How can the Internet have too many routes and not enough addresses?
- 21 August, 2014 03:12
The depletion of Internet addresses would seem to spell relief for aged routers that are struggling to deal with the Internet's growth, but the complicated interplay between those trends might cause even more problems.
Last Wednesday, some older routers and switches stumbled when the Internet's table of routes surpassed 512,000 entries, the maximum they could hold in a special form of memory called TCAM (Ternary Content Addressable Memory). The event drew widespread attention, though it was actually the third time in this young century that the Internet had broken through such a threshold. The number of routes exceeded 128,000 around 2003 and 256,000 in 2008, each time causing problems for some outmoded gear.
Devices that don't have room for all the routes may reboot themselves or fail to route some traffic, but the affected gear was fairly old. Cisco Systems says all the routing products it's sold for at least the past two years have had enough room in TCAM for more than 512,000 routes. Routers designed for the cores of carrier networks surpassed that long before. Juniper Networks, Cisco's longtime router rival, said it updated its gear for this problem more than 10 years ago. Alcatel-Lucent said its routers use a different memory architecture from the devices that got hit with the problem.
Because almost all the addresses defined by IPv4 (Internet Protocol version 4) have already been handed out to Internet service providers or end users, the number of routes allocated under that system may not grow much more, according to Cisco engineers. That would be one silver lining on a cloud that's hung over the network of networks for years.
"IPv4 cannot grow forever. We already reached a certain limit, so we personally wouldn't expect it to grow much larger," said Sasa Rasovic, incident manager at Cisco's Product Security Incident Response Team.
However, another danger remains, and it comes from the address depletion itself. With fewer IPv4 addresses at hand, users or service providers may want to split them up into smaller routes.
By common agreement among Internet engineers, the smallest accepted route on the Internet today points to a block of 256 consecutive IP addresses. (Using private addresses, companies and service providers can hook up many more devices behind those globally unique ones.) Now, some network operators want to break up those blocks so they can satisfy more customers, said Jim Cowie, chief scientist at Dyn, a traffic management company that recently acquired Internet analysis firm Renesys. Then, instead of one Internet route to reach the 256 addresses, there would be two.
"People are trying to do more with less," Cowie said.
Along the way, some may also be putting profit ahead of the Internet's ease of use. IP addresses officially are handed out free by nonprofit regional authorities, but their supplies are mostly gone. The mad dash for IPv4 addresses has led to some unseemly practices by those who already got their addresses.
"As the IPv4 address space is now depleted, a few smaller routes ... are being sold to other entities. Apart from a number of other more serious issues this is causing to the Internet community at large, this also has potential to cause a growth of the routing table size," Cisco's Rasovic said in an email message. "It's hard to predict just how fast and how big of an impact this will have in the future."
If some service providers start to split up the smallest blocks into even smaller ones, that could even affect whether all users can reach everyone else on the Internet, Dyn's Cowie said. Other operators might filter out the smaller routes, keeping their own routing tables a more reasonable size but not offering access to some addresses, he said.
And though it's impossible to say how many new routes might result, routers would continue to face a growing number of them. Like new party guests who want a piece of the same pie, Internet address holders could cut the IPv4 address space into ever smaller pieces, and it would fall onto the routers to keep track of all the slivers.
Dave Schaeffer, CEO of ISP Cogent Communications, thinks the routing tables will keep growing just from new addresses coming online.
"There's still a big, dark pool out there of IPv4 addresses in the hands of service providers that can be routed, that are not routed (yet)," Schaeffer said.
Migrating to IPv6 would eliminate the address shortage, because the newer protocol has an almost unlimited supply. Few users have adopted IPv6 even though most systems and networking gear have long been equipped for it. The IPv6 routing table still only has about 20,000 routes in it, Cowie said. That's what makes it feasible for Cisco to suggest, among other things, that network operators reassign some of the memory in their routers that was automatically set aside for IPv6 routes and give it to IPv4 routes.
But the short supply of older addresses and the expected growth of the Internet of Things eventually will bring more IPv6 addresses into service, Cowie said. That will raise issues of its own.
"Now that IPv6 has been introduced, more and more devices are going to be connected," Rasovic said. "The tables are different [in IPv6], and they're managed differently in memory." It's hard to know how many IPv6 routes there could eventually be, Cowie said. Those routes will all take up more memory, because an IPv6 address is much longer than one from the older version. Network engineering groups are already trying to figure out how to manage IPv6 routes, according to Cisco.
The IPv4 route-table problem will be with us for a long time, according to Cogent's Schaeffer. Some of Cogent's customers were affected by the surge in routes last week, though on their own equipment rather than Cogent's, Schaeffer said.
Conscientious ISPs may aggregate their own routes to help bring the tables back from the limit, but the reprieve will only be temporary, he said.
"We may, for a short period, fall below 512, but inexorably, the trend is larger tables."