Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: Long distance links

Paul (Bottorff i.e.),

> I agree if this is what is done. If the effective data rate coming out of
> the Campus is 9.584640, then there is no problem. If, on the other hand,
> the data from the campus is sourced at 10.000 and is carried on dark fiber
> for say 1000 Km (using a few repeaters), then enters a DWDM network the
> device at the edge of the DWDM network must slow the data rate down by a
> flow control mechanism.

Now I'm confused!  If one's able to send packets, through repeaters, over 1000
km at 10Gbps, this has to be someone's dark fiber and not a carrier's managed
DWDM service, right?  If so, then it stands to reason that the same party would
likely want to put a bridge/switch at the end of this long link before passing
the packets to the carrier network to ensure that the "private" link is "managed
internally" and not at fault, in case argument arise with the carrier as to
where the network problem is.

I raised the opposite question some time ago.  A pair of CPE switches connected
to their respective transponders over relatively short distance, which are in
turn connected by long haul DWDM network.  Assuming the CPE switches are "rate
matched" to 9.58464Gbps such that the transponders don't need big buffers, it'll
still be incumbent on the owner/operator of the CPE switches to ensure that the
switches have adequate buffer to allow for the round-trip delay over the long
haul link such that MAC "flow control" will still be effective.  But this
additional hassle, which may be overlooked, doesn't seem to make the whole
proposition any more appealing to the user.  I'd think that it'll make more
sense for the carrier to position a switch between the long haul DWDM network
and the CPE, thereby guaranteeing the performance of the transport service
provided.  If so, then it follows that the transponder may well be integrated
into this "WAN access switch" to further reduce the cost, resulting in a WAN
"access" I/O card with the requisite large buffer (and associated cost).  In
both cases, there's no buffering problem at the transponder, but suggest the
need of WAN I/F for the switches.

Did I miss some key assumptions in the logic that would result in a different