The 2006 Wireless Networking Summit
Posted Apr 13, 2006 7:21 UTC (Thu) by tialaramex
Parent article: The 2006 Wireless Networking Summit
Wireless _Ethernet_ is the whole point of the exercise. To complain that down in the engine room these devices aren't actually 100Mbit Ethernet controllers with a UTP PHY is to completely miss the point of Ethernet and its success.
A huge number of protocols, and protocol implementations have been written to run on top of an Ethernet link layer, some of them very tightly bound to Ethernet's specific features (e.g ARP). If you invent something quite different, let's call it Tenrehte, for your wireless networking, then none of that stuff works. No TCP/IP, no custom homebrew network code, nothing. So once you've started to ship engineering samples you need to go through a multi-month process to standardise layering of old protocols onto your new link layer. You need an IETF IP-over-Tenrehte working group, it produces specifications, a few people implement them, the specifications become standards track documents, updated software trickles out to end users and all the while you're doing this, a Wireless Ethernet product could be on shelves as a more or less drop-in replacement for wired products.
We can follow Tenrehte's story for real with IRDA and Bluetooth on the desktop (notice how it took a long time for Linux to support anything better than a serial link with these link layers?) or with ATM on bigger networks. It's less painful to convert newer protocols like IPv6 (no ARP, so the interface to the link layer is simpler), but it's still months of extra work, so it must only be done if it's really necessary to have a new link layer.
This is the same phenomenon as with PCI. Yes, PCI-X and PCIe aren't actually PCI hardware, the electrical connections are different, the speed is different, and down in the engine room of the kernel, there are different register settings, different hotplug behaviour, maybe different signalling and interrupt rules etc. But for most of the software outside that "engine room" these are just faster, better PCI devices. Or think of the Linux SCSI subsystem. A SATA disk isn't really a SCSI device, no conceivable electrical adaptor would make it plug into my UW SCSI controller, but for very good reasons Linux treats my SATA disks as SCSI devices and the SATA drivers in Linux do the extra work to accomodate this.
So, Wireless Ethernet is called so for a good reason even if it's a lot easier to write a Linux kernel driver for an NE2000 than for the IPW2200. The only NE2000 I have uses 10base2 anyway, so it won't even plug into today's "real" Ethernet networks, because those use 100baseT or better.
to post comments)