I suspect that on average the length of a thesis is inversely correlated to its brilliance. Computer science is a fairly established field at this point, so a mediocre thesis likely describes a large number of small improvements upon the status quo - along with charts, graphs, and explanations that compensate for lack of a single revolutionary finding. If someone can prove that P=NP (or that P!=NP) in 13 pages, there's no need for another hundred pages of explanations. On the other hand, improving the performance of Linux page cache by 5% on some workloads will clearly be accompanied with a hundred pages of methodology explanations and performance measurements.
Interestingly there are lots of long and brilliant dissertations - Chris Okasaki's functional data structures thesis comes to mind. However, I doubt there are very many (if any) short mediocre dissertations - in case of mediocrity lack of content must always be compensated by size.
Systems research is easy to criticize, but much harder to do right.
For starters, systems research deals with the real world. The real world is messy. How would you comprehensively represent every interesting workload in an analytical model? How would you derive it from real world applications? It is impossible in the general case (halting problem), and really, really hard to do even in specialized cases. If you think otherwise, the field of worst-case execution time analysis awaits your contribution eagerly (try modeling L1 and L2 instruction and data cache interactions in a multicore CPU).
Thus, in many cases evaluating different interesting workloads empirically is unavoidable. In such cases, 100 pages of setup, methodology, and analysis are a feature, and not necessarily a sign of mediocrity. There is a great danger of overlooking substantial flaws in brief descriptions.
So, yes, mediocrity can lead to inflated sections, but a good PhD committee will not led that slide. In your page cache example, I would absolutely expect to see significant experiments and analysis; I would probably not be convinced by just a few selected benchmarks. Presenting benchmarks well requires many more pages than a succinct proof might.
I'm probably biased because I do what might qualify as systems research for a living. It's hard to do right - much harder than a novice might think, but it's no rocket science. A reasonably intelligent person who has the stomach for rigor can be taught to do it well. However, I cannot easily get into computation theory - I tried my hand in it, and I just don't have the mathematical intuition. I could develop it to some degree with a lot of work, but I could never do first rate complexity research - that stuff requires real talent. I suppose it's common for people to slightly look down on what they do because they've experienced the mess first hand - the magic is always more spectacular on the other side of the fence :)
Saying you don't have a "talent" for something is another way of saying you don't enjoy doing it. If you don't enjoy it, you won't practice it, and if you won't practice it you won't be good at it.
I've known CS-theory people - who I know are very bright - say similar things about systems research.
Actually, I really enjoyed studying complexity theory. It just took far more effort than anything I've ever done before - it's really mind bending, and that's why I loved it. I just know that to get to the real meat, I'd have to work very hard for five years or so to develop the intuition and to get up to speed, and even then I'm not sure if I'd be able to cut it to do real first-rate work (a cop-out, I know).
I agree with you that people, especially laymen, tend to overrate talent - it's a good defense strategy that helps people avoid doing difficult things. But I disagree with you that there is no natural talent or inclination towards certain subjects in general. When it comes to hard math, I know people that just pick it much quicker and easier than I do, and it has been the case since I was very young (long story short, for a few years I went to school for gifted children, and some kids were just better than me no matter how hard I worked). I'm also convinced that some fields require more natural talent than others.
I agree. I wish the whole length == intelligence thing stopped after high school. In high school the teacher sets out a length that your essay needs to be and the student mindful of the length requirement fills the space with a few bullet points and copious amounts of BS which translates to a lifetime correlation for most people that length == A paper.
I think Pascal said it best: I have only made this longer, because I have not had the time to make it shorter.
This reminds me of a great essay, "How to Say Nothing in 500 Words or Less" - http://www.apostate.com/how-say-nothing-500-words. I think this nonsense should stop way before high school. I still remember the SPERM formula from my social studies class - social, political, economic, religious, and military. Pick three, write a paragraph on each, and stick them between an introduction and a conclusion (which is really an introduction rephrased). I learned more about good writing by debating various political issues on the gamedev boards I used to frequent than in any of my high school classes.
WEll, length == intelligence has been disproven by shamir and his paper about secret sharing. 2 pages of groundbreaking paper and the whole crypto world is shaking.
Most of the time, if people tell me to write more, I just ask them what they are missing, and either I can point them to a single paragraph where it is written or I need to write more, indeed. Most of the time, the other people give up and accept the short length after around 5-10 iterations :)
It's not a record by far, but my thesis of 82 pages (8 pages front matter, 68 pages contents, 6 pages references) is definitely at the short end of the spectrum -- somewhere around 0.5th percentile for Oxford DPhil theses according to my search of Bodleian records.
A think that this discussion is a little biased. For one hand, the form of a thesis depends a lot on the field. The more theoretical, the more it can be concise, but in some fields you need a lot of experimental data.
Another issue is that most dissertations are not made from a single, ground breaking contribution, but from several incremental contributions and you need to integrate them into a coherent text.
I've not finished my dissertation, but it fall in the latter case and I think it will be some 100 pages long.
By the way, it will be interesting to see the length of the PhD thesis of the people that claims that "the length of a thesis is inversely correlated to its brilliance" (only cperciva declared it)
Very equation heavy, which I think contributes to its brevity. Contrast that with some papers in other areas of math/TCS where you tend to be less equation heavy and more wordy. For example, lots of algorithms papers don't make as heavy use of equations (though there are still some of course).
The guy who wrote the thesis is the same who popularized a notation about the limiting behavior of functions which was introduced by the German mathematician Bachman. The notation is therefore called the Bachmann-Landau notation. Alternatively it is known as the Big-O notation.
can someone explain what the "theses" are on the last page of the thesis in the link? i don't understand how they are connected in any way to the rest of the work (i am not a mathematician). thanks.
I was wondering about the same thing. Here's the text:
------------
1. It is desirable during every existence proof of a mathematical quantity
to be led, at the same time on the way to the result, to the actual
existing quantity.
2. A boundary between arithmetic and analytic areas of mathematics
cannot be drawn.
3. The concept of the semiconvergent series is a relative concept.
4. Out of the impossibility of perpetual motion of second
kind comes the proof of the second law of thermodynamics.
5. It did not succeed, the justifying of psychology on an exactly mathematical
basis.
I've only ever written one undergrad thesis. The dry subject matter and the extended length of it makes me believe that the last eyes that will be laid upon it will be those of my thesis advisor.
Seriously though, why do research papers have to be so long? Verbosity makes for a high barrier to entry and it is unnecessarily boring for everyone involved.
Educational institutions could learn a thing or two from the 5 W's of reporting.
Interestingly there are lots of long and brilliant dissertations - Chris Okasaki's functional data structures thesis comes to mind. However, I doubt there are very many (if any) short mediocre dissertations - in case of mediocrity lack of content must always be compensated by size.