| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Dan, While I share some of your frustration with the lack of
progress during this week’s HSSG meeting, and some of the individual
points in your presentation, I do not support it. I am in disagreement with your
Post-Debate Conclusions, and find their tone as not conducive to good
discussion of the best way to move forward within the HSSG. I am aware of the
passions generated during this week’s debate and understand why you wrote
your presentation, but wish that it had not been sent out. I am confident that
after your vacation travel in Good discussion of how to move forward is critically
dependant on acknowledging that 1) 100GE Broad Market Potential, and 2) 40GE Broad
Market Potential have been established well above the threshold for 802.3 Five Criteria.
Continued debate of this will only lead to delay in addressing the substantive
issue of what is the best way to move forward in developing 100GE and 40GE
standards. A possible framework for this discussion is outlined in “HSSG
Next Steps Proposal” presentation that HSSG participants authorized as a post-deadline
meeting submission. http://www.ieee802.org/3/hssg/public/may07/cole_03_0507.pdf An insight that has come out of the HSSG discussion of 100GE
and 40GE rates during the past several months is that fundamental development
cycles for new network equipment architectures and new server architectures appear
to be different. A new data switch architecture development requires massive investment,
which leads network equipment developers to want large jumps in data rate, like
factors of 10x, to allow a return on that investment. An intermediate data rate
causes an increase in the overall development investment, and a shortening of
the useful life of network equipment which reduces the return on that investment. Economics of server development appear to be different and
more favorable to shorter development cycles, i.e. more frequent architecture changes.
This drives the need for more granular jumps in data rate, like 4x. Ethernet
has not done this historically, but that may be because protocols other then
Ethernet were used to bridge the gap. Going forward, more granular Ethernet
data rate steps may become the norm, rather then just a one time anomaly. This
difference in development economics is also consistent with how often some end
users replace servers versus networking equipment. It suggests that moving forward,
network data rates may go from 10G to 100G to 1T, while server data rates go
from 10G to 40G to 100G to 400G to 1T. This difference in data rate needs is
not necessarily bad for either industry. More frequent server replacement may
extend the useful life of network equipment (good for network equipment
developers), and large jumps in network data rates assures availability of
aggregation capacity to support multiple server cycles (good for server
equipment developers.) As we discuss how to best move forward with developing 100GE
and 40GE standards, any approach needs to have the following two
characteristics; 1) permits network equipment developers to have a single 100GE
architecture, i.e. does not force them into developing a dual rate 100GE/40GE
architecture, and 2) gives server equipment developers a 40GE server data rate.
It may also need the recognition that some data rates are optimized for server
interconnect and not intended for networking. I look forward to a constructive discussion on how best to move
both standards forward. Chris Subject:
[HSSG] Soliciting Support for my Presentation Date:
Thu, 31 May 2007 05:25:12 -0000
|