RFC: Attributes and XML-RPC

[Julian Reschke]

I would expect that gzip sees the attribute as

     ="
    name
    "

while it sees elements as

     <
    name>
    </
    name>

Meaning three substrings for attributes, while 4 substrings for elements,
where only 1 of them occurs twice.

Seems that the corrected benchmark proves that my surprise was not without
reason :-) (posting attached again).
  -----Original Message-----
  From: Tyler Baker [mailto:tyler at infinet.com]
  Sent: Sunday, September 26, 1999 10:04 AM
  To: Julian Reschke
  Cc: Mark Nutter; xml-dev at ic.ac.uk
  Subject: Re: RFC: Attributes and XML-RPC


  Julian Reschke wrote:
     I find this way surprising. Yes, it's clear that opening and closing
tags compress well, but I would still have expected the attribute version to
be smaller...

  Each quote (2 of them), comma, equal, and attribute name take up 5-10
bytes for each attribute when mapped to a symbol table, while the element
style takes up 2-4 bytes for the start tag and the end tag. The exact number
of bytes used for each attribute or element is dependent on the compression
algorithm and the number of different attribute or element names used. So it
is no surprise to me that using GZIP produces a compressed attribute file
that is larger (-:

  Tyler



  --

  > -----Original Message-----

  > From: Mark Nutter [SMTP:mnutter at fore.com]

  > Sent: Thursday, September 23, 1999 6:34 AM

  > To: Blair Murri; xml-dev at ic.ac.uk

  > Subject: RE: RFC: Attributes and XML-RPC

  >

  > Doh! "Code in haste, debug in leisure." Ok, here are the corrected

  > results:

  >

  > $ ./make.pl

  > $ ls -l

  > total 14206

  > -rw-rw-r-- 1 mnutter mnutter 5811846 Sep 23 08:28 attrib.xml

  > -rw-rw-r-- 1 mnutter mnutter 8672170 Sep 23 08:28 child.xml

  > -rwxr-xr-x 1 mnutter mnutter 999 Sep 23 08:26 make.pl

  > -rwxr-xr-x 1 mnutter mnutter 976 Sep 22 13:16 make.pl~

  > $ gzip attrib.xml

  > $ gzip child.xml

  > $ ls -l

  > total 1332

  > -rw-rw-r-- 1 mnutter mnutter 670757 Sep 23 08:28 attrib.xml.gz

  > -rw-rw-r-- 1 mnutter mnutter 681080 Sep 23 08:28 child.xml.gz

  > -rwxr-xr-x 1 mnutter mnutter 999 Sep 23 08:26 make.pl

  > -rwxr-xr-x 1 mnutter mnutter 976 Sep 22 13:16 make.pl~

  >

  > So child.xml.gz is slightly larger than attrib.xml.gz (about 1.5%,

  > assuming my math isn't as bad as my coding).

  >

  > That is what I would have expected. But, performance may be more than

  > network transport. I've got an app, where, for convenience, I have

  > wrapped MS's IE 5 MSXML parser, and I found a significant difference in

  > speed by placing stuff that there was only one of as an attribute
instead

  > of a child element. I don't know why yet, but the attributes where
faster

  > (I don't know if that was "recording" speed or "retrieval" speed) than
the

  > child elements using just that one parser. Of course, your mileage may

  > vary.

  >

  Blair L. Murri

  Sr. Programmer/etc.

  Wavo Corporation

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ic.ac.uk/pipermail/xml-dev/attachments/19990926/41903f30/attachment.htm