Re: [HSSG] 40G MAC Rate Discussion
On Apr 4, 2007, at 23:20, Mikael Abrahamsson wrote:
> On Wed, 4 Apr 2007, Jason Weil wrote:
>
>> It seems to me the real benefit/driver for a 40G MAC would be in
>> the Metro and Long-Haul Ethernet market and not as a host
>> interface. This thinking may be slightly myopic, but the real
>> question for me is assuming the standard is ratified in 2009 would
>> users be presented with two interface options at that time? If
>> this were the case unless economics outweighed versatility and
>> capacity I can't see why the 40 would be a better choice than 100
>> given both options.
>
> I'd have to agree here. Only reason to do 40GE would be if someone
> said they could have it shipping at latest during 2008, and we
> estimate 100GE is not going to be until 2010.
>
> 40G would have made sense NOW, and not 2009-2010.
From a network operators perspective this is also my viewpoint. 40GE
would be of interest to support the growing capacity demands in the
networks *now* (of course only if provided with enough port density on
switches and routers).
If the 40GE standard is developed in IEEE time scales (i.e being similar
to that of 100GE) and available not before end of 2009 it will be less
useful as an aggregation technology for servers or to scale networks. In
that timeframe for many applications even 100GE will be too little.
The scaling factor of 10x that has traditionally been used for Ethernet
might have been just "tradition", but it also fits nicely the traffic
growth that we see on average in the Internet (100% traffic growth per
year is roughly a factor of 10 every 3 to 4 years). The time between
previous ethernet standards (E=> FE, FE=GE , GE=> 10GE) has been in
the same order (3 to 4 years). Following this, a 100GE standard should
have been available 2005 and in the 2009/2010 timeframe a next 10x
standard should be available. Since the 100GE wasn't there and a next
higher standard will not be there in 2009/2010 we have to resort to
massive use of Link aggregation, now using 10GE links and when 100GE
comes available as a product, using 100GE links.
Back to 40GE: scaling link aggregation using 10GE for another 3 years
will be very hard. The use of 40GE might be of help here if it would
allow for standardized products to become available say second half
of 2008.
QUESTION: Is there a way to expedite the standardization process (and
subsequent product development) of a 40GE standard? Within or outside
of the IEEE?
If the answer to the above is "no" then I would say lets not spend any
time on anything other than 100GE so no delay is introduced in the
development of this standard and get it finished as soon as possible.
After all, though 40GE might be a better fitting standard for servers,
there is no pressing need in the coming years for servers to operate
at these speeds.
Another point: the combination 40GE/100GE does not really fit well.
Implemented on a switch, I see interface blade combinations with either
2*100GE or 5*40GE. Or alternatively 1*100GE or 2*40GE (wasting 20GE in
capacity to the backplane) or 2*40GE plus 2*10GE ??
Since 100GE has been adopted as a standard, if there is a need for a
smaller data rate to connect servers, it 40GE the right rate?
- Henk Steenman