/24s run amuck

Sean M.Doran smd at cesium.clock.org
Wed Jan 14 12:14:35 UTC 2004


> Sprint and a few others used to filter on /19s, 'cause that's what 
> ARIN & others handed out.  They changed that to /20s when the rules 
> changed.  Sprint gave that up.

The filtering was done on the /18 because that was what I expected we 
could easily afford to support in terms of memory and computation, in 
terms of maximum number of prefixes.

The move to /19s was driven by two arguments: firstly, the regional 
internet registries explained how they would not allocate out half the 
available /19s within a generation of routing equipment, and secondly, 
it squelched many of the usual sources of complaint.

The deployment of progressive flap-damping further relieved the need to 
filter on short prefixes, and the subsequent complementary deployment 
of progressive maximum prefix count limits have essentially eliminated 
the need to do prefix-length filtering at all.   Long prefixes now are 
simply less reliable than the covering shorter prefixes allocated by 
the RIRs.   Just how unreliable a given prefix is would be difficult to 
predict, which is a misfeature, but the routing system as a whole is 
much more robust than it was a decade ago.

Unfortunately there has been a macroeconomic cost to the growth of 
background noise in the Internet -- and the noise is still there -- 
which has made the Internet as a whole more expensive and less widely 
available than it ought to be.  However, there are much larger 
contributions of such waste outside the public Internet's routing 
system that dwarf the cost of the unnecessary demands on router power 
resulting from poor aggregation, poor hygiene, and poor stabilization 
practices.

> Almost everyone filters on /24s - they do not want to see /32s in the 
> global table.

Why not?   I'm curious about why /24s are OK but /32s are not.

I suggest that if there is no reason other than a watered down version 
of the voodoo mentality you've accused me personally of having with 
respect to long prefixes -- i.e., if you think I'm right about the 
problem but too aggressive about the limit -- that there is a business 
opportunity still waiting to be exploited by someone enterprising.

With respect to that, for my part I wish I could go back in time and 
complete the next phase of the filtering, viz. a web page which would 
accept a credit card number from anyone who wanted to have a particular 
prefix allowed through the access-list, for a small recurring fee.

Live and learn...

	Sean.




More information about the NANOG mailing list