Another Y2K-like problem, this time Internet routers are the problem

Read an article today in Wired about The Internet has grown too big for its aging infrastructure showing up as a serious problem that’s soon to be more widespread.

This Y2K-like problem is associated with the Border Gateway Protocol  (BGP) routing tables entries which represent IP address prefixes.  Internet routers keep BGP tables in Tertiary Content Addressable Memory (TCAM, sort of like a virtual memory page table only for router addresses) and there are physical limits as to how many BGP entries will fit into any specific Internet router.  Some routers crash when they exceed their TCAM limit and others just ignore the BGP entries that exceed their limits – neither approach seems workable long term.

Apparently we are approaching one of those hard and fast limits, at least for older routers, as the BGP routing tables reach over 512K entries.  As of May 2014, there were in excess of 500,000 BGP prefixes (table entries).

Smoking gun points to …

It appears that this time Verizon was the perpetrator. Yesterday they added 15K BGP entries to the Internet BGP table, kicking some routers over their 512K limit. This was no doubt in anticipation of some growth in Internet addresses on their networks.

The result was that LiquidWeb’s network went down. Supposedly they have an older Cisco 7600 router and the latest addition to BGP entries exceeded its TCAM capacity, crashing their router. Oops!

Verizon quickly withdrew the offending 15K BGP entry addition and things seem back to normal for the moment. But we are once again close to some arbitrary computerized limit. Only this problem won’t happen at midnight December 31st. It won’t take that long to exceed the current BGP entry limits again and next time it might not be that easy to back out.

But it’s almost like there’s no stopping it…

Just guessing here but these types of routers probably have similar limits for BGP entries exceeding 1024K entries, 2048K, 4096K, etc. With the number of internet connected devices growing exponentially, especially with the Internet of Things, I predict similar problems over the coming years. Indeed, we went from ~400K to ~500K BGP entries in just under two years and the rate of growth seems to be accelerating.

It’s really just a matter of time before even todays routers run out of TCAM slots. Y2K-like, only this time there’s no way to stop it from happening again and again in the future.  I suppose it would be better if the routers just ignored the new BGP entries rather than crashing but that would seem to put some segment of Internet routers out of their reach?  There’s got to be a way to intelligently ignore some updates or summarize prefix updates when a router runs out of TCAM entries.

Welcome to the new 512K problem.

~~~~

Comments?

Photo Credit(s): Cisco 7609 @ itb for INHERENT by Affan Basalamah

 

 

 

This entry was posted in Distributed computing, Internet of Things, Internet traffic, Networking, System effectiveness and tagged , , , , . Bookmark the permalink.