North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Which pipe dream should we be smoking? (was Re: Possible topic?)

  • From: Steven Schnell
  • Date: Wed Oct 22 09:38:00 1997

At 19:24 10/21/97 -0700, J.D. Falk wrote:
>On Oct 21, "Dorian R. Kim" <[email protected]> wrote: 
>> Then perhaps it might of interest to start a meta discussion about
>> problems of scalability of Gigaswitch based exchange points as well as
>> possible future direction of such large exchange points?
>	AFAIK, there are only two types of exchanges out there today:
>		1. Switched Ethernet and/or FDDI
>		2. ATM
>	We all know switched ethernet doesn't scale by itself,
>	which is why we've gone to switched FDDI.
>	And, it's quite possible that we're reaching the end of
>	scalability for switched FDDI -- unless, of course, some
>	amazing new technolgy comes along and fixes everything in
>	one swell foop.
FDDI is a finished standard and no more bandwidth upgrades are planned.
Ethernet, however, is now in the gigabit range.  It may be extended to
operate in the multi-gigabit range, if the IEEE decides the frame format
and the media access method will work at such speeds.  Products based on
GbE are in use today.  A NAP can drop in one or two GbE switches, connect
them to existing GIGAswitch/FDDIs, and scale (somewhat) from there.  Come
to NANOG this next week to see how this can be done.

The nice thing about the GbE switches (I note the Prominent Cajun P550 and
the DEC GIGAswitch/Ethernet switches) is that they give you instant access
to over 45 Gbps of (almost) non-blocking switch fabric. GigaLabs has a
monster of a switch with over 128 Gbps of non-blocking fabric (GbE uplinks
only however).  The only downsides right now with most current GbE switches
is that they only support 10/100BASE and GbE uplinks, offer no QoS and
limited flow-control mechanisms.  The flow-control protocol for GbE has not
yet been finalized by the IEEE.  All implementations of this are
vendor-proprietary.  (Be sure your vendor upgrades you to the Standard at
no cost!).

Connecting a router to a switch using a standard a Fast Ethernet adapter
gives you 200 Mbps of full-duplex throughput.  This is adequate for anyone
with an OC-3 behind their router.  Hell, anyone with more bandwidth into
the NAP can probably purchase multiple Fast Ethernet links from the NAP
operator.  These adapters are less costly than full-duplex FDDI adapters,
and the prices for switch ports and router adapters are declining.

I think that GbE technology permits the NAPs to support increasing
bandwidth requirements for the foreseeable future.  Beyond that, maybe the
ISPs with heavy-duty requirements should consider private interconnects
outside the NAPs.


>	Are the ATM-based NAP's having any problems with scalability
>	these days?  How far can they go?
>	Personally, I'd also be really curious as to any testing
>	that people have done or are doing in terms of other forms
>	of NAP fabric.  Will gigabit ethernet save us all?
>J.D. Falk                         voice: +1-650-482-2840 	
>Supervisor, Network Operations      fax: +1-650-482-2844
>"The People You Know.  The People You Trust."

                    Phone:  1.816.854.2113
                      Fax:  1.816.854.2201

            Numeric Pages:  1.800.724.3329, PIN 398.6644
              Alpha Pages:  1.800.724.3508, PIN 398.6644
             Outside U.S.: +1.619.279.8495, PIN 398.6644

   Text Page via Internet:  [email protected]