Third fighter joins FireWire, USB 2.0 fray
Why choose 1394 or USB when 1355's better than both?
Reader's Comment Following our coverage of the 'Intel snubs FireWire' and USB 2.0 debate, we received the following letter from Paul Walker, a representative of the IEEE 1355 Association: Thanks for the articles about Intel's USB 2 and the response from James Snider of the 1394 Trade Association. Mr Snider did not mention that the 1394 standard is currently under revision, to both 1394a and 1394b. Both revisions have taken far longer than planned, and the 1394a revision recently failed its ballot for approval. The reason for the delay is that the standard has been bent to applications it was never intended for, and which have requirements that 1394 fundamentally fails to meet. The Working Groups deserve great credit for their work, but they face a Herculean task, and to do the job properly may require a fundamental shift. Intel, with USB 2, has clearly decided that such a shift is necessary. But USB, while simpler than FireWire, is still a bus, and all the modern network technologies, such as Switched Ethernet, ATM, FibreChannel, Myrinet and IEEE 1355, are switched. Intel's own NGIO is switched. So any new technology which does not adopt the switched paradigm is obsolete before it hits the streets. Some of the switched technologies are, admittedly, more expensive than the 'simple' buses. But actually it is a lot easier if the logical topology of the network matches the physical topology, which is now always a set of point-to-point links. Incidentally, it is more ecological also if the packets go directly between A and B, rather than having to be bussed past all A to Z. Consider IEEE 1355, an equivalent of a UART can be produced which fits the smallest Spartan $2.95 FPGA from Xilinx, and which uses less than a third of the logic of a 16550 UART in Xilinx. I have the 1355 "UART" running on the bench at 140MBaud in the slowest XCS05 chips; it is asynchronous, AutoBaud, flow-controlled, and carries a very simple packet protocol. A four-port switch for 1355, including the ports, fits in less than half the amount of Xilinx logic required for a three-port USB Hub. The bandwidth of a switched network scales. So that a large network of 1000 nodes of 1355 has bandwidth of many tens of GBps, far higher than any conceivable bus. There is no constraint on the topology --- loops are allowed and indeed are necessary for reliable systems. (1394 disallows loops so that a break anywhere in the network splits the network into two completely separate networks, neither of which can talk to the other.) 1355 can be honoured with Intel's Mr Gelsinger's reference to "Niche Technology" (see this story), because it is being used in Space, where they need reliability above everything else. 1355 is also being used in other niche applications, such as super-computers (for performance), in data acquisition (for low latency), and in telecoms (for low cost). But Intel's announcement of USB 2 and description of 1394 as niche technology has thrown open the wider debate once again. Sooner or later, the industry will have to shed the legacy of the bus and move to simple, switched, networks. The sooner the legacy is shed, the less work and investment will have to be thrown away and done over again. The combination of reliability, much lower fundamental cost than even USB, scalable performance, ease of use and flexibility, make 1355 a pretty good, genuinely "Universal", choice. ® Paul Walker is Editor to the 1355 Association Web Site, www.1355.org The Register welcomes letters for publication. We reserve the right to edit for clarity and conciseness.
Sponsored: Hyper-scale data management