jshelton@deimos.ADS.COM (John L. Shelton) (10/12/89)
We are running out of bandwidth on several of our seven ethernet subnets, and are considering the options. The quick-and-dirty option is to add more subnets, and upgrade our Cisco gateway. I'm not convinced this is a good long-range solution, however. I think a given is that we will not upgrade 100+ CPUs to FDDI; we can assume that all CPUs have Ethernet hardware interface, and will continue to have them for at least 2 years. I thought I had run across a network product last year that implements ethernet over fiber optic, but provides 100mb/s bandwidth. You attach to the fiber with interconnects that support up to 8 ethernet-based hosts (sort of like a DELNI), and the box gets the packets going on the fiber at very high speed. WHile any one host is limited to 10mb/s performance (ethernet limit), the aggregate network capacity is much higher. The benefit to such a scheme is clear: no one host can swamp a network; no need for subnetting; no gateway to slow you down; no worries about balancing load between subnets; no pulling more ethernets around a building; no trying to hook up to N ethernets is an already cramped cable closet. Can anyone give pointers to this (perhaps hypothetical) network system? Thanks. =John=
chimiak@umbc3.UMBC.EDU (Mr. William J. Chimiak ) (10/14/89)
In Article 3780 of comp.dcom.lans: John L. Shelton writes >We are running out of bandwidth on several of our seven ethernet >subnets, and are considering the options. The quick-and-dirty option >is to add more subnets, and upgrade our Cisco gateway. I'm not >convinced this is a good long-range solution, however. > >I think a given is that we will not upgrade 100+ CPUs to FDDI; we can >assume that all CPUs have Ethernet hardware interface, and will >continue to have them for at least 2 years. >I thought I had run across a network product last year that implements >ethernet over fiber optic, but provides 100mb/s bandwidth. You attach >to the fiber with interconnects that support up to 8 ethernet-based >hosts (sort of like a DELNI), and the box gets the packets going on >the fiber at very high speed. WHile any one host is limited to 10mb/s >performance (ethernet limit), the aggregate network capacity is much >higher. >Can anyone give pointers to this (perhaps hypothetical) network >system? Thanks. It seems to me that you are just implementing an FDDI concentrator. Each FDDI concentrator can then be placed in a ring. The LLCs of IEEE 802.3 and FDDI are the same but admittedly the drivers will be quite different. If the ethernet bandwidths and concomittent delays are satisfactory for your applications, an FDDI concentrator scheme might be the way to go while maintaining compliance with a strong standard so upward system evolution would be less painful. For new systems coming on line, be sure to do cost and performance trade-offs between FDDI and ethernet before choosing either one. For further thoughts on FDDI/Ethernet hybrid networks, see the Data Communications issue on designing FDDI networks - it is quite good. I think it appeared June 1989.
vjs@rhyolite.wpd.sgi.com (Vernon Schryver) (10/15/89)
In article <2406@umbc3.UMBC.EDU>, chimiak@umbc3.UMBC.EDU (Mr. William J. Chimiak ) writes: > > It seems to me that you are just implementing an FDDI concentrator.... An FDDI concentrator would require all existing ethernet stations to do 802.3 instead of ethernet. Since essentially no machines on ethernets use 802.3, that would be a problem. There is lots of talk about ether-to-FDDI bridges in the standards committees. However, there are some very big problems such as thru-put and the >4KByte MTU on FDDI. The original author's requirements might be met by one of the FDDI "tunnels" for ethernet. I think someone like Fibercom makes such devices. (Please excuse me if they don't or if others also do.) An FDDI-to-ether IP router might work. For example, you could buy one of the products shown at Interop to gateway between ether and fiber. As I understand "ethernet fiber repeaters" and "ethernet fiber transcievers", they have nothing to do with FDDI, except perhaps the fiber itself. Even the fiber connectors differ (ST's vs. MIC's). An number of companies make them and lots of places are now using them. They may be what the original author was thinking of. Vernon Schryver Silicon Graphics vjs@sgi.com
goodloe@b11.ingr.com (Tony Goodloe) (10/16/89)
In article <43037@sgi.sgi.com>, vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes: > Since essentially no machines on ethernets use > 802.3, that would be a problem. Something that I should, don't -- Is only difference between an Ethernet frame and an 802.3 frame the usage of the type/length field. I see how this causes problems. Are there any other differences that I don't know. I have a real 802.3 spec, but no Ethernet "spec", if one exists. tony "ground me but once" goodloe
goodloe@b11.ingr.com (Tony Goodloe) (10/16/89)
Some words got deleted the first time I posted this. (sorry). Something that I should <KNOW>, <BUT> don't -- Is <THE> only difference between an Ethernet frame and an 802.3 frame the usage of the type/length field. I see how this causes problems. Are there any other differences that I don't know. I have a real 802.3 spec, but no Ethernet "spec", if one exists. tony "ground me but once" goodloe
vjs@rhyolite.wpd.sgi.com (Vernon Schryver) (10/20/89)
In article <6207@b11.ingr.com>, goodloe@b11.ingr.com (Tony Goodloe) writes: > > Something that I should <KNOW>, <BUT> don't -- Is <THE> only difference > between an Ethernet frame and an 802.3 frame the usage of the > type/length field. I see how this causes problems. Are there any other > differences that I don't know. I have a real 802.3 spec, but no > Ethernet "spec", if one exists. > > tony "ground me but once" goodloe The "only" difference is the type/length field. However, that difference implies an extra 8-byte LLC header before the good stuff in IP, ARP, and ICMP packets (except where it is/was 3-bytes), as well as code to respond to some LLC ethernet packets. There are also changes in the "hardware type code" of ARP packets and perhaps RFC-951/bootp packets. The groady details are in RFC-1042 and IEEE 802.2. Similar things for FDDI (where we have no option) are in the soon to be replaced and/or updated RFC-1103. Vernon Schryver Silicon Graphics vjs@sgi.com P.S. There may be a difference in stuff like dribble bits, but who cares enough about such to look it up?