Latency is dependent on the application. If you want a *minimum*, then sending asynch requests *typically* gets to it's destination in 5usec or less (for short msgs... add on appropriate numbers for longer msgs, worst case packets [2k bytes at 400 Mbits/sec] would be about 80 usec). If there is a lot of traffic on the bus, and you need deterministic access with a lower max, then you need to use "isoch" requests which deliver a packet every 125 usec (with some jitter depending on the current traffic load... worst case delay is about 200 usec).... and the 1394b spec defines how to run gavanically isolated using UTP and various forms of optical fiber
I beg your pardon... FW costs about $3-$4, depending on complexity, providing, of course, that you are in high volume production.
Where does this $25 come from??? There is no combination of parts built this century that would cost so much!
The "Firewire" name was chosen by a bunch of engineers drinking too much beer after hours just before Comdex '93, when the project was about to go public (IBM, Apple, TI, WD, Maxtor, Seagate were all showing drives/silicon/systems). We were under the gun since the marketing droids would have picked some name like "Performa" if we hadn't acted soon. (The original project name, BTW, was "Chefcat"... the name of my favorite coffee cup). I don't remember which of us actually came up with "Firewire", but I was the one that pushed it through the chain of command.
The marketing guys made one change: from "Firewire" to "FireWire"...
Not so. Apple is not an IC manufacturer... it's a systems house. Zayante also was not an IC manufacturer, we were a provider of technology (IC designs, software, etc.) Some of the big IC houses licensed our designs for 1394b parts (Agere, for instance). There is no competition here...
BTW, many of the current Firewire IC houses got their start by licensing the designs that our team did while we were at Apple the first time. TI, Philips, Fujifilm, LSI Logic, Sony and Adaptec all started by licensing the Firewire designs that were developed at Apple from 1990-95. Only IBM, Fujitsu and AT&T/Lucent/Agere did independent designs back then.
Oxford, BTW, is an excellent little company, and the whole FW community thinks very well of them.
... Mike Teener, ex-CTO Zayante, ex-Godfather of Firewire.
1) No monopoly... Sony, TI, Panasonic, and a gaggle of others own bits of the IEEE 1394 technology (hence, there is a patent pool somewhat like MPEG and PCI). Zayante had significant presence in the 1394b (800mbit/sec -> 3.2 Gbit/sec) work, but so did Intel, HP, TI and Agere.
2) 1394 *is* the open standard. It was developed using IEEE rules (fully open meetings, just show up, no NDAs, no invitations required, patents must be disclosed). USB was an invitation-only thing. Apple tried to get involved with the USB 1.0 effort back in 1995 but was rebuffed ("first publicly drop ADB and Firewire and then we'll let you join the club").
I was the technical lead on Firewire at Apple from 1990 - 96, and was a founder of Zayante. There is no attempt at a monopoly, just an attempt to do a superior job.
Windows ME and Windows XP supports IP 1394 (try it!). The data rate isn't cool (only about 70 Mbit/sec) for all the reasons outlined earlier (crappy DMA design by Ethernet bigots that had no idea how 1394 really worked). The Unibrain people have shown how to do it right using a better (and simpler, BTW) DMA structure, and they get about 320 Mbit/sec.
Nope. 1394b supports shielded twisted pair at 3.2Gbit/sec at 4.5 m. The connector is different, however. There is a special "bilngual" port design (used by Apple) that allows either "a" or "b" operation, but the connector itself is new and not compatible with the old 6- or 4- pin versions. Sorry. You'll need a new cable...
The optical version is only needed when you want to go beyond 4.5 m. Glass optical fiber gets 1394b to 100 m at 3.2 Gbit/sec.
Yes, Apple will lose royalties from FireWire if developers move to it.
Uh, Apple doesn't make but a few cents from the FW "royalties"... it is just one firm (among about 10) that is part of the "1394 Licensing Authority" that gets $0.25 per end user system. ("end user system" means something sold at retail as a single package... like a car, a PC, or an airplane... no matter how many 1394 ports/ICs/devices are in the "end user system").
BTW, don't the 3Gb transfer rates depend on an optical connection?
Nope. Works fine on a shielded twisted pair (very much like existing FW cables). The connector, on the other hand, is different (and better... smaller, better shielding, etc.). Macs will use "bilingual" ports that are compatible with both legacy 1394a and new 1394b connections. Initial 1394b implementations will be 800 Mbit/sec, but the first 1.6 Gbit.sec versions will start to appear in 2002 (late-ish)... analog test chips have already been fabbed, but the actual deployment will depend on the success rate of the first S800 versions.
The standard doc (at draft 1.2) calls for up to 1.6 Gbit/sec over shielded twisted pair (same construction as current cable) with full specs. The new cable and connector have been tested at 3.2 Gbit/sec at 4.5 m, but none of the Si vendors are ready to sign up for a performance spec for the required signal yet. Maybe in a year or so...
3.2 Gbit/sec has also been tested at 100 m over 50 um multimode glass fiber (same as used by Gbit Ethernet and Fibre Channel).
Slashdot Mirror
Latency is dependent on the application. If you want a *minimum*, then sending asynch requests *typically* gets to it's destination in 5usec or less (for short msgs ... add on appropriate numbers for longer msgs, worst case packets [2k bytes at 400 Mbits/sec] would be about 80 usec). If there is a lot of traffic on the bus, and you need deterministic access with a lower max, then you need to use "isoch" requests which deliver a packet every 125 usec (with some jitter depending on the current traffic load ... worst case delay is about 200 usec). ... and the 1394b spec defines how to run gavanically isolated using UTP and various forms of optical fiber
I beg your pardon ... FW costs about $3-$4, depending on complexity, providing, of course, that you are in high volume production.
Where does this $25 come from??? There is no combination of parts built this century that would cost so much!
The "Firewire" name was chosen by a bunch of engineers drinking too much beer after hours just before Comdex '93, when the project was about to go public (IBM, Apple, TI, WD, Maxtor, Seagate were all showing drives/silicon/systems). We were under the gun since the marketing droids would have picked some name like "Performa" if we hadn't acted soon. (The original project name, BTW, was "Chefcat" ... the name of my favorite coffee cup). I don't remember which of us actually came up with "Firewire", but I was the one that pushed it through the chain of command.
The marketing guys made one change: from "Firewire" to "FireWire" ...
BTW, many of the current Firewire IC houses got their start by licensing the designs that our team did while we were at Apple the first time. TI, Philips, Fujifilm, LSI Logic, Sony and Adaptec all started by licensing the Firewire designs that were developed at Apple from 1990-95. Only IBM, Fujitsu and AT&T/Lucent/Agere did independent designs back then.
Oxford, BTW, is an excellent little company, and the whole FW community thinks very well of them.
2) 1394 *is* the open standard. It was developed using IEEE rules (fully open meetings, just show up, no NDAs, no invitations required, patents must be disclosed). USB was an invitation-only thing. Apple tried to get involved with the USB 1.0 effort back in 1995 but was rebuffed ("first publicly drop ADB and Firewire and then we'll let you join the club"). I was the technical lead on Firewire at Apple from 1990 - 96, and was a founder of Zayante. There is no attempt at a monopoly, just an attempt to do a superior job.
Windows ME and Windows XP supports IP 1394 (try it!). The data rate isn't cool (only about 70 Mbit/sec) for all the reasons outlined earlier (crappy DMA design by Ethernet bigots that had no idea how 1394 really worked). The Unibrain people have shown how to do it right using a better (and simpler, BTW) DMA structure, and they get about 320 Mbit/sec.
The optical version is only needed when you want to go beyond 4.5 m. Glass optical fiber gets 1394b to 100 m at 3.2 Gbit/sec.
Uh, Apple doesn't make but a few cents from the FW "royalties" ... it is just one firm (among about 10) that is part of the "1394 Licensing Authority" that gets $0.25 per end user system. ("end user system" means something sold at retail as a single package ... like a car, a PC, or an airplane ... no matter how many 1394 ports/ICs/devices are in the "end user system").
Nope. Works fine on a shielded twisted pair (very much like existing FW cables). The connector, on the other hand, is different (and better ... smaller, better shielding, etc.). Macs will use "bilingual" ports that are compatible with both legacy 1394a and new 1394b connections. Initial 1394b implementations will be 800 Mbit/sec, but the first 1.6 Gbit.sec versions will start to appear in 2002 (late-ish) ... analog test chips have already been fabbed, but the actual deployment will depend on the success rate of the first S800 versions.
The standard doc (at draft 1.2) calls for up to 1.6 Gbit/sec over shielded twisted pair (same construction as current cable) with full specs. The new cable and connector have been tested at 3.2 Gbit/sec at 4.5 m, but none of the Si vendors are ready to sign up for a performance spec for the required signal yet. Maybe in a year or so ...
3.2 Gbit/sec has also been tested at 100 m over 50 um multimode glass fiber (same as used by Gbit Ethernet and Fibre Channel).