OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.


Help: OASIS Mailing Lists Help | MarkMail Help



   RE: [xml-dev] Data streams

[ Lists Home | Date Index | Thread Index ]

I tried Steven's experiment from a different angle. I filled an Excel XP
spreadsheet with a single-digit number, saved it in both XML and in a
comma-delimited text file (CSV). I then compressed both with WinZip and then
opened both with Excel. Here's what I found:

The XML file was 840MB, the CSV 34MB -- a 2,500% difference
Compressed, the XML file was 2.5MB, the CSV 0.00015MB (150KB) -- a 1,670%

Equally dramatic is the time it took to uncompress and render the files as
an Excel spreadsheet: It took about 20 minutes with the XML file; the CSV
took 1 minute -- a 2,000% difference.

My conclusion is that delimited text files handle large arrays of data more
efficiently. This stems, in part, from the fact that a comma delimiter (or
some other single character) carries much less overhead than tags; CSV
requires only a comma, while XML requires a minimum of 5 characters (<></>)
-- that's makes CSV a minimum of 500% more efficient ... and when you add
the semantic labels and attributes to the tags, and the size of XML
increases dramatically.

Note, however, that when dealing with large blocks of text instead of
numbers (or small text strings), the difference between XML and delimited
text files is considerably less.

Of course, XML offers benefits that a plain data array in a CSV file does
not, such as attribute definitions and hierarchical associations between the
data (if that's necessary) ... even though there are ways comma-delimited
data can be used to perform the same functions of XML when rendering
serialized data arrays as charts.


-----Original Message-----
From: Rick Marshall [mailto:rjm@zenucom.com] 
Sent: Saturday, December 04, 2004 4:50 PM
To: Steven J. DeRose
Cc: xml-dev@lists.xml.org
Subject: Re: [xml-dev] Data streams

thank you steven. that was the experiment i proposed a month or so ago. 
and you have just shown very neatly that the entropy of the message 
hasn't changed with the representation.

the only piece missing is to put out a file of a million 32 bit integers 
(4MB by definition) and see how much it compresses - ie more than 50%? 
then we really do have a lower bound on the entropy. i'm choosing to 
ignore the compact formula/algorithmic representation at this stage 
because that's not a general solution.



Steven J. DeRose wrote:

> At 16:31 +0000 2004-12-04, Michael Kay wrote:
> At 10:42 -0500 2004-12-04, tedd wrote:
>>>  > In everything I have read, it appears that every chunk of content
>>>>  must be encapsulated by tags, such as:
>>>  > <data>123.456</data>
>> These are both legitimate XML documents. The question is, what are you
>> trying to achieve? If you use XML markup around the data, then an XML 
>> parser
>> (and tools such as XSLT) will understand it. If you use commas, then you
>> have to parse it yourself. If you want to parse the data by hand, 
>> then why
>> use XML in the first place?
> Michael makes a good point -- how to do it depends on your goals.
> But to be extra clear, there is no real way to make XML *itself* aware 
> of fields that are delimited only by commas (or some other single 
> character delimiter). Such syntax was considered as a possibility but 
> rejected. SGML can do this via the SHORTREF feature, if you're 
> absolutely set on it.
> For cases like your example, where there is very little structure to 
> demarcate, it seems important: a million copies of "<data></data>" 
> versus "," adds up. However, consider:
> 1: if your data is "text files that are literally tens of thousands of 
> characters in length", that is small enough that the overhead won't 
> disturb most software running even on a cell phone. If we were talking 
> many millions or billions of *records*, then this would be more of an 
> issue (as it is for some users).
> 2: If you want the data formatted by CSS or XSL-FO, or transformed by 
> XSLT, or whatever, having all the data in one syntax that the 
> applications *already* know about is much easier than rewriting the 
> applications or working around them to add some syntax (like commas) 
> that they *don't* know about. You'll never have to debug the XML 
> parser you use to parse all those "<data>" tags, but you will spend a 
> lot of time if you try to introduce a new syntax in your process.
> 3: Any text file that contains zillions of instances of a certain 
> string, is necessarily very compressible. The first thing a 
> compression program will do is discover that "<data>" is real common, 
> and assign it a really short code. A comma-delimited file is 
> inherently less compressible.
> Here are some empirical results:
> I created a file with the numbers from one to a million, delimited in 
> different ways. zero.dat has just a linefeed between numbers; 
> comma.dat just has a comma and a linefeed; tag01 has a start and 
> end-tag with the one-character element type "d" (and the linefeed); 
> tag02 has element type "da", on up to tag20 which has a 
> 20-character-long element type. 5-line Ruby program available on request.
> Here are the original sizes:
>  6888888  4 Dec 13:35 zero.dat
>  7888887  4 Dec 12:50 comma.dat
> 13888881  4 Dec 12:51 tag01.dat
> 15888879  4 Dec 12:52 tag02.dat
> 17888877  4 Dec 12:53 tag03.dat
> 19888875  4 Dec 12:56 tag04.dat
> 21888873  4 Dec 13:06 tag05.dat
> 31888863  4 Dec 13:07 tag10.dat
> 51888843  4 Dec 13:08 tag20.dat
> Here are the sizes after gzipping:
>  2129148  4 Dec 13:35 zero.dat.gz
>  2130082  4 Dec 12:50 comma.dat.gz
>  2377733  4 Dec 12:51 tag01.dat.gz
>  2376912  4 Dec 12:52 tag02.dat.gz
>  2518197  4 Dec 12:53 tag03.dat.gz
>  2638489  4 Dec 12:56 tag04.dat.gz
>  2631120  4 Dec 13:06 tag05.dat.gz
>  2661673  4 Dec 13:07 tag10.dat.gz
>  2596261  4 Dec 13:08 tag20.dat.gz
> You can see that:
> the linefeed-only file reduces to 2130082 / 6888888 = 31% of its 
> original size
> the 20-char tagged file reduces to 2596261 / 51888843  = 5% of its 
> original size
> And even though the 20-char tagged file was over 7.5 times bigger than 
> the linefeed-only file when uncompressed, once they're compressed it 
> is only about 1.2 times bigger -- a mere 22% increase despite every 
> number having 2 tags with 20-character tag names, instead of nothing 
> but a line break.
> I wouldn't worry about the extra bytes much. If you've got enough data 
> for it to matter, buy a disk-compression utility and you can forget 
> the issue.
> Steve DeRose


News | XML in Industry | Calendar | XML Registry
Marketplace | Resources | MyXML.org | Sponsors | Privacy Statement

Copyright 2001 XML.org. This site is hosted by OASIS