ] or in times of an idle connection [Natchetoi, Kaufman, and Shapiro, ].…”