nanog mailing list archives
Re: what about 48 bits?
From: Stephen Sprunk <stephen () sprunk org>
Date: Tue, 06 Apr 2010 21:39:44 -0500
On 05 Apr 2010 12:43, Valdis.Kletnieks () vt edu wrote:
On Mon, 05 Apr 2010 13:29:20 EDT, Jay Nakamura said:I would have attributed the success of Ethernet to price!You've got the causality wrong -- it wasn't cheap, way back when.I remember back in '93~94ish (I think) you could get a off brand 10BT card for less than $100, as oppose to Token Ring which was $300~400. I can't remember anything else that was cheaper back then. If you go back before that, I don't know.Steve is talking mid-80s pricing, not mid-90s. By '93 or so, the fact that Ethernet was becoming ubiquitous had already forced the price down.
Ah, but what _caused_ Ethernet to become ubiquitous, given the price was initially comparable? The only explanation I can think of is the raft of cheap NE2000 knock-offs that hit the market in the late 1980s, which gave Ethernet a major price advantage over Token Ring (the chips for which all vendors _had_ to buy from IBM at ridiculous cost). That, in turn, led to mass adoption and further economies of scale, pushing the price lower and lower in a virtuous cycle. Still, lots of shops stuck with TR well into the mid- and even late 1990s because Ethernet didn't perform as well as TR under moderate to high utilization by multiple hosts, not to mention IBM's insistence that TR was required for SNA. It wasn't until Ethernet switching came out, mostly solving CSMA/CD's performance problems and eventually leading to full-duplex operation, that it was entirely obvious which was going to win, and I spent several years doing almost nothing but helping large enterprises convert to Ethernet (usually with the help of DLSw). By that point, off-brand Ethernet chips cost _less than 1%_ of what IBM's TR chips did, thanks to competition and sheer volume, so vendors had started including them "for free" on every PC and server, and that was the final nail in TR's coffin. (LocalTalk, ARCnet, and a variety of other physical layers suffered a similar fate, but unlike IBM, their backers quickly switched to Ethernet when they realized they couldn't compete with it on price _or_ on performance given their limited volumes, so those deaths were more sudden and absolute than TR's.) As to why no other technology has managed to dislodge Ethernet, though, I think it's fairly clear that's because the various successors to 10BaseT have all maintained the same connector and the same framing, which makes for trivial upgrades that deliver regular (and significant) performance improvements as customers' equipment replacement cycle turns. S -- Stephen Sprunk "God does not play dice." --Albert Einstein CCIE #3723 "God is an inveterate gambler, and He throws the K5SSS dice at every possible opportunity." --Stephen Hawking
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature
Current thread:
- Re: what about 48 bits?, (continued)
- Re: what about 48 bits? Steven Bellovin (Apr 05)
- Re: what about 48 bits? Jay Nakamura (Apr 05)
- Re: what about 48 bits? Valdis . Kletnieks (Apr 05)
- Re: what about 48 bits? Steven Bellovin (Apr 05)
- Re: what about 48 bits? Nick Hilliard (Apr 05)
- Re: what about 48 bits? Jeroen van Aart (Apr 07)
- Re: what about 48 bits? Vince Fuller (Apr 08)
- Re: what about 48 bits? Barry Shein (Apr 05)
- Re: what about 48 bits? Steven Bellovin (Apr 05)
- Re: what about 48 bits? Bill Stewart (Apr 05)
- Re: what about 48 bits? Stephen Sprunk (Apr 06)
- Re: what about 48 bits? joel jaeggli (Apr 06)
- Re: what about 48 bits? John Kristoff (Apr 07)
- Re: what about 48 bits? Roland Perry (Apr 07)
- Re: what about 48 bits? Joe Greco (Apr 07)
- Re: what about 48 bits? Roland Perry (Apr 07)
- Re: what about 48 bits? Joe Greco (Apr 07)
- Hubs on a NIC (was:Re: what about 48 bits?) Lamar Owen (Apr 07)
- Re: Hubs on a NIC (was:Re: what about 48 bits?) Joe Greco (Apr 07)
- Re: Hubs on a NIC (was:Re: what about 48 bits?) Steven Bellovin (Apr 07)
- Re: Hubs on a NIC (was:Re: what about 48 bits?) Joe Greco (Apr 08)