|Thread Links||Date Links|
|Thread Prev||Thread Next||Thread Index||Date Prev||Date Next||Date Index|
Below is an email that Menachem sent out earlier today. I am re-sending it for him, as the reflector has undergone considerable growth today! J.
We will have a lot of ground to cover, so I encourage all to use the reflector and let’s start having the conversations that we will need to have.
From: Menachem Abraham
John encouraged me to restart the discussion thread (ref: my old e-mail titled “Question” sent to the previous hssg reflector we used) regarding topics we may want to put on the agenda for the next meeting(s).
Please add to this.
Here it is with one addition at the end:
(a) Technical feasibility of a 100 Gb/s (or whatever speed we choose) Medium Access Control function is a topic that justifies more discussion than we had so far. I hope we can see some presentations that will go further than just saying “higher clock speeds and wider data path” make this feasible. Should we look at the feasibility of implementing the CRC function, for example, at this speed?
(b) Technical feasibility/aspects of Physical Layer Aggregation. Perhaps there are no issues here and for some this is trivial. However for many people in the study group and 802.3 WG this isn’t “intuitively obvious”. A minimum size packet at 100Gb/s lasts about 5 ns. That is about the same as the propagation time through 1 meter of cable. If we look at a 1 Km link we could have 1000 packets in transit (or 10,000 packets on a 10Km link and many more on long haul links coast to coast). What is the skew we would allow between lanes? How do we reconstruct the packet which traveled 10 Km on 10 lanes? Again, I am sure there are a few among us who have all the answers but we need to educate the entire group so eventually we can vote yes on technical feasibility.
(c) Compatibility. Are there higher layer implications to the fact that on a long distance 100 Gb/s link we will have 10 x the number of packets in transit (e.g. retransmission windows)?
(d) Economic Feasibility. There seems to be a lot of support to reuse the 10Gb/s PMDs. That may indeed end up as a good choice but before we make such a choice, would it be worthwhile to discuss the merits of 1500 nm vs. 1300 nm optics as this relates to packaging 10 x 10 Gb/s xcvrs into one physical device? Perhaps there are fundamental economic feasibility differences between the two (I am not saying that there is or there isn’t)?
(e) On the topic of distinct identity. The fundamental assumption for this study has been that Equal Cost Routing and 802.3 Link Aggregation are not good solutions for point to point routes that require greater than 10 Gb/s. Since this is at the basis of justifying our study group I believe it would be beneficial to listen to some presentations (tutorial in nature) on these two approaches (what is it, how it works in practical terms when a network manager sets this up and more details about why these techniques are not as good as we would want them to be).
(f) Another technical feasibility and economic feasibility related topic is Forward Error Correction. It would be great to hear about tradeoffs in implementing FEC on the individual lanes (NxM FEC implementations) versus at the new high speed (e.g. at 100 Gb/s)