[Beowulf] Teaching Scientific Computation (looking for the
Robert G. Brown
rgb at phy.duke.edu
Wed Nov 21 07:36:43 PST 2007
On Wed, 21 Nov 2007, Jim Lux wrote:
> Quoting "Robert G. Brown" <rgb at phy.duke.edu>, on Wed 21 Nov 2007 05:56:51 AM
>> On Tue, 20 Nov 2007, Jim Lux wrote:
>>> Octave is nice, but.... the graphics are MUCH better in Matlab, and
>>> there's all those toolboxes full of cool stuff (signal processing,
>>> control systems, maps, etc.)
>>> And, an academic license for Matlab is only $100. That's less than the
>>> textbook likely costs. Granted Matlab isn't quite as cool as the
>> "Only $100" is IMO highway robbery, especially for students.
> For a copy that they get to have on their very own personal computer (as
> opposed to a shared class license, which is cheaper, on a per seat basis)?
> I don't know that it's highway robbery.. It's comparable to the cost of a
> textbook these days. It's less than the cost of a couple concert tickets or a
> live theater presentation like "Wicked". (No starving students buying
> tickets to see the Stones or the Eagles these days...)
Oooo, thus speaketh a man who must not be facing tuition bills as high
as $40,000 per year for several children coming of age.
And textbook prices are, frankly, highway robbery as well. I have a
couple of them online for free, and students all over the world use
them. A textbook SHOULD cost around $25, not $125. And one day
(probably one day quite soon), they will. There is a certain amount of
"open source" pressure on that market, and with POD and Lulu and
alternative ways for people to publish a text and sell it for a
reasonable cost, market competition will eventually bring this back down
> I think their pricing model is quiet different. If the department buys the
> copies and lets the students share licenses, it's less on a per seat basis.
I'm just describing what Duke had, which I'm pretty familiar with
because we faced these actual costs a couple of years ago and had to go
renegotiate our whole deal with them. It is a bit better now -- I think
Duke just dickered out an unlimited use license, period, so anybody on
campus can install it on their machine(s), but for a while there that
was true only for specific subpopulations (and I shudder to think what
the current arrangement is likely costing the University).
> The individual purchase model is more like a textbook. Your department
> doesn't supply textbooks to your students, do they?
No, but textbook companies give all instructors desk copies. And
students don't carry textbooks out to GM and say, "We need to buy 500
copies of this textbook to use in our design labs" after they graduate.
Matlab really is NOT like a textbook model, it is more like a training
model. If Duke is willing to train students in the use of piece of
software that isn't designed just for student use, that software should
IMO be GIVEN away to student and school, or sold at a true pittance
> And, granted, the academic world IS different in terms of computing cost
> models, but for most medium to large businesses, a computer sitting on
> someone's desktop costs about $300/month for equipment lease/replacement,
> maintenance and help desk support, and a basic load of software. (about half
> the cost is equipment, half is support).
Ah, here we could go the rounds for a long time. Agreeing that around
half the cost is human support -- I might have gone over half, even --
the split between hard and software is more like 50-50 on the remainder.
Microsoft office alone, per seat, can cost a significant fraction of the
cost of the hardware at corporate rates. XP Pro or Vista
Business/Ultimate are also hundreds of dollars (around $300 for the
former, I don't know what for the latter). Antivirus adds a bunch. A
MS based business tricked out for JUST PLAIN DESKTOP stuff -- a login
interface, Explorer, mail, Office Pro, on top of a pro base version of
the OS -- can easily cost $800 in software alone, and that doesn't
really include the cost of the server side software needed to make it
all run. Real "applications" outside of this come on top, and of course
are going to typically cost $200 and up.
Which is truly absurd to me, since my linux boxes do all of this at zero
cost in software including the server side. And that server side can be
managed far, far more efficiently, and the OS can be installed far more
cheaply and efficiently and is zero maintenance once installed. I run a
mixed lin-win network at home, and all my #!%#@ problems come from the
Win side, except for those that come about from trying to make the Win
side work with the Lin side.
Only since I actually BOUGHT a vmware workstation license for my laptop
has using Win been anything like tame and sane. Need it? Click a
"Power up" button. Done with it? Click "Power Down" and make it go
away without saving a snap. Every now and then I have to relent and
actually let it run an update.
But we've sung this song and danced this dance before.
> In the world of $200 Walmart computers and free-for-the-downloading Ubuntu,
> yes, $100 is significant.
The only piece of software I've encountered in ten years that is worth
$100 is VMware workstation (which not terribly coincidentally is the
only piece of software that I've bought for myself in the last ten
years) and I still think that IT is too expensive, but when I have Xen a
try it just did bad things (this was FC 6ish). Xen is reportedly
improving fast, and KVM is in the wings, so I might be able to do away
with even VMware soon enough.
>> With all that said, for engineering types (matlab use is taught and
>> required in the Engineering school) and for various research groups or
>> student uses matlab is OK. But if you don't need a toolbox, octave
>> really is just as good and almost identical in terms of its command set,
>> and free is a lot better than $100, even for students.
> I agree, and use both...
> I just get spoiled with the substantially better graphing capabilities of
> Matlab, and some of the toolbox things that one gets used to. Sure, one
> could probably find equivalents (or write them) for Octave, but it's nice
> when it's just right there.
I don't "use" either one -- I use them somewhat reluctantly to teach
certain intro courses where there may or may not be some benefit to
letting students integrate EOM and see trajectories happen. I've tried
using octave a couple of times for quickie purposes -- solving an ODE of
my own just to see what the solution looked like -- and it was OK, but
no easier than mathematica or maple (both fully site licensed by the
university) and not MUCH easier than my popping out a chunk of GSL-based
C to do the same thing 80x faster and with a lot more confidence in the
result. But then, I have a working C program template and a huge
library of personally written code, so for me C programming is all about
code reuse and cut-and-paste and (as you all know:-) FAST typing in jove
with my fingers right on those home keys at all times, die mice die.
Robert G. Brown
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Lulu Bookstore: http://stores.lulu.com/store.php?fAcctID=877977
More information about the Beowulf