Slashdot Mirror

Try this: http://www.cs.wlu.edu/~levy/software/pecon/ I wrote it and have run it on Linux and Mac OS X. It's a free, bare-bones alternative to something that Mathworks sells for a lot more money: http://www.mathworks.com/products/distriben/ Of course, if you want *real* high-performance, you won't run Matlab, but lots of people have tons of Matlab code and many idle processors, like the original poster.

Cluster them and use the MATLAB Distributed Computing Toolkit. It parallelizes many of the underlying linear algebra algorithms used in matlab (using SCALAPACK) to take advantage of as many cores as you let it. Its not perfect by any means (and if you try and use it in a multi-user environment like a computational grid it is an absolute pain in the ass) but it does significantly speed up many types of matlab computation. Check it out here.

The latest release, MATLAB R2007a, is supported on Vista. The previous release, MATLAB R2006b, was released before Vista was released.

Our guiding principle is "Do the Right Thing." This means doing what is best for our staff members, customers, business partners, and communities for the long term, and believing that "right" answers exist. It also means measuring our success, not merely in financial terms, but by how consistently we act according to this principle.

There is a beta OS X version of Matlab available. I got a copy from them for free here:
http://www.mathworks.com/support/faq/macintel.html

A bit buggy, but matlab code executes fine on my MBP. I sent in a bug report and corresponded directly with a developer about the problem (print preview has problems), which was an interesting experience for me as a student.

For a different, somewhat more technical, but more succint discussion, Cleve Moler [of Matlab fame] wrote another view of this topic, about 5 years ago.

The math is the same, of course, but two points of view may provide a greater sense of perspective. So to speak. And Cleve is always worth listening to.

A calculator with a comparable set of capabilities to MATLAB, etc., even with the same monochrome screen, probably wouldn't last too long on a set of AAAs. Even back when I was using my TI-83 for long problem sets in college, I could still get months of daily use, probably a few hundred hours, out of a single set of batteries. No portable device of such power and flexibility can come close.

I'll concede the point about the price-point, though. For $100 you can get an entry-level PDA with color screen. It won't be nearly as rock-solid reliable, though, nor have as much of a user base and support.

I agree that they don't seem to have changed much over the last decade. But, I would contend that a a TI graphing calculator can do an aweful lot. I'm not talking about graphing a Calabi-Yau manifold, or something handled by one of Matlab's extensive toolboxes, but the actual manipulation and display of numbers and algebra. What do you wish they could do that they don't do now? If you'll look at user-pages for TI and other graphing calculators, you'll see that people have been able to program them to do amazingly complex things.

I don't use my graphing calculator for much these days, but that's mostly because it would take me less time to use MATLAB, mainly due to having a full-sized screen, keyboard and mouse.

This is interesting but the sample size is too small to let us know how accurate this technique really is.
http://www.mathworks.com/company/user_stories/user story10433.html?by=company

Let's start with facts, shall we?

Here's the price sheet: https://tagteamdbserver.mathworks.com/ttserverroot /Download/33872_91012v25_NA_INDV.pdf

RIGHT NOW, the single-copy United States price for Matlab for commercial use is $1900. The various add-on toolboxes cost anywhere from a few hundred dollars to several thousand dollars.

Those are ONE-TIME purchase prices, not annual license fees. Annual maintenance contracts, which get you upgrades as they become available, are typically around 1/5 of the purchase price.

As for US programmers costing $15/hour, GET REAL. Last time I looked, minimum rate for an entry-level new college grad was $50K, or $25/hr. Furthermore, you have to add overhead to that hourly rate: the guy typically will cost the company his raw salary AGAIN, in benefits, physical plant, support resources (desk, power, light, PC, network, copy paper...).

Figure a quarter of a million dollars a year per programmer or engineer, total burdened cost, and you aren't that far off.

There is an easy cross-check for this. Look at a healthy technology company, one whose principal product is software development, so that cost of good sold (raw materials, subcontract, etc., is negligible. Divide total sales by number of employees, and that gives you an upper bound on what the employees are costing the company. Look at an UNHEALTHY company, do the same calculation, observe that the unhealthy company is NOT succeeding in making enough money to cover the cost of the employees, and you have a lower bound. When I did this exercise in Dallas a few years ago, I found that the magic number was somewhere between $200,000/yr and $300,000/yr. That's $100/hr-$150/hr. At that rate, $1900 for a copy of Matlab, amortized over 2000 hours, is $1/hr: less than 1% of the total.

And, if you want to get picky, figuring maintenance at 20% of purchase price per year, you're looking at $4000/5 years, or $800/year, or about forty cents an hour, or about three dollars a day. Bathroom breaks cost more.

If you want to do cost arguments, always start with real numbers, not made-up ones.

If someone is developing something in Matlab I assume that what they are doing is non-trivial, like processing a signal for instance. They probably start by using Matlab's tool kits and block sets. Their problems may go a bit beyond transcription. For instance, can they expect the same floating point results from Matlab and C++. That one could provide some strange and annoying results. Finding a library with the functions available in Matlab could be a challenge. People are implementing some amazing stuff in Matlab and from what I've seen, transcribing it to C++ is a little more difficult than mere transcription.

Here's a good one; a very faithful rendition of 802.11 done in Matlab. http://www.mathworks.com/matlabcentral/fileexchang e/loadFile.do?objectId=3540&objectType=file

You can call matlab libraries from C++ code, which would seem to be the best of both worlds. Then you wouldn't have to port anything.

Unless you want to distribute your application to people who don't have MATLAB. Or is the MATLAB runtime engine free to distribute?

While my complex data visualization stuff is a long way off from being done in Java, the sort of simple data visualization stuff that I was doing in 1997 under Windows works quite well under Java, and it works equally well under Linux.

(a) lend themselves extremely well to optimisation

(b) lend themselves extremely well to incorporation in subroutine libraries

(c) tend to isolate the most compute-intensive low-level operations used in scientific computation

SGEMM

If you read the article, you will find (among others) a reference to a operation called "SGEMM". This stands for Single precision General Matrix Multiplication. This is the sort of routines that make up the BLAS library (Basic Linear Algebra Subprograms) (see e.g. http://www.netlib.org/blas/). High performance computation typically starts with creating optimised implementation of the BLAS routines (if necessary handcoded at assembler level), sparse-matrix equivalents of them, Fast Fourier routines, and the LAPACK library.

ATLAS

There is a general movement away from optimised assembly language coding for the BLAS, as embodied in the ATLAS software package (Automatically Tuned Linear Algebra Software; see e.g. http://math-atlas.sourceforge.net/). The ATLAS package provides the BLAS routines but produces fairly optimal code on any machine using nothing but ordinary compilers. How? If you run a makefile for the ATLAS package, it may take about 12 hours (depending on your computer of course; this is a typical number for a PC) or so to compile. In this time the makefile will simply run through multiple switches and for the BLAS routines and run testsuites for all its routines for varying problem sizes. And then it picks the best possible combination of switches for each routine and each problem size for the machine architecture on which it's being run. In particular it takes account of the size of caches. That's why it produces much faster subroutine libraries than those produced by simply compiling e.g. the BLAS routines with an -O3 optimisation switch thrown in.

Specially tuned versus automatic?: MATLAB

The question is of course: who wins? Specially tuned code or automatic optimisation? This can be illustrated with the example of the well-known MATLAB package. Perhaps you have used MATLAB on PC's, and wondered why its matrix and vector operations are so fast? That's because for Intel and AMD processors it uses a specially (vendor-optimised) subroutine library (see http://www.mathworks.com/access/helpdesk/help/tech doc/rn/r14sp1_v7_0_1_math.html) For SUN machines, it uses SUN's optimised subroutine library. For other processors (for which there are no optimised libraries) Matlab uses the ATLAS routines. Despite the great progress and portability that the ATLAS library provides, carefully optimised libraries can still beat it (see the Intel Math Kernel Library at http://www.intel.com/cd/software/products/asmo-na/ eng/266858.htm)

Summary

In summary:

-large tracts of Scientific computation depend on optimised subroutine libraries

-hand-crafted assembly-language optimisation can still outperform machine-optimised code.

Therefore the objections that the hand-crafted routines described in the article distort the comparison or are not representative of real-world performance are invalid.

However ... it's so expensive and difficult that you only ever want to do it if you absolutely must. For scientific computation this typically means that you only consider handcrafting "inner loop primitives" such as the BLAS routines, FFT's, SPARSEPACK routines etc. for this treatment, and that you just don't attempt to do that yourself.

It is easy to do analysis and plotting of scientific data from databases. All you have to do is actually use the right tool for the job. Do you have gobs of numerical data that you want to crunch, analyse and plot? Then you probably want Matlab or R. Is that data in a database? Then you'll probably be using Matlab's database toolbox or R's database connectivity to import and export data (based on whatever database query you care to use - you can import whole tables easily if you like). Once the data is imported (which is trivial) you have no shortage of tools to process, munge, analyse, and plot to your heart's content.

Jedidiah.

"MATLAB is a high-level language and interactive environment that enables you to perform computationally intensive tasks faster than with traditional programming languages such as C, C++, and Fortran."

It depends on how extensively this application makes use of the JIT Accelerator that was introduced in MATLAB 6.5:

i think you're working under the (very) false assumption that the kind of "data analysis" you care about is the only kind there is. i've done lots of data analysis using simple spreadsheets (seldom Excel specifically, generally open source alternatives). it's a great tool for that job. but the kind of data analysis i care about rarely exceeds one data point a day, averages lower than that, and we're rarely looking at more than a few years. for doing data analysis against that, it's great.

besides, Matlab prices itself right out of the water for any data analysis project i've ever had to work on.

none of which is intended to dispute the fact that Excel is frequently misused. i've seen whole businesses run by chains of inter-linked excel workbooks, any one of which would make a grown man cry.

Any unclassified code written with taxpayer money should be Open Source and available for the cost of duplication. Of course, all the ITAR nonsense will get in your way in some cases, but this can be overcome by following procedures. Free and open source supports the long term protection of data integrity and the inability of a proprietary vendor to hold the government hostage to upgrades and maintenance.

As for COTS, well, it is desirable that it be FOSS. However, some extremely useful COTS stuff just doesn't have an adequate FOSS counterpart - e.g. Simulink.

Here are some more contests, accessible to everybody (not only students):

http://www.mathworks.com/contest/
  which is a collaborative(!) competition

http://codewalkers.com/php-contest.php
  holds a nice PHP contest, but seems to be inactive

http://dinsights.com/POTM/
  holds a nice permanent contest, but there is nothing to win

http://www.recmath.org/contest/index.php
  holds a tough contest, which just finished.
  With IFCP, it's the hardest contest I'm aware of.

http://www.mathworks.com/products/symbolic/

I had the good fortune of meeting the gentleman when I interviewed with Mathworks a couple of years ago. I was taken aback by his humility, and the poor guy was embarrassed when I requested his autograph :) He has a former license plate in his office that reads "YACCMAN".

You want non-linear contrast stretching, so that the values of greatest interest are differentiated the best.

On that note, I'd like to mention that Matlab's adapthiseq (contrast-limited adaptive histogram equalization) totally rocks. It's rather nice for pre-processing before applying more sophisticated techniques.

death to MATLAB

Exactly. MATLAB runs on Linux. Linux must die, therefore MATLAB must die. You are either with us or you are with the communists.

OUR PLAN IS WORKING!

The basic problem is that a Harrier has more major flight controls than the pilot has hands. There's a nozzle angle control and a throttle control, along with the usual stick and rudder pedals. VTOL operation requires coordinated operation of the nozzle and throttle controls. Both have significant lag. That's a tough control problem, worse than a helicopter.

Everything has been tried. Better pilot training. New flying approaches. Simulator training. A redesign (the Harrier II). Stability augmentation systems. Avoiding VTOL whenever possible. Harriers still crash a lot. (The Harrier has a good ejection system, so the pilots usually survive.)

One of the stability augmentation systems was the VAAC Harrier Study. This was an experimental effort to use computer control to get the three inputs that affect longitudinal stability (stick, throttle, and nozzle angle) down to two. This was supposedly successful but was not deployed.

This new thing seems to be a further step in that direction.

Let's not forget the FDIV bug. Here's another article that discusses the details of that problem.

Nope, they are working for various ISVs. I'd actually imagine that it would be harder for H1-Bs to get jobs in the public sector, given all of the civil service requirements etc, but what do I know...

But, of course, my experience doesn't match up with your preferred script, so naturally you blow it off.

Funny, I was thinking the exact same thing about your dismissive reactions to my experiences.

That's ok, it's to be expected from a soulless shill like you.

Ooooohhh!!! Namecalling!!! With an attitude like that, I can now see why you are worried about keeping your precious job.

One day, when this program is abolished, I'll hoist a drink and think of you.

I'm actually fine with abolishing the H1-V program, provided that it is replaced with something as equally capable of cherrypicking the world's talent as the existing program.

Gnuplot does work, but its not the most user friendly piece of software, and while I've used it for rough-and-ready visualisation, I've never been able to get attractive (i.e. publishable quality) results out of it.

Since you don't specify free software, can I recommend the graphical capabilities of Matlab? As well as being endlessly scriptable and versatile, there's a GUI so you can place text and symbols anywhere you like. There's a native Linux version, and as you're a student you should be able to buy it at a more reasonable price. It's also quite possible that your university/department has a site licence of some description.

Gnuplot does work, but its not the most user friendly piece of software, and while I've used it for rough-and-ready visualisation, I've never been able to get attractive (i.e. publishable quality) results out of it.

Since you don't specify free software, can I recommend the graphical capabilities of Matlab? As well as being endlessly scriptable and versatile, there's a GUI so you can place text and symbols anywhere you like. There's a native Linux version, and as you're a student you should be able to buy it at a more reasonable price. It's also quite possible that your university/department has a site licence of some description.

MATLAB started as a Fortran library it seems. As handy as it is, MATLAB has some dire limitations: its performance and syntax. While certain vectorized operations can be speedy, many folks I know end up recoding in C because MATLAB just crawls for everything else. On top of this the language itself is just plain ugly. It's reminicent of BASIC with random bits of bash scripting and other oddities thrown in to make a patois that is decidedly disgusting.

I myself have switched over to using R for statistical computing mainly because it's nicer to look at than matlab and has some really great statistical facilities built right in. For more analytical stuff, Mathematica or its open source cousin Maxima are definitely a better choice.

What excites me about Fortress is that it's both cleaner looking than MATLAB and has some neat features like traits. I'll be curious to see how it pans out.

You are wrong about Matlab's support for OSX (note the order there), Checkit In fact, it even runs under Linux (after the most difficult install procedure I have ever had).

I ALMOST agree with you. As far as "standard" languages go, FORTRAN does do this best. The problem is, there's a much better solution now.

1) Array, vector, and matrix processing can not only be done better, but with FAR less work in MATLAB, and using their converter, will nicely produce C (C++?) on the other end.
2) Parallelization of vector processing can also be done with >version R11 of Matlab, again with many fewer lines of codes, using a toolkit. (see here)

A couple years ago I was working on some "Cutting Stock" optimization problems. The current system was written in Fortran, but Matlab was the language of choice for the new one. Its probably more expensive, but talk about saving you some trouble in the long run...

It would be silly to write a memory manager in UML.

It would be silly to write a memory manager in UML.

The Discrete Wavelet Transform involves dividing and filtering the time domain signal which will take as long as any FFT based algorithm. Do you have a "linear computational complexity" reference ?

Funnily enough the recent Matlab comp was along these lines: Main Page. Given a number of unknown grids of boxes of different weights a program was needed to solve the rearrangement problem as efficiently as possible.

Where's my compilable flowchart?
I believe this is what you're looking for:
Stateflow Coder

what they do is say that you can't distribute a product that contains certain bits of Matlab or provides access to the Matlab command line unless you're distributing it to someone who already has a license agreement with Mathworks.

You might want to try Jython and the Numerical Python for Jython.
I have not used either for a long time, but use plain Python and Numerical Python a lot; sure beats Matlab and Mathematica for most things. Right now for solving optimization problems with 10k+ s.t. constraints.

You do realize that MatLab runs in Linux if you're willing to licence it, which it seems you are under windows...

Anyway, a quick freshmeat search showed me that Nulab, Yorick, Scilab, FrAid and Lush are all possible replacements, depending on the application. Moreover, many of those refer to Octave which might be suitable, depending on your needs.

Likewise National Instruments makes LabVIEW for Linux, and freshmeat says to look at Flow Designer and TACO as potential free replacements.

If the two are used for related purposes, then consider RobotFlow which came as a result under both searches...

Just in case you decide to retry the system at a later date...

RTFA'd, saw the movies. Where does it say Linux?

Nowhere, that's how I got to the conclusion that this story must have been submitted by one of my colleagues (am associated to that department, myself)... :)

As far as I remember, the computer controlling the electro-servo hydraulics actually *is* powered by Linux. I suppose it was RTLT, because the students and several of my colleagues did much modelling in Simulink.

Well, there is some more information available, but this year's students did not as equally good a job of documenting their project as the 2003 students (Swedish only, though)...

m2html -- a GPL HTML documentation system for Matlab can generate TODO lists out of keywords found in source files for years as well. It can also run in "real time" - just put it to cron. Can Microsoft patent something which is out there for years?

1. Excel is an extremely poor tool for doing anything other than basic graphs and calculations. For engineering purposes, it's near useless.

I also miss having an equivelant to the Excel solver utility, which can optimize hundreds of variables at once to minimize/maximize a result.

Whoa! You can't be serious. If you have to optimize something with hundreds of variables, you should look into real programs to do the task. To be frank, results from excel solver are shit. Optimization is a large field of applied mathematics and can't be reduced to MS Excel click-through feature. See for example this and this.

--
Jari

Not true. I purchase quite a lot of closed source software for Linux. Some of it I run under Wine, but where possible, I buy Linux versions.

When I was working on my CS degree I purchased copies of Matlab and Mathematica, I also would have purchased Maya (went to the lab instead) if it had been less than $400 for the student version. All my windows friends found them on irc/p2p.

I used to purchase my copies of Redhat, I have the boxed sets (other than those I gave away) for every release between 5 and 7.3. However, I quit using Redhat since I received the exact same benefits for buying it as someone that downloaded it for free. I would have stuck with them if they had allowed me a year of priority access to up2date with the box set purchase. They didn't. I left. I now use Debian and am much happier with it. It isn't because I want something for free, but because I when I pay I want something more than is given away for free. If I get something better for a few dollars than I could get for nothing then I'll pay a few dollars. That is why my display is running Accellerated X. The display drivers in Xfree86 could not handle my laptops screen/videocard (1400x1050 lcd with an intel i830M graphics card) and left a nasty black border around the screen. I tried the Accellerated X demo and it worked perfect.

There IS software available to purchase for Linux. Much of it is better than the free stuff, and lots of us use it. Many people have no idea that some of it even exists. I think that many of the companies that sell Linux software and have superior products just need to spend a little more on marketing so they get some name recognition (this means you XIG). The companies also tend to get "stuck" on a distro. Many of them only release RPMs and refuse to provide instructions for other non-RPM distros, even when it works perfectly on the distro (this means you again XIG).

I suppose Fortran has a lot of momentum, (I think my alma matter just dropped it from the ME requirements in the last five years or so) and thus many non-CS persons who use it out of habbit. However, it seems that once someone learns how to program in Matlab, they never go back. It's just so damn quick to prototype ideas with.

... at the Lenslet page, the unit actually has several components. The VMM (vector matrix multiplier) does 8000 MAC (matrix array calculations) but there is a VPU (vector processing unit) that comes in at 128 Giga-ops and which would be the bottleneck in the whole setup. No question this is a huge improvement BUT to put it in perspective, it is a DSP only, not a computer system (although some neural network weenies might see a way of turning this into something more than just a DSP). In any case, the bottlenecks will come from the equipment it has to operate with both onboard and off.

Still, note that it's developed with Matlab. Now surely that is the Holy Grail of research, a bitchin' language with an awesome tailored processor. Imagine the logo Matlab [Lenslet Inside].

Actually, it DOES work. $100 for the full suit of Matlab applications, for example, is MUCH less expensive than the $5000 or so you would have to pay to buy a commercial licence. Also, schools require students to purchase software like this. In order to be in Engineering at Virginia Tech, for example, students must have Inventor, Matlab, Windows XP, Acrobat, CAD and a whole host of other software. The package costs about $900, but commercial licenses for similar software would be well over $100,000.

---I don't understand. They're comparing a bunch of X Servers versus a bunch of Dell PC's?

That's what it seems. Yeah, I know. The compairison sucks.

---What about the guy who's playing MP3's at his desk?

What? Lopster and XMMS arent good enough for him?

---What about the guy who wants to sync to his Palm Pilot?

There's already good sync software for Linux. Just un-endorsed. Hell, They might actually make a "legit" tool if stuff like this happens.

---What about the guy who's using Messenger?

There's buttloads of tools for IM on Linux.

---What about the guy who *NEEDS* a specific piece of software to communicate with his peers?

In limited cases, WIndows is the only answer for now. But as sysad, you could put heavy pressure on a company who does such.

---What about the guy who's burning DVD's of classroom presentations?

Get him a Mac. Most unix dudes could get one working.

---What about the guy who wants to run mid-priced shrink wrapped applications like Mathematica or MATLAB or IDL (all probably less than $10,000 for a single user license, but could get expensive for a big machine).

OK... Your point ?

---What about the guy who runs small simulations -- the kind of thing a reasonable desktop could do in an evening or a weekend? People who run computer centers often complain about 40 hours of computer time on the big boxes.

Help his department build a small cluster for job crunching. COuld even be a "beowulf" cluster if his apps support it. Then he could 'job' out time to other departments. That'll avoid cpu munchers on the main system.

---In short, what about all the flexibility that the Personal Computer gives the user? Why ins't that included in their "TCO" at all?

How about the flexibility of "use the tool that works for the job"? Trust me, you really dont NEED windows anymore.Look at all 3 links at your Math program question.

It's not necessarily true that for loops are always bad. MathWorks (the company that makes MATLAB) have added an accelerator to MATLAB 6.5 that can make for loops almost as fast as C (which isn't bad for an interpreted language.) Look here for more info.

This is not such an issue any more as The Mathworks have introduced new technology into the interpreter. The interpreter now contains a "Just In Time" JIT compiler that phenomenly increases the speed of looping code.

"The JIT-Accelerator gives you the flexibility to run your code faster without having to perform vectorization.....Today with the JIT-Accelerator, you no longer need to vectorize code to speed up the execution of many applications."