Date: Tue, 05 Nov 1996 21:15:13 GMTServer: NCSA/1.5Content-type: text/htmlLast-modified: Fri, 13 Sep 1996 15:34:46 GMTContent-length: 13481<title>Glew's Thoughts about (Higher) Education</title><h1>Glew's Thoughts about (Higher) Education</h1><H2>Why</H2>I want to go back to school to finish my Ph.D.Why?<ul><li>Bob Colwell assures me that I don't need a Ph.D. to establish myposition in industry. He's probably right<li> Sometimes I dream about having a less stressful job, where I cando 40-50 hours a week of world-class work - rather than work 60-80hours a week as at Intel, spending most of that time doing cruft.  ButI'm probably fooling myself.</li><li> I'd like to take some time and think up new ideas.  Intel haseaten a whole slew of the ideas that I had stored up when I came outof my M.S., and working at Intel gives you very little opportunity forthe type of deep though that I, at least, need to come up with goodideas.  I am very much afraid that, if I do not go back to school, Iwill reproduce the job path of many of the people I see around me atIntel - some of whom are Fellows - who are now idea consumers, insteadof idea generators.</li><li> Of course, I have some hope that Intel's new MicrocomputerResearch Labs, of which I am a member, will provide the sort ofopportunity for deep thought that I think we need.</li><li> I want to work on one or two more chips, after P6, but then Iwould like to have the opportunity to "retire" to a university.</li><li> I would like some day to go home to Canada. I only meant to workin the U.S. 3 years. But, there aren't many jobs for computerdesigners in Canada, so having a Ph.D. gives me an option.</li><li>I left my Ph.D. because of my wife. Now that my wife has left me,I'll return to my Ph.D.</li><li> Because if I don't do my Ph.D now, I won't ever do it(assming I eventually have a family).</lI.<li> Because I like thinking deeply. </li><li> People used to accuse me of being too ivory tower - in fact, they    probably still do.  But at least, after 10 years in industry,     I have probably proven that I am not one of "Those who can, do; those who can't teach"    if I decide to become a professor.</li><li> Because I would like to do something more lasting for the world than just design    CPUs. Because I think that there are fundamental things to be learned from CPU    design that apply to other areas - like psychology, government, etc.</li><li> Because I want to have a role in designing the artificial intelligences    that will eventually replace mankind. At the moment, this most needs    faster computers.</ul><H2>Sob Story</H2>I sometimes feel that I missed out a lot in my university career.    <DL>    <dt>Undergrad - McGill <dd>	<p>	After discovering computers, learning how to skip classes,	and losing my scholarship, I worked my way through with	a succession of computer jobs, working up to 35 hours a week.	</p>	</dd>    <dt>First Job Phase<dd>	<p>	5 years at "The Little Software House on the Prairie".	I must have been the only person to move to Urbana,	without knowing that the University of Illinois was there.	</p>	</dd>    <dt>Masters - Illinois<dd>	<p>	I was about to leave Urbana, when Wen-Mei Hwu arrived.	Since we shared an interest in out-of-order processor design,	I stayed.	<p>	I worked through the $%^$$!! classes part-time at the rate 	of one a semester	- not too heavy a load, but enough to distract me from work	at Gould/Motorola.	<p>	After years of this, my then-wife finally finished her undergraduate	degree. Deal was, she could apply for jobs anyplace in the U.S. she wanted,	so long as there was a good computer architecture school for me to do 	my Ph.D. at. 	But she didn't really look very hard, and she chose - <em> Chicago</em>.	At that time did not have any good schools for Computer Architecture.	<p>	So she moved to Chicago, and I went part-time at work.	I TA'ed a VLSI design class at Bell Labs' Indian Hill facility,	so that UIUC would pay for my trips to Chicago to visit her.	<p>	But this didn't work, so I eventually chose the quickest topic	I could to finish my MS and get out - "Snoopy Cache Test-and-test-and-set without	Excessive Bus Contention" - proving that I have some interest and knowledge	of MP and memory ordering.	<p>	Even though all of my actual research - 4 or 5 notebooks - had been in out-of-order	CPU design, specifically, HaRRM, a Hardware Register Renaming Mechanism,	which addressed what Wen-Mei told me was the worst timing path of HPSm	by removing the associative logic from the register file and having an explicit	renaming stage. Overall, the machine I was working on looked a lot like P6.	At the end, I was looking at mechanisms for "Convergent Code" - which Gus Uht reminds me	is "Minimal Control Dependencies" - mechanism to avoid throwing away work after a	branch misprediction that turns out to be independent of the path through the branch.	I was hoping to combine this with extremely long trace cache lines to get high ILP.	(Although I've always had this feeling that ILP is overrated, and that out-of-order	by itself, in a non-superscalar situation, has advantages.)	<p>	(There was a hidden aspect to my research that I didn't understand when I was there,	but which was later explained to me. I chose to work with Wen-Mei because of HPSm,	but it became apparent, after a while, that Wen-Mei was moving towards compilers	and VLIW, and away from dynamic scheduling. Wen-mei at first 	had a number of hardware students, who dropped off after a while - I was his last	hardware student,	because I was part time and paying my own way.)	</p>		</dd>	    <dt>Intel - P6<dd>	<p>	Finally, I got my M.S. with my quick-and-dirty MP cache coherence topic,	and came to Intel.	<p>	P6 basically consumed a whole slew of ideas that I had stored up.	But, Intel does not give you space to think of new things.	New ideas were not wanted throughout most of the P6 project.	(My biggest memory of my job interview with Gould was Steve Bunch  saying	"You're a smart guy. We don't need any more smart people 	with lots of new ideas on NP1 [the sexy new CPU].")	<p>	The best new idea I came up with since joining Intel was the use of redundant	arithmetic to index the cache. I had been thinking about using redundant	arithmetic in a general purpose CPU for my last two years at Illinois - ever since 	I took classes with Robertson - but I had not solved the memory indexing problem	until I read the paper "A+B=K without Carry Propagation" by two Spaniards,	who didn't realize what they had done,	shortly after I came to Intel.	So I consider that my first wholly Intel idea.	</p>	</dd>    <dt> <p>Why have I talked so much about university?  Basically, because I never got to	be a full time student - through my own fault and obstreperousness - except	for the last spurt of research in my MS.  I never got to enjoy being at university.	But I really enjoyed that first taste of academic research, 	and, frankly, I feel a bit cheated. </p>	<dd>	    <p>	    I've also got a great big chip on my shoulder,	    because I never did well in my undergrad,	    never published the stuff I worked so hard on in my MS,	    have never published in a classy setting,	    and have no real proof that I'm a smart guy.	    <p>	    Sure, I worked on P6, and I think I contributed a lot	    - I probably think I contributed more than I did,	    there were a hell of a lot of good people working on P6 - 	    but again, there's no proof. What did Bob, Glenn, Dave,	    Mike, Andy do? Why am I the only P6 architect not to ever	    get called to the front of the room to receive a plaque?	    	    <p>	    Yes, I'm paranoid and insecure.	    But at least I'm honest about it.	    (Maybe if I felt secure I would become complacent	    and never come up with an original idea again?)	    <p>	    So I guess that a Ph.D. is a way of showing that I am smart	    - that there are some ideas that are incontrovertibly <b><em>mine</em></b>	    </p>	    <dd>.    <dt>Ph.D. - T.B.D.</dt>	    ...    </dl></ul><H2>Fear</H2>I am, however, quite a bit scared of returning to school.I don't like doing multiple classes at once.I like doing one thing, intensely - crash courses.My time management skills need work.<p> I am not not very good at deferring my interest in a topic.If I am seized by an idea, I will exhaust myself working on itfor upto 60 hours straight- and often not be able to make it to class the next morning.Even when there is a test.<em>Hell: I often don't even know what day it is!</em><p> I can hope that my ten years of experience in industry has tamedthis tendency a bit, but, y'know, I don't know that I want it to betamed.<p> So, I'm actually quite afraid that I will get into trouble withthe class requirements to do my Ph.D.<p> Doing world-class research I'm not afraid of - although I amsomewhat afraid that I may have told the various members of mycommittee what I think of their work.<H2>Hell</H2><p>Bottom line: I'm a glutton for punishment.(Moreover, I'm about to punish you with the image below.Does anyone have recommendations for a good image editor?Why doesn't HTML support vector, non-pixel, art?)<p><!WA0><IMG SRC="http://www.cs.wisc.edu/~glew/gradhell.gif"     ALT="Matt Groening's `Life in Hell' Lesson 19: Grad School - some people never learn"     UNITS=en     WIDTH=660     Height=660><h2>Advice</H2>Do as I say, not as I did:<ol><li>Finish undergrad as quickly as possible.Yes, I know it's frustrating to go to boring classes when you so much want to investigate new ideas. Tough it out if at all possible.Get to the interesting stuff as quickly as possible.<p><li>But if you are like me, a square peg in a round hole,persevere!  Choose your goals, and work towards them. Don't let failing your classes, losing your scholarship,or having your ideas laughed at get you down!<p>99% of all smart people just drift around according to what interesting ideas their betters and mentors give them.Which makes life easy because then you are working on what yourbetters and mentors think you should be.<p>But we rebels choose our own topics.If we are lucky, we choose the right ones.</p><li>I'd like to hire some smart non-conformists like me.But it's hard to find them in a mass of resumes.All you can sort on is grade point averages, classes, degrees.<p>If you don't fit in, make yourself stand out! Flaunt your successes!Maybe even your failures (I got my first good job with a resume that containeda graph pointing out the bimodal distribution of my grades).</p><li>Create a portfolio of the work that you have done outsideformal settings.<p><li>Engage in shameless self promotion, even if it isn't something you feelcomfortable with.  The drones do it anyway, with their old biy networksand references and honour societies.Beat them at their own game, and then beat them where it counts - by producingsuccessful products.<p><li>Use the Web! The Internet is a wonderfully democratic place. The drones may saythat "There isn't anything good in comp.arch" (or your favorite newsgroup).But there just may be  someone like me reading, making note of the people who make sense- irrespective of whether they come from a fancy gradschool or a podunk chop shop.<p><li>Shop around!<p>When I had my problems with undergrad, I thought that all universities were the same.All taught in the same way.<p>But it's not necessarily true.<p>There are different styles of learning: Some people like multipleclasses. Some, like me, like crash courses. Some people like homework.Others like exams.<p>American schools emphasize classes.British schools emphasize lecture series and examinations.<p>Some law schools use the case method. Others use conventional classes.<p>Shop around for the style of education that you do best with.<p>One of the tragedies of modern life is the number of people who havebeen turned off by the first style of education they encountered,but who would have done well with another.<p> There is no single  right style.But then again, it is useless to cry about the mismatch.Solve it!</p><li>I shouldn't say "drones"  above. It's a pejorative.I mean by "drones" the people for whom everything seems easy,who do well in everything they do,but who do not <em>choose</em> what they do, but rather drift according to whatever is placed in front of them.<p>As opposed to troublemakers like me.<p>Can't you tell that I resent them?<p>But beware: the "drones" are not necessarily any less smart or less creative than the "rebels".In fact, they may well be smarter: maybe they aren't drifting, but reallyhave intellectual goals they want to acheive, but go about things the easy way.Certainly, they are luckier than us rebels.<p> But they aren't any less creative, and they aren't any less smart.And they contribute as much, or more, to society as we rebels do.<p> It's easy for a rebel to fall into the trap  of thinking thatrebellion in itself is valuable. It isn't - at least not necessarily.<p>All I want to say is that, if you're rebellious by nature, it doesn't have to beall or nothing. Work with your nature, take advantage of it.Beat the drones at their own game,and when you run the show, change the rules.<p>And never let the bastards get you down.</p></ol> <hr>$Header: /u/g/l/glew/public/html/RCS/higher-education.html,v 1.2 1996/09/13 15:34:46 glew Exp $