Re: [HSSG] 40G and 100G
Sure, arguments can be made in anyway, but for the server NIC and the fabric inteconnect as well 40G doesn't seem to make a whole lot of sense. For the server NIC next step by 2010/12, IMO, 20G seems to make more sense, specially, if the 10G optics (and catX) scales to 20G without much added cost. For the fabric interconnect faster is better (like 100G).
On 6/13/07, Hays, Robert <email@example.com> wrote:
While the "volume server" market need for 40G interconnects is ~5 years
out, our colleagues at IBM and Sun Microsystems have demonstrated that
there is earlier demand for 40G interconnects in High-Performance
Compute (HPC) clusters. I don't think it would be appropriate for IEEE
to gate the development of a standard based on arbitrary volume
thresholds. Vendors & end-users will make different decisions on when
to adopt new speeds based on their particular needs and budgets.
I agree that much of the 40G standard will be able to leverage work to
be done for 100G, but I don't think that necessitates delaying 40G. To
address your concern, we will work on a proposal for how 40G and 100G
standards can be developed in the most efficient way possible without
artificially delaying one or the other.
From: Frank Chang [mailto:ychang@VITESSE.COM]
Sent: Wednesday, June 13, 2007 3:24 PM
Subject: Re: [HSSG] 40G and 100G
In reviewing the thread, I have one quick comment regarding the market
urgency. I believe server interconnection for 40g is huge enough to
justify itself. And I heard broad end users stating that 100g is
desperately needed, so 100g has market urgency today.
While the dominant server-switch interconnect is 1G today, 40G is
projected to have its prime time in the 2012 time frame, which is 5 yrs
out. I f the 40g product isnot immediately needed, what is the
arithmetic behind to develop its standard well ahead of its prime time?
I think this is important point to clarify in the July presentations.
Also I heard 40g plans to piggyback on almost every aspect of the 100g
std and could be able to quickly catch up and make cost-effectively.
Could it be better to develop 40g later, which is more beneficial to
establish the viability of 40g products in the market? This would help
avoid the fragmentation of the R&D efforts for most of us, so the 40g
and 100g can be developed in serial instead of parallel.
From: Hays, Robert [mailto:
Sent: Wed 6/13/2007 12:21 PM
Subject: Re: [HSSG] 40G and 100G
Thanks for your note Chris. It helped to focus the discussion and next
steps in my mind.
For anyone who is not yet convinced that a 40G objective is worthy of an
IEEE 802.3 standardization effort... Would you please help us prepare
for the July Plenary by letting us know if you disagree with Chris's
assessment or have any additional items you believe need to be
addressed? My take-ways from his assessment are:
1. Broad Market Potential, Technical Feasibility, and Compatibility
have been sufficiently demonstrated for 40G
2. 40G and 100G Distinct Identities need to be clarified by careful
wording regarding target applications
3. More data is needed to demonstrate Economic Feasibility of 40G
Your feedback will be helpful in developing our presentations. Thank
From: Chris Cole [mailto:
Sent: Wednesday, June 13, 2007 10:27 AM
Subject: Re: [HSSG] 40G and 100G
I support the approach outlined by Mark for how to make progress at the
July meeting; specifically to work the details of completing the 5
Criteria for 100G and 40G.
With respect to 100G, all 5 Criteria have been met and out of the July
meeting the HSSG should forward a PAR for 100Gbps Ethernet as currently
We have again seen a challenge to the 100G Broad Market Potential; using
the recent argument that 100G does not have Broad Market Potential
without 40G. The arithmetic behind this argument is a mystery. End users
have stated that 100G is desperately needed today, when the dominant
server-switch interconnect is 1G. When 100G starts deployment, it has
further been stated that Nx100G LAG configurations will be immediately
required for second level switch-switch interconnect. Data centers are
only now looking to transition server-switch interconnect to 10G, with
the start of 10G high volume ramp projected for the 2009 time frame. End
users have stated that when they transition servers to 10G, they will
need 1T Ethernet to support the resulting traffic. On the other hand,
40G is projected to start high volume ramp in the 2012 time frame, so it
will be a number of years after 2012 that 40G server ports will approach
10G server ports. So any appreciable impact of 40G on 100G demand!
(outside of leading edge applications like HPC,) is 7 to 10 years out.
With respect to 40G, a separate set of 5 Criteria need to be drafted
that meet the requirements of a distinct project. Broad Market
Potential, Technical Feasibility and Compatibility have clearly been
met. The two criteria that need further work are Distinct Identity and
Economic Feasibility. There is no question about 40G having a Distinct
Identity; it is simply an issue of crafting the wording to differentiate
40G from 10G, and to differentiate 40G from 100G. The latter is
straightforward as the two address different applications (at least
The 40G Criteria that continues to have genuine disagreement is Economic
Feasibility. I am personally persuaded that 40G has meet this criteria,
versus for example 4x10G LAG alternative, thanks to Shimon's, Howard's
and Schelto's excellent presentations at the May meeting. There are
others that remain unconvinced as can be seen from Dan's recent email. I
suggest that the concern of these HSSG participants be taken as genuine
(not as a HSSG 40G derailing tactic,) and that through a combination of
additional supporting material and good crafting of the wording we
convince those that have a concern to support 40G Economic Feasibility.
I am confident that they will continue to have an open mind to
additional supporting material.
40G advocates should see this as an opportunity, not as a problem. A lot
of those who are skeptics are end users, and this gives the future 40G
suppliers a forum to start educating their future customers why 40G will
offer them benefits over 10G. If this argument can not be made
persuasively, there should be genuine concern at server and switch
companies (not just at the HSSG) about the viability of these products
in the market.
Given the few items that remain to be done, we should be able be able to
forward a distinct PAR for 40Gbps Ethernet out of the July meeting.
From: Mark Nowell (mnowell) [mailto:mnowell@CISCO.COM]
Sent: Tuesday, June 12, 2007 11:44 AM
Subject: Re: [HSSG] Soliciting Support for my Presentation
I have to strongly object to many of your comments and think they are
not helpful in the path forward.
I think some perspective is needed to substantiate these comments and
your recent attendance has perhaps robbed you of that perspective.
The HSSG was formed nearly a year ago now and has done considerable work
trying to define a project. It is true that the initial focus was on
the networking and aggregation applications and many rates and reaches
have been debated and discussed. During these numerous presentations,
debates, motions, and straw polls, the process of winnowing down the
proposals into a set of objectives and then 5 Criteria responses was
happening. It should be noted that, for this application a number of
rates were decided against including 40G, 80G, 120G and I think even
160G was mentioned.
Then, rather late in the process, compared to the progress already made
on the networking application, Shimon from Sun pointed out that there
was another market/application that wasn't being addressed - the server
market and this too needed a higher speed solution with 100G being too
Admittedly, there was considerable healthy debate about why alternative
technologies couldn't be used (such as LAG). It is unfortunate that
this application was not brought into HSSG clearly until later in the
process and you can't expect a major shift like this to happen without
debate and discussion so everyone is clear as to the problem statement.
I think this took until the Geneva meeting to happen properly. I don't
think I have ever heard a comment that the server application should not
The discussion about bringing two major new Ethernet rates out at the
same time, which has never been done before, is a valid one to have and
is really the background to many of the questions as market confusion is
something we all prefer to avoid if possible.
So rather than continuing an unconstructive mud-slinging activity, I
think we all need to look forward. There is a real need to develop a
networking solution @ 100GE. There is a real need to develop a server
interconnect solution @ 40GE. There is a real need to facilitate the
industry's understanding of these two applications. These two
applications, as defined, are complementary to one another.
I'm disappointed to see any suggestion of a 'divide and conquer
strategy' or 'strategy of kicking 40GE out of the HSSG and then nipping
it in the bud'. You are the first person I have ever heard suggest
that. I think a path forward can be found which is acceptable for both
applications and it was proposed a couple of times during Geneva. To
make it happen though, much offline work needs to happen between now and
July. So I urge everyone to put away their Robert Ludlum spy novels and
re-enter the relatively boring engineering world of standards and focus
on how to solve a problem.