Sunday, May 29, 2005

The Arrogance of Intellect

Yesterday I went to National Chiao Tung University (NCTU) for Ph. D admission oral exam of the computer science (CS) department. By the end of it, I was disappointed, shocked, and had lost my respect for one of the professors.

The oral exam was divided into four 10-minute sections, each of which had a different focus and two professors doing the interview. My first section was on Motivation and Prospective Future Research. The young professors seemed impressed when I talked about my early research experience, my plan to build a massive simulator using peer-to-peer (P2P) computing, and the long-term goal to start up a company. That was the only section in which I felt my value as a person was appreciated.

The next two sections were on Mathematics and Information. I had been weak in math and unfamiliar with algorithm theory. So I had some difficult time answering even basic questions posed by the professors. I would be willing to take the responsibility myself, for not having studied enough the fundamentals in computing theory. However, I was unable to answer mostly because these algorithms or mathematics were unrelated to my present research, so I did not possess a good working knowledge. On the other hand, I was confident that I would be able to answer if they were relevant to my current research. However, one professor in the Mathematics section won my respect when he tried to rephrase several times the questions and provided guiding clues as to test the degree of my understanding as opposed to a simple yes/no of my ability. I could feel that he intended to teach and help a student, as opposed to intimidate or to embarrass.

The final section was on Master Thesis, in which the examinee had 5 minutes to explain his or her master thesis and 5 minutes to answer questions. Having devoted the last 2 years of my life on researching highly scalable virtual environment (VE) systems, published in the affiliated workshop of ACM SIGCOMM (one of the largest conferences in the field), and submitted a paper to IEEE Network Magazine (one of the leading journals in computer science, with an impact factor of 3.871 in 2003), I was confident of doing an interesting presentation to demonstrate my real strength: a project-based learner who has the interest and inclination to do serious and independent research.

However, much of what happened next shocked me, and I was puzzled for the rest of the day. I began the presentation by explaining that I intend to build scalable virtual environment system and defining the problem. But after I mentioned that I used P2P computing (a kind of novel networking method that has given birth to new applications used by millions of simultaneous users), the professors quickly denounced: “that’s absurd!” and “how is that possible?”

They began by saying it was “impossible” to synchronize the topology in such environment (which was in fact possible with proper control), questioning if I had done simulation or not (which I had), and asking why I made certain design decisions (which I had to admit that I probably did not answer well due to my slower response caused by the intimidation). One of the professors was opening and closing his eyes, leaning back and forth, to express his impatience as he listened to my explanations.

Then, as I explained that the topic was chosen because “P2P can potentially be more scalable than client-server architecture”, that professor quickly remarked: “P2P is the least scalable (emphasized by his tone) architecture in building large systems!”

I listened to his remark in shock, while trying to think of a response in panic. As the premise of my work was that P2P can be more scalable than client-server, if his comment was true, the value of my entire research would be invalidated. In fact, that seemed to be what the professors believed when the section ended, and their faces had the expression: “your research is ridiculous and absurd!” Though I had some diagrams in mind that might help to clarify, my time was up, and the professors wanted me out of the room, expressing no interest in listening to any further explanation.

As I walked out, it just came to me how pathetic the professor’s comment was -- not just the rudeness and unfriendliness of criticizing without an effort to understand, but because of the sheer and plain ignorance behind that comment. While P2P was, in the past, associated with a fully-connected architecture, which would not be scalable, during the past few years, research efforts from talented people have focused on building scalable P2P, where each node only connected to limited number of other nodes to avoid overloading any particular node. With the introduction of worldwide file-sharing P2P networks already used by millions of users, the scalability of P2P systems was obviously no longer questionable by any respectable researcher. A blanket statement such as “P2P is the least scalable architecture” only indicates a lack of knowledge and familiarity with recent P2P development and research during at least the past two or three years.

In some sense, what that professor did was not just insulting my intelligence and efforts, but disregarding the reviewers of my previous publication (in ACM SIGCOMM’s affiliated workshop), the authors of other similar publications (which appeared in leading conference such as INFOCOM), and the judgment of IEEE Network’s Editor-in-Chief, who sent the manuscript of my thesis’s work for outside peer review. Had my research idea been so absurd and worthless, it should have been turned down by the many experts who reviewed such works a long time ago.

When the two professors left the room for break, the other professor told me “he is the expert,” in the sense of telling me to get lost and stop wasting my time on dreaming on impossible designs. I could sense that they also questioned how my advisor had allowed me to graduate with such an “undoable” thesis topic.

I tried to contain my puzzlement over his statement, and learn of the name of this “expert professor.” The first thing I did after getting back was to search for this professor’s publications and works, to find out how come after two years of studying the scalability problem in virtual environments and conducting extensive survey on existing literatures in various fields, maintaining contacts with both academic and industrial people, that I have not even heard of this expert professor’s name.

And this was what I found: the “expert professor” had in indeed built a platform that was not well-publicized and on which no real system was built; yet he published not even one paper about this infinitely scalable platform (in the words of the promotional document). He co-authored numerous papers on various topics unrelated to virtual environment, but had not been a first author since 1995, the year he became a full-professor. Among journal articles he published recently, the 2003 JCR impact factors were between 0.2 and 0.5. In other words, he had not been academically active for the past 10 years and had not published in leading journals or conferences.

Probably not a big surprise, his platform used client-server architecture, which my work posed as inherently less scalable than P2P.

I felt humiliated, puzzled, and a bit sad at how things turned out. Certainly this “expert” professor had lost my respect, and the experience also changed my perception of NCTU’s CS department, the most prestigious engineering school in Taiwan, for assigning such professor to conduct important admission interviews. The professor was obviously a learned man, so his ignorance was likely not due to the lack of ability but something else – perhaps arrogance?

Friday, May 06, 2005

The Ultimate Simulator

This will be my first blog. I found this service after reading an article in Fortune magazine describing the war between Google and Bill Gates.

Yesterday I talked with two friends about the possibility to use peer-to-peer (P2P) computing for protein folding simulation. The idea was simple: connect large number of volunteer computers in a P2P fashion, and each node is responsible to model one atom in a protein molecule. As the simulation goes, each node would constantly discover neighboring atoms within a pre-specified radius and would communicate the most recent coordinates with those neighbors within the "interaction sphere".

One friend expressed concern about the potentially large amount of communication required to keep the various nodes synchronized, while the other friend thought the idea was "very interesting."

I, on the other hand, was just pondering on the intriguing possibility: if protein folding were to be simulated in the most parallel fashion, how much more can you get than each atom is simulated by ONE processor? And if such thing does work, wouldn't it be in effect, the ultimate simulator?