By: David W. Hess (dwhess.delete@this.banishedsouls.org), September 26, 2007 1:22 am
Room: Moderated Discussions
Jonathan Kang (johnbk@gmail.com) on 9/25/07 wrote:
---------------------------
>Michael S (already5chosen@yahoo.com) on 9/25/07 wrote:
>---------------------------
>>Aaron Spink (aaronspink@earthlink.net) on 9/24/07 wrote:
>>---------------------------
>>>Michael S (already5chosen@yahoo.com) on 9/24/07 wrote:
>>>---------------------------
>>>>I don't know whether it is blocking or not, but pass-through routing fall out of
>>>>favor in high-speed network infrastructure equipment a long time ago. I'd guess
>>>>there were reasons for preferring store-and-forward implementations over pass-through
>>>>probably related to corner cases in error recovery. So I expect that at least first
>>>>few generation of CSI would avoid path-through.
>>>>
>>>
>>>Many high performance networks today are designed to MAXIMIZE pass-through routing
>>>via means of virtual cut-through mechanisms.
>>>
>>
>>Not in GBE switching. I didn't check recently, but 2-3 years ago all popular switches
>>were store-and-forward. On the other hand, older 100BaseT switches are mostly pass-through.
>>
>
>This is most likely because 100BaseT networks were broadcast networks where the
>medium was shared. It made sense to pass-through as the switch basically acted as a physical medium.
>
Wouldn't medium sharing only apply to a hub and not a switch in which case no routing is performed? I never came across many 100BaseT broadcast networks although in at least 2 cases I replaced switches with hubs to correct flow control problems.
Cut-Through switches begin forwarding the packet to the destination after interpreting the header which allows lower latency (7.5 microseconds versus about 160 microseconds for a 1500 byte payload on 100 Mbit/s ethernet per hop) at the expense of CRC error checking which has to be done at the destination. I presumed that it was not used with gigabit ethernet because of the store-and-forward latency improvements that came with increased transmission speed.
From what I remember reading the details of AMD's Hypertransport switching matrix, store-and-forward is always used with multiple packet queuing at every transmit port.
---------------------------
>Michael S (already5chosen@yahoo.com) on 9/25/07 wrote:
>---------------------------
>>Aaron Spink (aaronspink@earthlink.net) on 9/24/07 wrote:
>>---------------------------
>>>Michael S (already5chosen@yahoo.com) on 9/24/07 wrote:
>>>---------------------------
>>>>I don't know whether it is blocking or not, but pass-through routing fall out of
>>>>favor in high-speed network infrastructure equipment a long time ago. I'd guess
>>>>there were reasons for preferring store-and-forward implementations over pass-through
>>>>probably related to corner cases in error recovery. So I expect that at least first
>>>>few generation of CSI would avoid path-through.
>>>>
>>>
>>>Many high performance networks today are designed to MAXIMIZE pass-through routing
>>>via means of virtual cut-through mechanisms.
>>>
>>
>>Not in GBE switching. I didn't check recently, but 2-3 years ago all popular switches
>>were store-and-forward. On the other hand, older 100BaseT switches are mostly pass-through.
>>
>
>This is most likely because 100BaseT networks were broadcast networks where the
>medium was shared. It made sense to pass-through as the switch basically acted as a physical medium.
>
Wouldn't medium sharing only apply to a hub and not a switch in which case no routing is performed? I never came across many 100BaseT broadcast networks although in at least 2 cases I replaced switches with hubs to correct flow control problems.
Cut-Through switches begin forwarding the packet to the destination after interpreting the header which allows lower latency (7.5 microseconds versus about 160 microseconds for a 1500 byte payload on 100 Mbit/s ethernet per hop) at the expense of CRC error checking which has to be done at the destination. I presumed that it was not used with gigabit ethernet because of the store-and-forward latency improvements that came with increased transmission speed.
From what I remember reading the details of AMD's Hypertransport switching matrix, store-and-forward is always used with multiple packet queuing at every transmit port.