Re-discussion on the practicability of Network accelerator card (TOE)

It seems that some readers of the Toe Network accelerator card Working mechanism is not clear, it is necessary to explain.

Just now, an old friend (really "veteran") sent me an email stating the necessity of toe. The full text of the email is published below to share with you:

"I totally agree with the teacher's opinion," Shimen said. I'm an engineer, so I just want to add a few more figures, here are some of my additions:

"IP network in 2, 30 years ago, only dozens of K speed, CPU packaging and unpacking can also cope." And the results of the network is increasing speed, today has gigabit Ethernet to the desktop, Gigabit Ethernet running on the LAN, backbone network has been upgraded to 40G (40,000 trillion) wavelength division multiplexing. But the speed of the CPU can not catch up with the speed of the network,-in general, 2, 30 years to speed up the network hundreds of times, CPU frequency only speed up dozens of times times, good, the problem. Think that readers may know the principle of handling a TCP/IP bit that requires a single hertz frequency. Then bi-directional Gigabit IP network will consume 0.8 2.4G frequency of Ben four core, million IP network will consume 8 cores. The kernel consists of cpu,cpu that make up the server, which means that the processing of these packets requires a large number of servers. ”

I (ie Shimen) think: In fact, the concept of toe has long been, not something new. But making "products" has been a matter of recent years (see the history of the American Intilop Company). In our country, the market for toe is now almost nil. However, the market is what we need to create, not born in existence, otherwise, we need to do business.

Someone started cursing toe, which of course is perfectly OK. However, I hope it is not because of my personal reasons (and abuse). ...... I may not be able to explain the principle of toe. In the future, I must continue to go on, to understand it.