Applying FLC-based dynamic buffer size tuning to shorten the information retrieval round trip time in mobile location-aware environments

Abstract

This paper proposes to use the novel fuzzy logic controller (FLC), which is a dynamic buffer size tuning technique, to support more dependable location-aware information retrieval in a pervasive environment. Within a smart space of a mobile/pervasive environment a user may request service from the dedicated surrogate via a portable small-form-factor (SFF) device (e.g. PDA). In effect, the SFF is the client and the surrogate is the server, which may serve many clients at the same time. The one-surrogate-to-many-SFF relationship or asymmetric rendezvous easily makes the surrogate buffer overflow in the peak demand seasons. The request traffic to the surrogate is tied in with the transient traveler mass traffic through the smart space, for example, in an airport. The transit travelers may switch on their mobile SFF at will to demand service from the local surrogate. If the surrogate buffer overflows, the users do not benefit from the cyber foraging provided by the pervasive computing infrastructure in the background. The FLC adaptively tunes the surrogate buffer size so that it always covers the queue length by the given Δ safety margin. As a result the SFF-surrogate asymmetric rendezvous becomes dependable because the chance of buffer overflow is eliminated. The dependability leads to shorter information retrieval roundtrip time (RTT) and happier users.

Appendix: The essence of the convergence algorithm

The convergence algorithm (CA) is an internet end-to-end performance measurement (IEPM) technique. It estimates the mean service roundtrip time (RTT) of a logical channel quickly and accurately. The Java-based CA prototype, M ² RT, was verified and validated as a macro tool. In its macro form the tool must be installed at the two nodes that represent the ends of the logical channel. Micro IEPM tools differ by operating as a logical entity to be invoked anytime and anywhere for service by message passing. The positive mean message response time (M ² RT) experience led to the development of the Java-based micro CA (MCA) prototype: micro mean message response time (M ^3RT) [17]. The CA treats a traffic pattern simply as a waveform, and its speed and accuracy are independent of the traffic pattern because it is based on the Central Limit Theorem. The CA operation is summarized by (2) and (3), where M _i is the estimated mean of the distribution for the time window in which the F (flush limit) number of data samples is collected; F = 14 yields the fastest estimation convergence. The other parameters are: (a) M _{i − 1} is the feedback of the last estimated mean to the current estimation cycle, (b) m ⁱ _j is the jth sample in the jth M _i estimation cycle, j =  1,2,3,(F − 1), and (c) M ₀ is the first data sample when the CA has started. Figure 16 is an example in which M _i always settles to the average channel RTT of 480 ms in the steady state.

$$ M_{i} = \frac{{M_{{i - 1}} + {\sum_{j = 1}^{j = F - 1} {m^{i}_{j} } }}}{F}; $$

(5)

$$ M_{0} = m^{{i = 1}}_{{j = 0}};\quad i \geq 1 $$

(6)

Fig. 16

An example M _i of prediction for a TCP channel