sorry, but ...

[Peter Fairbrother]

On 13/08/12 21:57, Derek Fawcus wrote:
> On Wed, Aug 01, 2012 at 10:05:57AM +0100, Andrew Bangs wrote:
>> Intercepting at switches (vs literally 'on the wire') always reminds me of Zeno's arrow paradox.
>> We somehow have to pretend that the thing to be intercepted is stationary if this is really different from doing it on the wire,
>> or at least that it's received and then has stuff done to it before being sent on its way.
>
> No pretense is necessary,  that is usually how they operate;  with store and forward on the routers/switches.
>
>> (for anyone thirsty for further reading, "cut-through switching" is a phrase your favourite search engine could investigate)
>
> I think there were some early switches which did cut through,  as a way of reducing
> latency,  but the down side is that damaged packets can then propogate.
>
> I don't know of any IP routers which do cut through forwarding when operating at the IP level.
> Likewise for most MPLS switches.  The packet is fully received,  and checksums verified before
> retransmission.
>
> I think I may have read of some new optical switches (with an MPLS varient) which do something
> like cut-through,  but it is more a case of lambda switching;  where certain MPLS tags get
> associated with specific colours.  This would generally be in the core of the network,
> and one is then in to having to use optical splices to intercept.
>
> I would suggest that 90% (or more) of current internet routers/switches are store and forward,
> even when the have hardware forwarding engines.



IIRC that's the way the internet was designed; it is built on packets 
rather than connections.

Cut through switching is considerably more complicated and expensive 
than store-and-forward; the expense is only really justifiable where 
very low latency is critical, eg in supercomputer networks.

It's not needed at all for the everyday internet.



[and I expect even cut through switches store-and-forward at least one 
bit, for a gazillionth of a second - a bit is not a communication, but 
then a packet is not usually the whole of a communication either]


[not even I would expect judges to fall for that... but what do I kno'?]


-- Peter F.




> .pdf
>
>