When I started tracking security accelerator ICs a few years back, Xerox Parc scientist Tom Berson shared with me a piece of his wisdom. Tom contends that security will become ubiquitous in communications when it is a nearly free feature-one that is transparent to end users and mostly transparent to hardware designers.
It strikes me that Tom's wisdom applies fairly well to most new technologies. In the same way that security acceleration captured the imagination of engineers a few years back, TCP/IP offload engines (TOE engines) are capturing their minds today. Moreover, I contend that TOE engines quickly will become ubiquitous. The problem, for those trying to make money off the technology itself, is that these engines will be integrated as just another feature on system-on-chip silicon. Hence, it will be transparent and free.
In a nutshell, TOE engines have come into existence because higher networking speeds-particularly Gigabit Ethernet and 10-Gigabit Ethernet-are taking too many cycles from host microprocessors in PCs, servers and workstations. A rough rule of thumb is that it takes 1 microprocessor hertz to handle 1 bit per second of TCP/IP over Ethernet traffic. So, without TCP/IP hardware acceleration, a rocking 2-GHz Pentium 4 processor wastes half its processing power on networking overhead. Seems a pity.
Until now, interest in TOE engines was highest among those with ambitions in storage, particularly iSCSI. However, where TOE engines will find ubiquity is in Ethernet silicon that resides on network interface cards or LAN-on-motherboard implementations. As Ethernet functionality migrates into the PC core logic chip set, so too will TOE engine functionality.
There will be winners and losers as TOE engines rapidly become free and ubiquitous. End users win with better computing system performance at zero extra cost. Provided Intel and Broadcom integrate TOE engines fast enough, they win, too, by defending their turf in the Ethernet adapter silicon market.
The losers are likely to be standalone TOE engine companies simply because integration will be so rapid and there is enough competition that intellectual property licensing revenues will be limited. There will, of course, be opportunities for standalone TOE engines inside of communications and storage infrastructure. But the big homerun for the technology will be inside of Ethernet silicon in PCs, servers and workstations.
Jeremey Donovan ( Jeremey.Donovan@Gartner.com) is chief analyst at Gartner Dataquest.