Although software is the most important ingredient of the next-generation Internet — from web applications to cloud-computing platforms —hardware vendors are not shy about marketing their wares as integral parts of the equation. This week, Cisco claimed it will revolutionize the Internet — the cloud included — with its forthcoming CRS-3 router, and startups like Tilera are shilling new chips designed with large-scale web applications in mind. If we accept the premise that hardware upgrades are necessary (they certainly can’t hurt), the question remains as to who will sell us this new gear.
On the chip front, Tilera has promise – and an additional $25 million – but it also has its work cut out for it. Cramming 100 cores onto a low-power chip is one thing; defeating Intel and AMD at their own game is something else altogether. Certainly, customers love more cores, but Intel and AMD aren’t exactly sleeping on multicore development. Intel’s six-core Westmere and eight-core Nehalem chips are on the way, as is AMD’s 12-core Magny-Cours chip – the first-ever production x86 processor to reach that number (although Intel does have a 48-core prototype).
Furthermore, as Stacey Higginbotham points out in the Tilera article linked above, multicore chips still struggle with linear performance issues, and software-development skills aren’t necessarily keeping up with all these additional cores. For Tilera, is offering too many cores too soon a legitimate concern?
Speaking of x86, there also is no telling whether customers will want to write to Tilera’s RISC chip architecture when the x86 they already know is growing faster and more efficient with every incarnation. Impressive fourth-quarter server-department sales for Intel and AMD indicate that customers aren’t shying away from the market leaders as compute demand increases. And if web companies really do opt for ultra-low power and ultra-high density, Intel’s x86 Atom processor is a viable candidate in its own right. The fact that Tilera and its chip-startup cohorts keep raising money suggests there is reason to believe they can succeed, but stealing significant market share from AMD and, especially, Intel is a lofty goal, indeed.
In the networking sector, it’s market leader Cisco that could find itself in a tougher fight than it might expect. The CRS-3 router promises performance gains that are impressive to say the least, and service providers will need more horsepower (if only to handle all the cloud-computing and video data Cisco will sending across their networks), but some believe an open platform like that offered by rival Juniper Networks is a better solution than “simply another box.” During the fourth quarter, in fact, Juniper bested Cisco in router revenue within the service-provider market (17 percent compared with 11 percent). So, will the CRS-3 truly revolutionize the Internet infrastructure and boost Cisco revenues, or is it, as the 451 Group’s Dan Kusnetzky suggests, merely a new transmission when providers are looking for a new car? Rumor has it that Cisco’s data-center rival HP wants in on the router market, too, and it certainly has the size to make a dent.
The numbers suggest we’ll see hardware evolutions to match the software advances that are driving Internet traffic, and service providers and web companies could be lining up to buy these new products. Whether they’ll be buying from their current vendors might depend on what they need.