Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/82888
Title: A novel dynamic cache size tuning mechanism to support fast response on the Internet with applications to mobile and telemedicine
Authors: Wu, Sui-lun
Degree: Ph.D.
Issue Date: 2006
Abstract: The aim of this thesis is to propose a framework for maintaining a minimum cache hit ratio in a dynamic manner. By doing so the service roundtrip time in a client/server interaction over the Internet is shortened. This is achieved because the dynamic cache size tuner obviates the second leg in the information retrieval process. The final framework proposed in this research after considerable verification is the MACSC (Model for Adaptive Cache Size Control). This MACSC research has achieved all the objectives, namely: a. Proposed the framework that leverages the minimum number of network parameters for accurate dynamic cache size tuning. b. Proposed compensation measures that help the framework work healthily under extreme conditions. c. Proposed a RTPD mechanism that is suitable for time-critical applications so that it can be included as part of the self-reconfigurable framework for dynamic cache size tuning on the fly. d. Verified the generic framework with simulated and live datasets/traces. e. Verified that the proposed framework can indeed support wired and wireless client/server interactions with similar efficacy. f. Verified that this efficacy applies to mobile and time-critical applications such as telemedicine. There are four approaches proposed for realizing the MACSC framework, which leverages the Zipf-like behavior as the conceptual basis. This behavior represents the relative data object popularity profile from which the popularity ratio is derived on the fly. This ratio determines the size of the dynamic cache size adjustment. The four approaches dictate how the standard deviation and the popularity ratio of the popularity distribution should be computed, and they are as follows: a. Point estimate (PE): This CLT (Central Limit Theorem) based method is sensitive to changes in the ROP profile, but it generates a lot of oscillations for having no feedback system. b. M3RT: This CLT based technique is transcribed from another problem domain, namely, Internet End-to-End Performance Measurement (IEPM) [Cottrel99], [CottrelOl]. Previous experience [Ip03] has confirmed that it always yields the mean of any waveform accurately in real-time applications because it has a feedback loop. The good quality of stability of this technique, however, becomes a liability for MACSC application because there is a need to strike a balance between stability and sensitivity, and this led to the proposal of the "fine-tune point estimate (F-PE) technique. c. F-PE: It combines the merits of PE sensitivity and M3RT stability and accuracy due to the presence of a feedback loop. d. Real-time Traffic Pattern Detection (RTPD): It was observed in the early experiments that the Internet traffic patterns have different ill effects on the MACSC accuracy. Therefore, a RTPD mechanism is included in the original MACSC framework to make it into the newer RTPD/MACSC framework. The aim is to let the RTPD mechanism identify the traffic pattern at the time so that the MACSC can "reconfigure" itself to neutralize the traffic ill effects. In the MACSC research context, the RTPD/MACSC(PE) solution is a form of reconfigurable dynamic cache size tuning. As a result four specialized solutions are derived from the general MACSC framework. The specialization is the technique whereby the SD is computed on the fly in each solution. Therefore, the four novel proposed conceptual solutions, which represents an evolutionary process, are: MACSC(PE), MACSC(M3RT), MACSC(F-PE), and RTPD/MACSC(PE).
Subjects: Hong Kong Polytechnic University -- Dissertations
Internet
Cache memory
High performance computing
Pages: xx, 201 leaves : ill. ; 30 cm
Appears in Collections:Thesis

Show full item record

Page views

50
Last Week
1
Last month
Citations as of Mar 24, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.