|Thread Links||Date Links|
|Thread Prev||Thread Next||Thread Index||Date Prev||Date Next||Date Index|
Many of your questions are answered in messages just exchanged between Geoff Thompson and me. To address some of the others:
A multi-hop solution without realignment/reassembly sounds difficult, if not impossible:
1) How would intermediate nodes read the destination MAC address?
2) The inter-channel skew would be effectively unbounded.
The question of what channel rates higher than 10G to use is related to the question of mixing and matching lanes of differing speeds. For starters, I think we should apply some judgment on the latter issue – i.e., would anyone really want to mix 10G and 1G lanes to build a really fat pipe? If we’re talking about a pipe of 20-100 G or more, do additional 1G increments really add any value? Regarding lane speeds higher than 10G, the question becomes, “Can we define aggregation of lanes whose speeds are as yet unknown, or do we have to make a decision today on speeds (and hence, technologies) that don’t exist yet?” Do we use 25G, 33.3G, 40G, 50G? This future-gazing is a tricky business.
Stephen J (Steve) [mailto:sjtrowbridge@xxxxxxxxxx]
In some respects, I think that (A) and (B) in your description might be good candidates to tackle in separate projects (PARs).
Let me comment on proposal (B) first:
I would characterize this as a study of "Multiple Lane Approaches". Here, there are a number of questions to be answered:
1) Is the number of "lanes" fixed or variable?
2) Can the interface operate at reduced bandwidth in the presence of failure of individual lanes?
3) What is the network architecture governing the multi-lane interconnection? Are multiple lanes carried over a single span only, or can the lanes be carried independently across a network and only reaggregated at the endpoints? (attractive feature here: not necessary to build the new interface on every network element along a path between ultra-high rate endpoints to provide an ultra-high rate service. Network elements transporting individual lanes can be blissfully unaware that that lane is part of a larger aggregate). This network architecture has an impact on the differential delay that may need to be accommodated across the lanes when they are re-aggregated.
4) Are the individual lanes carried over existing or new physical interfaces?
5) Can we learn from or reuse the capabilities from ITU-T Virtual Concatenation and LCAS to provide this kind of physical layer aggregation?
Back to proposal (A), I think that even those of us who think that (B) is an important problem to solve do not believe that 10G is the highest rate we will ever have for serial transport. So the way I would like to see (A) approached is to study what the next potential serial interface rate above 10G should be and what its characteristics are. Here, we could study 100G, 40G or other rates that provide a suitable evolution and are supported by the technology.
40G has some attractive characteristics since it could reuse components from transport interfaces at the same rate (SONET OC-768, SDH STM-256, or OTN OTU-3). Note that in one sense, we already have WAN PHY type interfaces at 40G if you consider Ethernet frames mapped via GFP-F into SONET OC-768c, SDH VC4-256c, or OTN OPU3.
Of course even 40G is not the highest rate interface that we will ever build, so even if 40G is considered to be an evolutionary step along the path of increasing bandwidth, it is worth studying what the next serial rate should be above 40G.