Re: Collisions in URLs using MD5

On Wed, 28 Jul 1999 adrian@creative.net.au wrote:

> Does anyone have figures for collisions of URL names when md5'ed ?
> I'm curious to know what it is like in the real world ..

I did some experiments in June 1997 using URLs from our SV cache. I
varied the length/size of an MD5 digest (in bytes) and varied the number
of days in the access log.

 trace length, number of number of MD5 collisions for
     days unique URLs a given URL digest length
                                 4 5 6 16
 ------------- ----------- ------ --- --- ---
             1 375066 13 0 0 0
             5 1494774 257 1 0 0
            10 2619168 817 2 0 0

Thus, for six byte and longer URL digests, there were no collisions in
the given set. A four byte URL digest gives negligible number of
collisions (817 or 0.04% for a 10 day trace). The standard MD5 digest
length is 16 bytes.

Alex.
Received on Tue Jul 29 2003 - 13:15:59 MDT