------------------------------------------------------------------------------ The discussion of periodic, aperiodic, and sporadic seems to be a combination of frequency concerns and demand/usage concerns. I'd like to see these issues separated more clearly. A Bayesian approach is briefly mentioned as "more precise that all the above". Please provide more information on this. How does it work, why is it better? Is it hard to implement? Should all other scheduling algorithms be abandoned? When should the scheduling algorithms (RM, EDF) be used. I assume that rate-monotonic is only appropriate for highly dynamic task arrival situations, since EDF always does better. The 70% limit on usage in RM seems to correspond to a well-known limits based on queuing theory for p robabilistic arrival times and service times. You discuss several different scheduling systems: MMOSS, SFQ, MTG, RT-Mach, Chorus, others? The paper would be easier to read if the transitions between these systems was more clearly indicated. Section breaks would be nice, as would introductory discussion of these systems. The half paragraph discussion of DSPs and their OSes seems out-of-place. I'd recommend expanding it or dropping it. The MTG discussion begins with the claim that other scheduling systems ignore precedence relationships. This was an odd claim following the SFQ discussion - to my thinking, the hierarchical system of SFQ is a precedence relationship. If the two systems are taking a different view of precedence, the difference should be made explicit. MTG claims that the utility function is "monotonically decreasing" and therefore more delays mean less quality. For this to be true, the utility function would need to be "strictly decreasing", a stronger requirement. Several of the paragraphs were extremely long and presented multiple concepts. Often you present a lengthy description of a problem and a substantial description of its solution in the same paragraph. Breaking these up would increase the readability of the paper. For example, pg. 2, second column, this sentence "A solution to this problem ... PCP ... PIP" makes a great paragraph break. I prefer citations to include the name of the author or the work. For example "Heidelberg [2] and Yartos [4] are examples of such systems." rather then "[2] and [4] are examples of such systems." It makes it easier to follow references that re-occur and I may be familiar with the heidelberg projects. Page 5 would _not_ print correctly on my system. Please check your formatting system to verify whether you are generating overly long pages. ------------------------------------------------------------------------------ Explanation of the task scheduling problem was thorough (Section II). Some of the examples were difficult to follow (i.e. the RT-Mach resorce reservation model). ------------------------------------------------------------------------------ Import 1 - 7 6. There's tons of papers about handling multimedia on the OS out there. Novelty 1 - 7 1. Absolutely nothing novel here. I don't believe the word "we" was ever used. This paper seems to be simply a re-stating of ideas presented in the set of papers, and little else. Quality 1 - 7 3. Long paragraphs, with little structure. Also, there was no salient point to be gleaned from the paper.Finally, certain pieces of classic OS were re-hashed in detail where there was no need to. Overall 1 - 7 3. Commentary: I had a tough time with this paper. Besides the grammar problems, there were five main problems: 1) (Most importantly) No salient "point". The paper listed a set of algorithms for handling multimedia task scheduling and some applicable areas, period. What was the point of this? Also, where was the writers' input on the subject matter? I had a hard time finding any opinions. 2) Underuse of references. There were many statements made which had no reference to where the ideas came from. Were these all original ideas, or were they from papers? For example, section IV.2 is almost devoid of references to either the authors or papers, but many assertions are presented. Also, on that topic, there were a couple bald assertions, e.g. pg 5, end of paragraph 2:"In a system really safety critical it might not be a good idea mixing different kinds of tasks in a processor." with no explanation as to why the authors felt this was so. 3) Confusing structure. The topics were not presented in a heirarchical manner; I was constantly looking back into the paper to figure out *why* I was reading what I was reading. For example, Section II spends almost 2 pages talking about classic task scheduling, without a single mention of Multimedia tasks. All the while I'm wondering, "what does all this have to do with anything?" The next section uses some of the material, but it does not flow well. In general, I felt that there was a lot of material, but not much of a train of thought behind it. 4) Unnecessary/Irrelevant expansion of topics. Especially the bit on pg 2, first paragraph (Processes are...) and the example in pg 2, second paragraph. Because these ideas are standard OS theory (i.e. "left as an exercise to the reader" standard) a sentence or two describing the idea would have sufficed. 5) Lots of technical blips: o In the introduction, there are 2 Section IVs mentioned. o Figure 3 is not a directed graph, although it is described as such. o Projects aren't mentioned by name at least once. o Paragraphs over a page long! No relief for the reader. ------------------------------------------------------------------------------ The author present (well) what the problems are (Introduction) Following this, the general task scheduling problem is presented (a nice presentation of what scheduling is) (Section II). In the same section real-time scheduling is introduced. Section III deals with multimedia tasks profiles. It identifies general characteristics of multimedia tasks, and introduces concepts nedeede to understand multimedia tasks scheduling. Section IV gives overviews on two classes of schemes of scheduling found in multimidia aplications. The Conclusion is given in Section V. Good points: Good english A newbie in the field finds out interesting things. Comments: The paper is a bit difficult to read. The main reason is that the paragraphs are too lengthy. I don't know what to recommend, but to make a better survey out of this, lots of stuff could be set aside. It still depends to whom it is addressed. This being a survey, one of the main weak points is that the authors focus too much on some scheduling schemes, while just mentioning other. The main problem with the paper is that no discussion is given in which different approaches are analyzied with respect to each others strong and weak points. In particular, section V restates known facts. ------------------------------------------------------------------------------ This is an excellent paper and I highly recomend it to be presented. The paper as a whole had excellent writting and was well structured as a whole. The reader is never left guessing as to where they are being led. The problem is laid out clearly at the begining of the paper with a nice example problem. The descriptions of Rate Monotonic and Earliest Deadline are well written. The were quite a few critiques, most of which were quite sound. They were almost always supported. Some of the problems with the paper are: - In equation 3, the Bi is never explained. - Figure 1 is pointless - Integrated Scheduling should be busted up into smaller sections - ( the IV.2 subsection header is kind of funny ) - The conclution was a bit weak. After such a great paper it would have been nice to have a great conclution. ------------------------------------------------------------------------------ Very difficult to read. The paragraphs were very long and I couldn't grasp what the key points were (if there were any at all). It looked like plain summary to me. In the algorithms section, there was just explanations of how the algorithms worked, I didn't see any analysis of the author's work. The english was good.