|Thread Links||Date Links|
|Thread Prev||Thread Next||Thread Index||Date Prev||Date Next||Date Index|
Some of the discussion of 100Gb protocols has been in the context of a 10 x 10Gb physical layer. It is early in the standards process to reach this conclusion, and our discussions would benefit by not making the assumption of 10 physical layer channels.
Part of our objective should be to define standards that will operate over the existing infrastructure (for example existing deployment distances: 10km, 40km, 80km,) and be amenable to low cost, high volume manufacturing. The 10GBASE-LR, 10km standard is a good example, as it is the most widely used 10G Ethernet SMF interface. It takes advantage of the low fiber dispersion at 1310nm to use low cost DFBs. A 10 channel CWDM 100Gb implementation would require a 200nm window, assuming today’s 20nm minimum wavelength spacing limit for un-cooled optics. This would cause dispersion of the shorter wavelength channels to approach 1550nm values. This can be compared to one of the previously listed alternate approaches; 5 x 20Gb. Since the dispersion effects at 20Gb are four times those at 10Gb, either approach will require a unique set of technical solutions to deal with this problem. It is not apparent today, which would result in lower implementation costs five years from now.
From: David Martin
Some history on the 10GigE project…
Re: ONE RATE AND FORMAT - if we had recognized early enough that WAN was an important application and tried to unify things to the greatest extent possible, we might have elected…
There were lengthy discussions on the applications for 10GigE. Both LAN and WAN were recognized as important. But the 2 had different fundamental rate requirements, leading to the values we have today. For example, a common test for an 802.1 bridge supporting the next generation Ethernet line rate on its aggregated side is to fully fill 10 interfaces at the current Ethernet line rate on the lower speed side and observe that no frames are lost through the higher rate interface. That was one practical, backwards compatibility-based argument for LAN = 10.000G. And you’re well aware of the WAN rate requirement.
The committee’s position was that a standard 802.1 bridge would suffice to convert between the 2 interface rates.
Remember that the 802 committee is the “LAN/MAN Standards Committee”, not WAN. It is a testament to the participants of P802.3ae that they were open-minded enough to consider the WAN PHY proposal.
Re: REPEATER DEFINITION - If we were to define a repeater that regenerates preamble & IPG and have a requirement that anything that works in a point-to-point configuration must also work across a repeater, we could discourage some of this proprietary stuff…
Since repeaters were not included within the scope of P802.3ae, such work was launched through T1X1.5 and then migrated to SG15, eventually becoming G.7041 GFP with related pieces in G.806, which of course you’re well familiar with.
So the standards world did do its job.
The different rates are not the issue. It’s the proprietary variants of the 10GigE LAN that are the problem. We could have had a similar problem now with the WAN PHY if a vendor had made proprietary use of its TOH. Then instead of directly connecting a 10GigE WAN PHY signal into an OC-192 regen (i.e., its original intent), the WAN PHY signal would have to be somehow entirely mapped into an OC-192, which of course wouldn’t fit and we might have over-clocked OC-192s today instead.
So how does a standards body crystal ball and preclude an equipment vendor from breaking a standard with proprietary changes? That’s the issue. And the answer is obvious.
David W. Martin
Stephen J (Steve) [mailto:sjtrowbridge@xxxxxxxxxx]
With all due respect, at 10G there were two rates and frame formats defined:
- The 10G LAN PHY, which some take to mean 10G BASE-R at 10.325 Gbit/s including the 64B/66B coding
- The 10G WAN PHY, which is built inside of a SONET/SDH wrapper at 9.58464 Gbit/s including 64B/66B coding.
As much as we wish that customers would all choose the LAN interface when they want to connect to a LAN and the WAN interface when they want to connect to a WAN, this hasn't happened. We see numerous cases for a variety of different reasons where people ask to take the 10G LAN PHY and carry it over a long distance.
There are some straightforward kinds of approaches to do this if you start from the assumption that Ethernet is really a packet interface and all you need to do is preserve the packets across the transport network. For example, you can carry your 10G LAN PHY by mapping it into an ODU2, decoding 64B/66B, dropping the preamble and IPG and replacing with GFP framing (saving 8-12 octets per packet) and using the OTN scrambler and get the full packet throughput of a 10G LAN PHY over a 10G transport network interface. At the far end, you reverse the process and regenerate a 10G LAN PHY signal with the same packet information content. But this breaks any proprietary stuff that people are doing with the preamble and IPG and seems to have an "image problem" with some operators that you are fiddling too much with the customer's stuff (even though most of us understand that this approach doesn't touch the payload).
Let me express the two suggestions that I am making for how we approach future interface definitions in terms of what it would have looked like had we done this at 10G:
- ONE RATE AND FORMAT - if we had recognized early enough that WAN was an important application and tried to unify things to the greatest extent possible, we might have elected that the XGMII being freshly defined wasn't exactly 10.00000000000 Gbit/s, but instead, say 9.282944 Gbit/s (if we decided that SONET/SDH was the dominant transport mode to be targeted) or 9.721329 Gbit/s (if we decided that OTN was the dominant transport mode to be targeted). Then you could have taken exactly the same 64B/66B coded bitstream that you used for the LAN and filled it directly into transport frames, and poof you are done. You wouldn't have had any of the market issues we see today.
- REPEATER DEFINITION - If we were to define a repeater that regenerates preamble & IPG and have a requirement that anything that works in a point-to-point configuration must also work across a repeater, we could discourage some of this proprietary stuff. If we then define that any layer 1 server layer network is supposed to meet the requirements for a repeater (appropriately defined), then you have the flexibility to use things like GFP framing over a transport network to get the extra throughput you need to fully carry all of the packets of a 10G LAN PHY Ethernet signal over a 10G transport network interface.