Snow days ahead for the Internet

It’s finally spring, at least here in the Northern Hemisphere. But the forecast for the Internet isn’t particularly warm and sunny, at least if some of the world’s leading experts on Internet technology can be believed.

Let me explain: Some years ago, one of my colleagues used to talk about “‘Internet snow days”‘ as a way to characterize the Internet’s unpredictable performance. In essence, he meant that just as the real world will experience unpredictable weather patches during which normal activities are curtailed, every now and then the Internet will experience unavoidable performance degradation.

I’ve always liked the Internet snow days concept — and it popped back into my head after I moderated a panel session on the future of the Internet at the recent Future-Net trade show. Panel participants included some of the brightest and most imaginative minds in the ‘Net — folks who were in many cases there from the very beginning.

The upshot? Batten down the storm windows; Internet snow days are coming.

If you’re a regular reader of this column, you’ll probably remember that back in the fall, Nemertes Research found that demand for Internet services was likely to exceed access capacity within the next five years or so. Panelists didn’t disagree, and according to them, that’s just the beginning.

The single biggest issue highlighted by the panel was economics: Internet service providers generally don’t make money from providing Internet connectivity services. Without the ability to ensure profits, providers (whether traditional telcos, cable companies or potential new entrants such as Google) have little incentive to invest in access, which means the problem is unlikely to get solved, at least in the short term.

There’s more, though. Several panelists stressed concerns over IPv4 address exhaustion, and sketched out how carriers are likely to handle the challenge going forward. Bottom line: Be prepared for carriers to be far more aggressive than previously about reclaiming address spaces, and parsimonious about handing them out. Depending on how (and whether) a transition to IPv6 occurs, there may also be situations in which you literally “‘can’t get there from here”‘: That is, users on one IP version can’t connect to users or resources on another.

Panelists also tended to agree that long-simmering issue of route-table expansion is poised to boil over. The issue, in a nutshell, is that route table size is increasing faster than route-lookup performance improvements, which means that routers will take an increasingly long time to locate routes. Moreover, the issue is CPU cycles — which means adding memory doesn’t fix the problem (and adding processors can be prohibitively expensive).

Finally, multihoming remains an unsolved (and knotty) problem. Note: moving to IPv6 doesn’t fix either route-table expansion or multihoming; arguably, it makes both worse.

None of this is new — folks have been warning about these and related issues for a decade or more. But what stood out on the panel discussion was a heightened sense of urgency, and a shortened prediction timeframe. My take? Don’t pack that snow shovel away just yet.