[Date Index] [Thread Index] [Author Index]

Re: 3.0 = Rip Roaring Resource Hog :-(

• To: mathgroup at smc.vnet.net
• Subject: [mg5162] Re: 3.0 = Rip Roaring Resource Hog :-(
• From: Eric Jacopin <jacopin at thomson-lcr.fr>
• Date: Wed, 6 Nov 1996 01:34:12 -0500
• Organization: Thomson-CSF, Laboratoire Central de Recherches, Orsay, France
• Sender: owner-wri-mathgroup at wolfram.com

Mark Evans wrote:
> 
> 3.0 = Rip Roaring Resource Hog :-(
> 
> I have used Mathematica for years and found it exceptionally useful in a
> wide variety of situations.  It has succeeded where other programs and
> compilers have not only failed, but failed miserably.  It has allowed me to
> do jobs I could not dream of accomplishing without its unique capabilities.
>  It is generally a pleasant and refreshing experience to work with
> Mathematica.
> 
> Yet there are some situations when Mathematica just falls down on its face.
>  I have encountered several of these in quick succession recently.  All of
> them relate to the manipulation of real-world data as opposed to small
> examples or algebraic expressions.
> 
> It is starting to sink into my brain that Mathematica is extremely wasteful
> of system resources.  For the first time, I am thinking about moving my
> work over to MATLAB.  I need a program that knows what machine numbers are!
> 
> I've recently posted some complaints about MMA 3.0's file system
> deficiencies.  I am informed that WRI is in the process of addressing these
> for version 3.1.  Hooray for that.
> 
> But consider the following demonstration of what a resource nightmare MMA
> 3.0 can be.
> 
> I have a single Windows WAV sound file that is exactly 463,094 bytes long.
>  This figure is reported by the Windows 95 Properties dialog.  That's about
> half a megabyte.  In principle, any program that reads in this data should
> not need more than about 0.5 to 1.0 megs to store the data.
> 
> MATHEMATICA 3.0 TAKES 9 MEGABYTES TO STORE 0.5 MEGABYTES OF DATA.  This is
> an inefficiency ratio of 18:1.  I AM HAVING SERIOUS DOUBTS ABOUT THE
> ABILITY OF THIS PROGRAM TO DEAL WITH REAL LABORATORY DATA SETS.
> 
> Here are the byte counts I got from Mathematica 3.0 Final.
> 
> <<Miscellaneous`Audio`
> 
> In[17]:=
> filename = "G:\\Telluric\\Data\\Original Data 8\\b1.wav"
> 
> Out[17]=
> G:\Telluric\Data\Original Data 8\b1.wav
> 
> In[18]:=
> FileByteCount[filename]  (* size of file in bytes *)
> 
> Out[18]=
> 463094
> 
> In[19]:=
> sound = ReadSoundfile[filename];  (* read it (2 min.) *)
> 
> In[20]:=
> ByteCount[sound] (* how big is the data in bytes *)
> 
> Out[20]=
> 9261024
> 
> In[21]:=
> N[ByteCount[sound] / 1024]  (* how big in kilobytes *)
> 
> Out[21]=
> 9043.97
> 
> In[22]:=
> nsound = N[sound];  (* turn integers into machine #'s *)
> 
> In[23]:=
> N[ByteCount[nsound] / 1024]  (* new size in KB *)
> 
> Out[23]=
> 9043.97
> 
> I tried Share[] and it did not help.  So whether the array is composed of
> exact integers or machine-precision numbers, it takes 9 megabytes of
> storage in Mathematica 3.0.
> 
> Nine megabytes of storage for an uncompressed file that takes 0.5 megs on
> disk is out of this world.  This is object-orientation gone berserk.  Does
> MMA 3.0 treat every component of the array as an "object"?
> 
> I recently complained to WRI about the price increases following release of
> 3.0.  You are looking at $2000 for a copy of Mathematica plus one
> application pack.  Two thousand dollars for a setup with this kind of
> inefficiency is too much money, by far.
> 
> I want to see Mathematica succeed, but I also want to see myself succeed.
> If the tool can't do the job, it should not be advertised as an all-purpose
> slicer/dicer and priced like one.  From the outputs shown above, it seems
> that any successful use of the application packs on real data of any
> significant size will require access to several hundred megs of RAM.  Most
> of us don't have that much.
> 
> Please, WRI, please focus all your efforts for 3.1 onto these efficiency
> problems.
> 
> Mark Evans
> evans at gte.net

Well, I got to admit that I *DO* agree with the extremely bad memory
management of Mathematica. Indeed, I experienced the same ratio, and
even much worse. For instance, I need almost 50 Mb (!!!!) to compute
fractal dimension from file with 60k of reals, whereas using C (and,
even scheme), I use only 2 Mb for such a thing.

I believe Mathematica is good for prototyping and teaching concepts,
which is not bad at all, but you should really have this in mind before
developing your things any further.

-- 
Eric Jacopin
ftp://ftp.ibp.fr/ibp/softs/laforia/pweak/Pweak.html
http://www-laforia.ibp.fr/~jacopin