a discussion by JAM

As the volume of data to be stored rises inexorably the tools for data management are ever more in demand. New technology, such as Personal Digital Assistants combined wireless computing is placing greater expectations on the DBMS’s ability to manage very distributed data over less reliable networks whilst consuming less resources. Whilst there is still a need for centralised systems to evolve over time in terms of upgrades and adaptability to new technology there is a strong trend in nomadic computing aiming it way towards the future 'dream' of ubiquitous computing.

Ubiquitous computing is what the late Mark Weiser termed in 1988 as the third wave in computing. "First were mainframes, each shared by lots of people. Now we are in the personal computing era, person and machine staring uneasily at each other across the desktop. Next comes ubiquitous computing, or the age of calm technology, when technology recedes into the background of our lives". That is, computers will be embedded in walls, in tabletops and everyday objects and is seen as the great vision of computing.

However, there remain a number of problems with wearable computers. It is more difficult to maintain localisation information e.g. when one room is updated then all wearable computers need updating. Also where the application is controlling devices is often better to let the device drivers reside in the device itself and not with the wearable unit as this will save CPU, network and power resources. Further resource management amongst multiple users is still difficult. Systems such as Hive uses the best of both ubiquitous and wearable computing producing an agent-based architecture. However the architecture is limited in terms of its resource discovery and data discovery and would still rely on large amounts of centralised systems (e.g. DBMS to maintain accurate data, an important issue which they fail to address). It also has limits on its scalability and extensibility, flexibility and adaptiveness, and no measure of lightweightedness is provided. All these characteristics are necessary to realise Weiser’s vision of clam computing.

Nomadic and moreover ubiquitous computing needs database systems which are lightweight and customisable to combine high performance and power efficiency, yet flexible enough to adapt to a changing environment such as the ability to adapt. Current DBMS technology, having none of these qualities, needed to be reassessed and new architectures developed.This has lead DBMS researchers to look at extensible and micro-kernel DBMS architectures. However, I believe that we need to take these ideas further. I believe the solution lies in component-based DBMS (CBDMS).

Current trends in Operating Systems have greatly influenced my work in this area. A high-speed history is thus:

• Until the 1980s almost all OS were inflexible in nature with all systems services incorporated into a monolithic kernel.
• In the mid-1980s, microkernels were introduced as a result of the spread of high-speed networks bringing higher demand from distributed systems users. Microkernels deployed object oriented software and were relatively lean and efficient, realizing only the necessary OS functions
• By 1994 it became clear that the requirements of all applications cannot be met by the OS in advance leading to what is termed extensible kernels.
• 1998-present has seen an interest in better performing OSs which aim to provide the lightweightness and customisablity to combine high performance and power efficiency and are yet flexible enough to adapt to changing environments. One example of this is our operating system Go!

an aside (Talking of adaptibility...meanwhile since 1990 processor manufacture research has been looking at Field Programmable Gate Array (FPGA) processors. These consist of a matrix of highly complex reprogrammable logic ICs and what is interesting is that they have the ability to dynamically reconfigure themselves. Taking this to an extreme, if we have hardware that can morph surely we should have software that can improve on this capability. Such technology surely is ideal for ubiquitous computing. )

• Commercial DBMS have matured to become large, expensive and lumbering pieces of systems software.
• However research has noticed this and has begun looking initially at microkernel DBMS.
• Further, more work has been done on extensible DBMS e.g.   Genesis [1988], Starburst [1990], Exodus [1990] and Monet [1995], Adaptive Component Architecture from Sybase,  Informix’s Universal server and P2 which aimed to be a extensible microkernel architecture.
• However there is still not a architecture looking at more adaptive and lightweight DBMS architectures and the only mention of components and DBMS is at the application level where DB components are distributed using CORBA.That is, these systems are not of the component granularity that we believe necessary for the optimal trade-off between highly tuned and highly reusable components.

To conclude, the introduction of components is advantageous for data management systems at two levels. Firstly, any DBMS that runs on a component based operating system will have the advantage that it can directly access the hardware resources bypassing the OS. Secondly, if the DBMS is broken into components, it then becomes more flexible, potentially highly tuned to particular applications, and able to share its components with other data management architectures.

If you are interested in any of this email me, all comments (even negative ones :-) ) are welcome.

(c) J.A.McCann. Please note that this discussion can only be used for review, criticism and as part of research activities. Using it for any other purpose without the permission of the author, may result in breach of copyright.