今天听了一个关于parallel computing的讲座, 觉得自己越来越落伍了, 而且不只是落伍, 还把以前学的忘得一干二净. 几天前收到CCR的信得知这个讲座, 查了一下演讲人的资料, 是个强人, 决定去长一长智, 顺道也复习一下以前学过的. 今天虽然她presentation的organization不算很好, 但是也至少把general idea传达给观众了. 不过观众群中很多也对parallel computing一知半解, 也奇怪那些天天做parallel computing研究的人那个时候去了那里, 搞得演讲的需要"降一降级"就是一些基础才能继续. 更奇怪的是机械系的人来得最多, 也许刚刚好排在平时seminar的schedule, 很多人以为是MAE的seminar series吧.
上HPC的时候, 明白了一番computer architecture, processor之间的communication算是有点简略带过, 可能这门学问真的很大, 尤其在computer science里, 之前就大概知道一些简单的topology. 今天收获可算大了, 虽然自己不在这方面继续研究. 演讲人做了一个framework涵盖parallel computing的networking问题. 她最大的motivation还是有heterogeneous processors的系统, 和这些系统的geographical location的问题, 然后怎样用数学模式predict这些computation问题. 我不是computer scientist, 所以很多还是听不进去, 就知道是个很难的问题. 其实也是, 因为我们机械工程的, 其实也不会想到networking那么长远的问题. 也不是因为没想到那么长远, 而是因为要把手上的问题parallelized都是一门学问. 然后还要去做task scheduling已经很难. 能够在计算上省下大概30-40%的时间应该很好了, 没有会在继续想能不能再省计算时间的地方的. (Muthukee可能应该可以给我input关于这些的一些comment吧?) 不过还是很多人忘了一点, 就是parallel computing的重要目的是省时间. 如果花更多时间去减少计算时间, 有时很不华算. 除非是一个每周要运行一次的计算, 而且是10年的project, 那花个2年来研究还算有道理. 可以说, 我们还没有到那个境界吧?
其实, 值得一提的还是heterogeneous network的issue, 尤其提到network backbone, 还有technology advances, 这是很practical的问题. Networking的protocol和medium还在持续发展, 现在wireless networking也越来越快, standard也一直在更新, 有时parallel codes也要能够跟上潮流才能efficient. 还有更实际的processor的technology, 一下就duo core, 一下又duo core 2了, 迟一些可能每一台电脑都有自己的parallelization的能力, 只是看algorithm有没有办法好好利用这些资源了.
还是有很多还会持续发展的, 不跟上脚步真的会落伍的. 不过有时也不一定. 几天前还在用DOS boot一台PC/104, 然后可以做real time control. 还好小的时候玩DOS也玩得还行, 没有把它忘得一干二净. My point is… 有时low level的东东还是要会一些的, 不然high level的不work, 不能够就不能troubleshoot而逃避问题的. 总的来说, 现代的工程师好像什么都要会一些, 不然很难保住饭碗啊.
P/S: 看看下面的trackback吧, 有一度我说过Pringle的事的, 呵呵…
I agree with what you said here, the powerpoint slides seem made in a rush… The interesting point is they are predicting the irregular gang activity in local area by inputing the culture, daily behaviors of those gangsters…I am not sure how accurate their results will be and how they validate them….haha, but the validation must be fun enough!! will they ask a bunch of gansters to hang around the streets and provide them enough money to rob, food to steal or car to ham???
Haha! Yeap, validation is always the fun part in research. I’m not sure how exactly they verify mathematical model in social activities, but I do hope that they kept some database from the previous criminal cases. If they’re serious about it, I’m sure they kept all the investigation files that they’ve made, and somehow verify them. But, I do admit that they are extremely "limitted" and case dependent. One way is to just use the current hypothesis, and try to investigate whatever that may happen in the direction of proving the hypothesis. However, this may itself lead to another area of research, in which even math or AI can’t explain them. Of course another problem (as happen often in any design problems) is that we would create faults to validation that the solution is faulty – too dangerous. As you said, won’t be asking the gangster to take vandalism in action and have residents experience such hardship.
Hehe… don’t get too much involved, you have your "dots" to take care of. But, it’s your research being oriented to such "psychological science". Haha! Good luck to you!