leonard@arizona.edu (08/28/90)
No, multiport repeaters don't buffer packets. The question is, what exactly does this Kalpana EtherSwitch thingy do when the segment it wants to transmit onto has a collision in effect? I think the answer is that it buffers. I got some literature from Kalpana and it says that there is a 256-packet buffer per Ethernet interface. The next question is, does the EtherSwitch act like a repeater and gobble preamble (so that you are limited to a max of 3 in series), or does it act like a bridge and reconstitute the header, so that the number you can have in series is limited only by propagation delay? The literature doesn't address this question. (I would assume that it regenerates the header.) Has anyone seen an EtherSwitch in action? Is anyone from Kalpana out there?
pat@hprnd.HP.COM (Pat Thaler) (08/29/90)
> > The next question is, does the EtherSwitch act like a repeater and > gobble preamble (so that you are limited to a max of 3 in series), > ---------- > Repeaters as defined in ANSI/IEEE Std 802.3c-1985 do not gobble preamble. They are required to transmit a full preamble (7 bytes of preamble followed by a byte of Start of Frame Delimiter). Not transmitting a full preamble can cause a false collision detect on coax due to analog details of coax segments and collision detect methods. The original Ethernet and IEEE 802.3 repeater definitions did not include this requirement. The original IEEE 802.3 repeater definition did include the note: "The fundamental concepts in the section are essentially correct, however, additional statements may be needed for a complete specification." The ISO publication of the same strengthened the note to: "Section 9 (Repeater Unit) of this standard is undergoing revision. DIS 8802-3/DAD 3 is intended to replace Section 9 and represents a substantial revision." ANSI/IEEE Std 802.3c-1985 has been published by IEEE in the _802.3_Supplements_. Since it has completed the ISO approval process, it will be published in ISO 8802-3 ANSI/IEEE Std 802.3, Second Edition replacing the original section 9. (This is expected to be in publication within a couple of months.) The repeater limit (4 according to 8.6.1 and 10.7.1 of IEEE 802.3) is based on round trip delay and on interpacket gap shrinkage. Pat Thaler
kwe@buit13.bu.edu (Kent England) (08/30/90)
In article <1990Aug27.143144.62@arizona.edu> leonard@arizona.edu writes: > >The question is, what exactly does this Kalpana EtherSwitch thingy do >when the segment it wants to transmit onto has a collision in effect? > >I think the answer is that it buffers. I got some literature from >Kalpana and it says that there is a 256-packet buffer per Ethernet >interface. I think you're right, but that sounds like a lot of buffer. The Kalpana has to have very fast table lookup for forwarding. Perhaps the 256 refers to the size of the MAC address table per interface? At any rate, I would be interested to know how big the MAC address table is, per interface, since that would affect where these things could be used. I can't imagine that Kalpana could accept arbitrary topologies with thousands of potential MAC addresses on some interfaces and still maintain claimed forwarding rates. --Kent