IPUrbia

The usually brilliant GigaOM wrote a little bit recently on the increased expansion of commodity computing, focusing, of course, on the x86 processor line. He threw in a mention to the upstart, vyatta, who is making a focused effort to bring commodity-based routing into enterprise and carrier markets, despite the fact that x86-based networking has had, at best, lukewarm success in its natural niche, SMB (usually in x86-based firewalls and content filters).

I’ve often had a lot of discussions on this line with friend of mine and have typically toed the Intel party line without fail. General sticker shock to any quote from Cisco/Juniper/ always brings out the “we can do it much cheaper mentality”, and, when dealing with small-scale networking projects, that is usually true. My gut instinct is that the raw speed of x86 processors and the low cost that allows you to simply throw quantity at the problem eventually wins in routing like it has in the server environment.

A number of extensive pricing comparisons later, I’m backing away from the Intel (well, I’m actually partial to AMD) love. A couple of days ago, I started pricing out SSL VPN solutions, while Intel won at small scale (500 concurrent sessions or under), Juniper, Cisco, and SonicWall all quickly starting more economical better after about 2000 concurrent sessions and almost 2.5x cheaper at 50,000 concurrent sessions. Did I mention that this is without counting software licensing costs for the Intel based solution? Did I also mention that this figure doesn’t even account for the cost of maintaining a whole rack of x86 1U servers, swapping out failed hard disks, and dealing with the increased maintenance of 25 small boxes compared to one huge monster. Still, to be fair to Intel, the cost calculation also doesn’t include the risk of back surgery after installing an 11U chassis .

This application is dealing in a space where x86 should shine–encryption is very CPU intensive, and my application is 100% Ethernet based, so it relies on commodity terminating hardware as well. The value proposition usually only gets worse once you start throwing in TDM and optical cards for x86. The other problem with the x86 kool-aid is that Cisco, Juniper, etc…have a lot of room to come down on price, as their pricing model seperates out the cost of the hardware and the cost of licensing. Considering their profit margins, if push comes to shove, the per-seat licensing might get dropped–and I’d definitely welcome any competition that can make that happen.

In a lot of ways, I think we may be seeing a downturn in commodity processing outside of the PC/server market. I remember back in the mid-to-late 90’s using Linux to turn old PCs into a firewall/router for small offices–a great deal when your alternative was a few thousand in Cisco hardware. That quickly eroded once Linksys threw the $50, idiot-proof device on the scene. In the VoIP space, I think things are going to come full circle again–the value proposition has been favoring x86/Asterisk (or whatever) for a couple of years now; Linksys/D-Link, etc…have been launching good $500-$1000 turnkey PBXs which greatly undermines the value proposition of x86-based solutions on that front. File servers? I’m going with the $200 NAS over the $1,000 server.

x86 processing does have its place–it has been the processing power of disruptive technology after disruptive technology. x86 has been the forerunner in the value priced SOHO market, but has always been supplanted by Linksys and company once any substantial demand develops. On the enterprise/carrier end goes, the economics again favor putting the money in hardware development to optimize routing hardware, etc. The server space favors commodity because it is so broad and generic; otherwise, the economics favor specialization.