Hej! Jag har kollat alla bidrag och tycker att #1 är klart bäst, tack vare sin originalitet, användbarhet och välgjorda presentation. --------------------------------------------------------------------------------------- 1 - How do you classify this thesis (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent)? 3 - Above average 2 - What is your overall expertise concerning the subject areas of this thesis? 2 – I am knowledgeable in the area, though not an expert 3 – Technical strength (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent) 2 - Average 4 – Presentation. Is the thesis well written? (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent) 3 - Above average 5 - Summary of the thesis (Short summary of the thesis) The thesis treats the problem of controlling the update procedure of a database with real-time data when the workload is varying. Simple models for database transactions and performance metrics are used. These allow the problem to be studied using standard methods from feedback scheduling and control theory, though the author choose a soft computing approach denoted "Imprecise Computation". Evalutation of a couple of schemes are done through simulations. 6 - Points in favour and against (also sent to the authors) The interesting application of scheduling transactions in real-time databases is considered, which seems not to have been extensively studied the literature. It's not clear from the thesis, however, how far the presented model is from a real database application. For example, to me, the considered steady-state and transient-state performance metrics seem quite hard to apply in practice. Also, the approach of "Imprecise Computation" is not logic to me, following the earlier discussion in the thesis on feedback control. Why not try a classical control approach first? -------------------------------------------------------------------------------------------------------------------- 1 - How do you classify this thesis (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent)? 4 2 - What is your overall expertise concerning the subject areas of this thesis? 1 – I am an expert,2 – I am knowledgeable in the area, though not an expert, 3 – I am not an expert. My evaluation is that of an informed outsider 3 3 – Technical strength (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent) 4 4 – Presentation. Is the thesis well written? (1 – Weak, 2 – Avarage, 3 – Above avarage, 4 - Exellent) 3+ 5 - Summary of the thesis (Short summary of the thesis) An approach for optimizing performance of a real--time database is presented. Data quality and transaction response time are balanced using a feedback control scheduling policy. Two job queues are used for the scheduler, one for database update transactions, and one for user transactions. Database update transactions are only commited if they are needed based on a variable (based on target utilization) data error criterion: A new value is only written to the database if it deviates more than a variable limit. User transactions are only scheduled if the target utilization allows. The feedback loop uses two controllers, one based on a deadline miss criteria for the database transactions and another based on target utilization. The control signal is chosen such that it satisfies the currently most demanding criterion (the minimum of the deadline miss control signal, target utilization control signal and their sums). User transactions may have optional subtransactions that increase transaction quality. These contribute less to the deadline miss criteria and are therefore less significant. Two algorithms are proposed to maintain Quality of Service and Quality of Data. One using the miss percentage controller (alg. 2) and the other using both miss percentage and utilization controllers (alg. 1). The algorithms are experimentally evaluated and it is concluded that the algorithms are effective in maintaining a specified QoS and QoD level of the database. Alg. 2 has better average performance while alg. 1 is better at avoiding temporary "overshoots" in performance. Several ideas for future research directions are presented. 6 - Points in favour and against (also sent to the authors) + Contributions * Proposal of two control strategies for load balancing of RTDB One based on utilization target One based on deadline miss percentage * Proposal of algorithm for predicting utilization for a given allowed data error * Experimental evaluation of strategies + Logic approach with background, problem description, solution and evaluation +Generally a very good presentation with perfect language and a scientific style. The thesis would fit as a conference paper (subject to research contribution strength, which I cannot finally judge) +Forefront research (based on the date of the referenced papers) +Good treatment of references -It took some time before it was explained how the control theory of Chapter 2 were to be applied in the RTS scheduling context -Many terms and abbreviations makes the reading difficult (to some extent inherent in this kind of study) -The assumption seems to be that computation of data items and comparison with database content is less time consuming than a database write. This is not clearly identified, and not justified (i.e. is there a gain in skipping the database update step when the value is already computed and compared with the current database value) +/- Narrow research problem makes the project perhaps less challenging, i.e. the problem formulation phase is simpified --------------------------------------------------------------------------------------------------------------- 1 - How do you classify this thesis (1 - Weak, 2 - Avarage, 3 - Above avarage, 4 - Exellent)? 4. Well worthy of winning. 2 - What is your overall expertise concerning the subject areas of this thesis? 1 - I am an expert,2 - I am knowledgeable in the area, though not an expert,3 - I am not an expert. My evaluation is that of an informed outsider 3 3 - Technical strength (1 - Weak, 2 - Avarage, 3 - Above avarage, 4 - Exellent) 4, as far as I can judge 4 - Presentation. Is the thesis well written? (1 - Weak, 2 - Avarage, 3 - Above avarage, 4 - Exellent) 4. Very readable, although it will probably take a couple of weeks to get through the formulae... :-) 5 - Summary of the thesis (Short summary of the thesis) It describes the problem of getting an RTDB (i.e., a DB that holds not just static data, but data that are defined as time series) to respond within the timing constraints of a real time system. It describes a possible tradeoff: you can get a quick answer, but you can keep calculating and get a better one. The thesis proposes two algorithms to manage this, and puts it in the context of control systems. 6 - Points in favour and against (also sent to the authors) Very comprehensive coverage. One of the most ambitious works I've seen at this level. The only thing I'm missing is that I would have liked a practical application example (RTDB applied to a real world problem), to put things in context.