Mohammad's Assignment 13

BUILDING AND MANAGING LARGE SCALE DISTRIBUTED SERVICES
BY
JIN LIANG, University of Illinois at Urbana-Champaign, 2007

Reviewed by: Mohammad Al-Mutawa

Mohammad's PhD Thesis review.pdf

Recent research in peer-to-peer and grid computing has made it possible to build Internet
Scale services such as content distribution, storage service, name service and publish/
subscribe. By utilizing large number of service nodes that collaborate in a decentralized
fashion, such services can potentially achieve high scalability, availability, reliability and QoS/performance. However, the design, implementation and management of such applications is a challenging process, due to the lack of design principles, software architectures, reusable code bases and powerful management tools.

In his dissertation, Liang addresses the challenges involved in developing infrastructure based large scale distributed services which is a special case of large distribute applications. Liang focuses on the design, implementation and management aspects of such applications. The dissertation goes into discussing and describing each aspect/phase in great details

For the design phase Liang introduces OCMA (Overlay Construction and Maintenance Architecture), a layered architecture for designing large distributed
applications. The great thing about OCMA architecture is that it is a general architecture applicable to both service applications and P2P applications. The layered approach of OCMA was inspired from the Internet and its success which is in large extent due to its layered architecture design. The OCMA architecture consist of 3 layer: a) Membership management layer, b) Overlay construction and maintenance layer, and c) Application specific processing layer.

For the implementation phase, Liang presents PPF ((Protocol Plug-in Framework) a reusable C++ framework for implementing large distributed applications. One of the very nice features about PPF is that it can run on both simulation and real world mode. In Simulation developers can simulate network topology and time, which makes debugging tasks much easier. The PPF framework consists of an event manager, a time manager, a socket manager, one or more peer nodes, and one or more protocol modules on each peer node. The event manager is the central component of the application. It consists of an event queue and a dispatcher. The time manager can return either the real time, or some virtual time, depending on the execution mode. The protocol module is the main component of the application, and it provides event handlers for timer and network events. The peer node is the component that implements network simulation.

For the third phase which is management, Liang presents Management Overlay Networks (MON), a simple and lightweight tool for managing large distributed service applications. MON uses a dynamic query (on-demand) approach to monitor the internal detail information about the application processes. Also for the management phase, infoEye is presents which is a system that can dynamically recognize application query patterns such as query arrival rate and attribute popularity and automatically configure itself to use either push or pull for different metrics in order to minimize information management overhead.

Over all, I found the dissertation clear and well organized. I liked the fact that everything was evaluated and supported with results from experiments. The thing that was missing is the security aspect, which is very critical especially when we are dealing with the global Internet; but then I found that it is included in the future work part of the dissertation.

Last modified 4 December 2007 at 11:37 pm by mohammad