400G ER4 wavelengths - two grids
Dear colleagues I have come across a weird mishmash of offers for QSFP56-DD 400G ER4 and ER4 lite transceivers in the market, and now I'm looking to compare notes. It seems to me that there were two separate development trains going on at once. Resulting in two incompatible grids of four wavelengths being used (and sold, see below): - 1295.56 nm 1300.05 nm 1304.58 nm 1309.14 nm - 1304.58 nm 1306.85 nm 1309.14 nm 1311.43 nm Why you would go tighter seems obvious to me: To stay close to the zero point of dispersion with all four lanes, and perhaps to make all four experience more similar levels of dispersion, to make dispersion compensation in the receiver easier. It seems that the wider spaced set is based on IEEE802.3 clause 122 (added in IEEE802.3bs), more specifically in Table 122–6 "200GBASE-LR4 and 200GBASE-ER4 wavelength-division-multiplexed lane assignments". (Note that it doesn't officially specify 400GBASE-ER4, this was apparently added in the 50G-per-lambda-time.) The tighter spaced seems to be based on the 100G Lambda MSA specification called "400G-ER4-30 Technical Specification 1.0" specifically Table 2-1 "Wavelength-division-multiplexed lane assignments" [0] FS.com and Fibermall sell transceivers with the wide spacing [1][2] Flexoptix sells transceivers with the narrow spacing [3] According to a contact at one of the shops the wide spacing is also called LWDM and the tight one is called nLWDM. Both now exist and are out there, so caveat emptor, I guess... Does anyone have any parts sourced directly from a router manufacturer? Could you share what they went with optically? Or any other pertinent insider knowledge? :-) Best regards, Joel [0] https://100glambda.com/specifications [1] https://resource.fs.com/mall/resource/qdd-er4l-400g-data-sheet.pdf [2] https://www.fibermall.com/sale-461143-qsfp-dd-400g-er4-lite-lwdm4-30km.htm [3] https://www.flexoptix.net/en/d-164hg-30-e.html?option875=1 -- Joel Busch, Network Engineer Switch, AS559
On 26/11/2025 15:40, Joel Busch via NANOG wrote:
Does anyone have any parts sourced directly from a router manufacturer? Could you share what they went with optically?
I have not seen any OEM's selling the 400G-ER4 or 400G-ER4 Lite in the wild. Most of the ones I have come across are from the open market. What OEM's seem to be selling are 400G-DR4/FR4/LR4.
Or any other pertinent insider knowledge? :-)
The bottom line is that the ER4-Lite was built for shorter distances (30km), while the ER4 proper is designed for up to 40km. Of course YMMV re: fibre quality, so the better numbers to work with are up to 15.5dB for the ER4-Lite and up to 18.5dB for the ER4 proper. The ER4-Lite is cheaper, and also explains why it has been out of stock for some time now. I see that suppliers have resumed providing units in the last 2 or so months. It would not be advisable to run an ER4 and ER4-Lite over a link. Better to match them for optical compatibility. Of course, it goes without saying that if you need to go father than 40km for 400G, or solve for poor fibre OSNR, you are better off going coherent. Hope this helps. Mark.
Hi Mark On 26.11.2025 17:42, Mark Tinka wrote:
I have not seen any OEM's selling the 400G-ER4 or 400G-ER4 Lite in the wild. Most of the ones I have come across are from the open market.
Hm not just the ones I checked then. I wonder why they aren't interested. Maybe not a big market, or they are waiting for IEEE standardization?
The bottom line is that the ER4-Lite was built for shorter distances (30km), while the ER4 proper is designed for up to 40km. Of course YMMV re: fibre quality, so the better numbers to work with are up to 15.5dB for the ER4-Lite and up to 18.5dB for the ER4 proper.
Yes I realize. Admittedly I do use the distances as shorthand though.
The ER4-Lite is cheaper, and also explains why it has been out of stock for some time now. I see that suppliers have resumed providing units in the last 2 or so months.
Interesting. I hadn't noticed this, probably because we only have three ER4 so far and no need for more until we get more routers :-)
It would not be advisable to run an ER4 and ER4-Lite over a link. Better to match them for optical compatibility.
I'm not sure if you caught my main issue: It's not that ER4 lite uses a different grid from ER4, it's that ER4 lite from different manufacturers use different grids! Flexoptix ER4 lite: 1304.6 nm, 1306.8 nm, 1309.1 nm, 1311.4 nm FS.com ER4 lite: 1295.56 nm, 1300.05 nm, 1304.58 nm, 1309.14 nm FiberMall ER4 lite: 1295.56 nm, 1300.05 nm, 1304.58 nm, 1309.14 nm One grid based on the 100G lambda MSA the other on extrapolation from IEEE802.3. For products sharing the same marketing name! Same goes for the ER4 too.
Of course, it goes without saying that if you need to go father than 40km for 400G, or solve for poor fibre OSNR, you are better off going coherent.
Of course, we do that transalpine for example.
Hope this helps.
Mark.
Thank you, always good to hear your views on optics. I was thinking you might respond :-) Joel -- Joel Busch, Network Engineer Switch, AS559
On 26/11/2025 19:31, Joel Busch wrote:
Hm not just the ones I checked then. I wonder why they aren't interested. Maybe not a big market, or they are waiting for IEEE standardization?
I've actually never asked the OEM's. My 1+1 suspicion would be that trying to drive 400G-ER4 PAM4 beyond 40km means they would need to get involved in optical planning for the customers as well. Moreover, the ER4 and ER4-Lite plugs are not so differently priced from the 400G-ZR+ coherent ones (within about US$1,000 - US$1,500 of each other). Since all the OEM's have some kind of 400G-ZR+ solution, I could see why they would be reluctant to support 400G-ER4 and ER4-Lite.
Interesting. I hadn't noticed this, probably because we only have three ER4 so far and no need for more until we get more routers :-)
It was between June - August, this year. The issue was due to an SOA shortage. Nearly every manufacturer was affected by this, on both the ER4 and ER4-Lite. Some manufacturers have EoL'ed the ER4-Lite builds, and are only focusing on ER4. So if you have access to ER4-Lite supply, you should take it all while you still can :-).
I'm not sure if you caught my main issue: It's not that ER4 lite uses a different grid from ER4, it's that ER4 lite from different manufacturers use different grids!
Flexoptix ER4 lite: 1304.6Â nm, 1306.8Â nm, 1309.1Â nm, 1311.4Â nm FS.com ER4 lite:Â Â Â 1295.56 nm, 1300.05 nm, 1304.58 nm, 1309.14 nm FiberMall ER4 lite: 1295.56 nm, 1300.05 nm, 1304.58 nm, 1309.14 nm
That is odd. Looks like Flex are the odd man out, here. The ones we use for our own network and sell via our hardware business are from Smartoptics. They are built similar to FS and FiberMall using 1296/1300/1305/1309nm wavelengths. For the ER4 ones we use and sell from Smartoptics also, they are running 1304/1306/1309/1311nm, which matches what you say Flex are using for ER4-Lite. I suspect the Flex numbers could be a simple documentation error where someone copied & pasted the ER4 numbers to the ER4-Lite page, but I could be wrong. I would write them and ask. Mark.
On 26.11.2025 19:22, Mark Tinka wrote:
It was between June - August, this year. The issue was due to an SOA shortage. Nearly every manufacturer was affected by this, on both the ER4 and ER4-Lite.
Some manufacturers have EoL'ed the ER4-Lite builds, and are only focusing on ER4. So if you have access to ER4-Lite supply, you should take it all while you still can :-).
Interesting and good to know, thank you!
The ones we use for our own network and sell via our hardware business are from Smartoptics. They are built similar to FS and FiberMall using 1296/1300/1305/1309nm wavelengths. For the ER4 ones we use and sell from Smartoptics also, they are running 1304/1306/1309/1311nm, which matches what you say Flex are using for ER4-Lite.
Oh that's actually the first I've heard of a manufacturer using different grids for their ER4 lite vs their ER4! Now I get why you were telling me not to combine ER4 lite and ER4, in your experience they use nLWDM and LWDM respectively. My impression was the opposite, that each seller sticks to one grid for both. Examples below: FiberMall list LWDM for both: - https://www.fibermall.com/sale-461143-qsfp-dd-400g-er4-lite-lwdm4-30km.htm - https://www.fibermall.com/sale-458209-qsfp-dd-400g-er4-lwdm4-40km.htm Flexoptix list nLWDM for both: - https://www.flexoptix.net/en/d-164hg-30-e.html - https://www.flexoptix.net/en/d-164hg-40-e.html EdgeOptic list nLWDM for both: - https://edgeoptic.com/products/400g-qsfp-dd/400g-qsfp-dd-41/ - https://edgeoptic.com/products/400g-qsfp-dd/400g-qsfp-dd-42/ FS only have the ER4 lite, Huber+Suhner only have the ER4, GBC Photonics only have the ER4, so they don't provide a useful datapoint here. I found another example for your case too: Fiberworks list LWDM for ER4 Lite and nLWDM for ER4: https://www.fiberworks.no/fiberworks-as/qdd-400g-er4l/qsfp-dd-400gbase-er4l-... https://www.fiberworks.no/fiberworks-as/qdd-400g-er4/qsfp-dd-400gbase-er4-4x... I don't have the insight into who actually manufactures and who sells who's stuff, so these examples might contain duplicate entries for what any one manufacturer does.
I suspect the Flex numbers could be a simple documentation error where someone copied & pasted the ER4 numbers to the ER4-Lite page, but I could be wrong. I would write them and ask.
I was talking to them before starting this thread and they confirmed that the nLWDM spacing is correct. They also let me know that they tested their ER4 lite against their ER4 and they were compatible. So unfortunately it's not so simple. We have a mess in the market, I think. Joel -- Joel Busch, Network Engineer Switch, AS559
On 27/11/2025 11:32, Joel Busch wrote:
Oh that's actually the first I've heard of a manufacturer using different grids for their ER4 lite vs their ER4! Now I get why you were telling me not to combine ER4 lite and ER4, in your experience they use nLWDM and LWDM respectively.
Not quite. Both ER4 and ER4-Lite are nLWDM, because of the chromatic dispersion sensitivity when using PAM4 modulation above 10km.
My impression was the opposite, that each seller sticks to one grid for both. Examples below:
So Smartoptics use different wavelengths for ER4 and ER4-Lite: Â Â https://smartoptics.com/product/tqd031-sn4c-so/ https://smartoptics.com/wp-content/uploads/2023/10/ds-tqd023-sl4c-so-qsfpdd-... And they are using nLWDM for both. The LWDM is primarily used for 100G-LR4, as there is no concern of chromatic dispersion below 10km.
I don't have the insight into who actually manufactures and who sells who's stuff, so these examples might contain duplicate entries for what any one manufacturer does.
It's quite possible. On our end, we've generally gone to 400G-ZR+ coherent for 400G requirements, even in the metro. So we've only had to deal with a handful of 400G-ER4/ER4-Lite plugs, and only from Smartoptics. We haven't tried to mix vendors, not for any reason other than there was no benefit in doing so.
I was talking to them before starting this thread and they confirmed that the nLWDM spacing is correct. They also let me know that they tested their ER4 lite against their ER4 and they were compatible. So unfortunately it's not so simple. We have a mess in the market, I think.
So do they explain why they are using the same sets of wavelengths for what they market as ER4 and ER4-Lite? And is there an appreciable difference in price between both types of plugs? Looking at Flex's web site, the ER4-Lite page says "not interoperable with D.164HG.30", which is odd because D.164HG.30 is Flex's ER4-Lite: Â Â https://www.flexoptix.net/en/d-164hg-30-e.html?option875=1 I think there is a documentation issue with Flex, but again, I could be wrong :-). Mark.
On 27.11.2025 10:49, Mark Tinka wrote:
On 27/11/2025 11:32, Joel Busch wrote:
Oh that's actually the first I've heard of a manufacturer using different grids for their ER4 lite vs their ER4! Now I get why you were telling me not to combine ER4 lite and ER4, in your experience they use nLWDM and LWDM respectively.
Not quite.
Both ER4 and ER4-Lite are nLWDM, because of the chromatic dispersion sensitivity when using PAM4 modulation above 10km.
My impression was the opposite, that each seller sticks to one grid for both. Examples below:
So Smartoptics use different wavelengths for ER4 and ER4-Lite:
https://smartoptics.com/product/tqd031-sn4c-so/ https://smartoptics.com/wp-content/uploads/2023/10/ds-tqd023-sl4c-so- qsfpdd-400g-er4-lite-c4-r6.1.pdf
And they are using nLWDM for both.
The LWDM is primarily used for 100G-LR4, as there is no concern of chromatic dispersion below 10km.
If I'm getting you right, you are saying there are three grids, two of which are types of nLWDM? I think you're mixing something up. Especially since 100G-LR4 already uses what Smartoptics uses for 400G-ER4-lite. In my view we have the following three common grids in the O-band: CWDM4: 1271 nm 1291 nm 1311 nm 1331 nm ( 20 nm spaced) LWDM: 1295.56 nm 1300.05 nm 1304.58 nm 1309.14 nm (800 GHz spaced) nLWDM: 1304.58 nm 1306.85 nm 1309.14 nm 1311.43 nm (400 GHz spaced) 100G-CWDM4 plugs existed but faded relatively soon in favour of 100G-LR4 plugs. 100G-LR4 plugs already use the LWDM wavelengths. See IEEE802.3 Table 88–7 "100GBASE-LR4 and 100GBASE-ER4 transmit characteristics" and IEEE802.3 Table 88–5 "Wavelength-division-multiplexed lane assignments": Center Range 1295.56 nm 1294.53 to 1296.59 nm 1300.05 nm 1299.02 to 1301.09 nm 1304.58 nm 1303.54 to 1305.63 nm 1309.14 nm 1308.09 to 1310.19 nm This is also true of the 100G-LR4 from Smartoptics: https://smartoptics.com/wp-content/uploads/2022/12/so-qsfp28-lr4-10l-r5.3.pd... If you compare that to the 400G-ER4-Lite from Smartoptics you linked you'll see the same wavelengths: https://smartoptics.com/wp-content/uploads/2023/10/ds-tqd023-sl4c-so-qsfpdd-... Later the 400G-FR4 and 400G-LR4 went back to the CWDM4 wavelengths. So 400G-LR4 ended up using wider spacing than 100G-LR4. A bit counter-intuitive, but true: https://smartoptics.com/wp-content/uploads/2022/12/so-qsfp-dd-4c-lr4-4-r5.6.... My guess is it was hard enough to handle 10 km chromatic dispersion - without FEC, remember - that people already moved to tighter spacings for 100G-LR4. But then they could relax it in the 400G era because now host KP4 FEC is mandatory. I think the LWDM spacing of 100G-LR4 is also what makes the existence of so called "LR4 ext" or "20km LR4" possible as a by-product.
So do they explain why they are using the same sets of wavelengths for what they market as ER4 and ER4-Lite?
No, because I didn't ask, because I then thought it was normal :-)
And is there an appreciable difference in price between both types of plugs?
Yeah around 30%, the lite makes do with an APD receiver and no SOA, according to the spec sheet. Maybe that's where the savings come from. But I didn't ask about that either. This is very similar to the specified difference between their 100G-ER4-lite and 100G-ER4, so I didn't question it much.
Looking at Flex's web site, the ER4-Lite page says "not interoperable with D.164HG.30", which is odd because D.164HG.30 is Flex's ER4-Lite:
  https://www.flexoptix.net/en/d-164hg-30-e.html?option875=1
I think there is a documentation issue with Flex, but again, I could be wrong :-).
Ah sorry I could have mentioned this ahead of time, since I ran into it too. The D.164HG.30 is actually an older version of the ER4 lite. They said it was cancelled, they can't sell it anymore. D.164HG.30.E and D.164HG.40.E are the current ones I was referring to. That's also what you and I both linked to. There is no shop page for the D.164HG.30 anymore. Joel -- Joel Busch, Network Engineer Switch, AS559
On 27/11/2025 15:29, Joel Busch wrote:
100G-CWDM4 plugs existed but faded relatively soon in favour of 100G-LR4 plugs.
You can still buy 100G-CWDM4 today. It is the 2km option for 4-lane 100G. On the single-lambda side, the 2km plug has been defined under 100G-FR1. While I like those the most because data centre x-connects will generally be under 2km long, experience is showing that operators are carrying their "LR4" habit into 100G-LR1 world :-). 100G-LR1 plugs are about 12% more expensive than 100G-FR1, on average, but it seems old habits die hard :-).
100G-LR4 plugs already use the LWDM wavelengths. See IEEE802.3 Table 88–7 "100GBASE-LR4 and 100GBASE-ER4 transmit characteristics" and IEEE802.3 Table 88–5 "Wavelength-division-multiplexed lane assignments":
Center             Range
1295.56 nm   1294.53 to 1296.59 nm 1300.05 nm   1299.02 to 1301.09 nm 1304.58 nm   1303.54 to 1305.63 nm 1309.14 nm   1308.09 to 1310.19 nm
This is also true of the 100G-LR4 from Smartoptics: https://smartoptics.com/wp-content/uploads/2022/12/so-qsfp28-lr4-10l-r5.3.pd...
If you compare that to the 400G-ER4-Lite from Smartoptics you linked you'll see the same wavelengths: https://smartoptics.com/wp-content/uploads/2023/10/ds-tqd023-sl4c-so-qsfpdd-...
Later the 400G-FR4 and 400G-LR4 went back to the CWDM4 wavelengths. So 400G-LR4 ended up using wider spacing than 100G-LR4. A bit counter-intuitive, but true: https://smartoptics.com/wp-content/uploads/2022/12/so-qsfp-dd-4c-lr4-4-r5.6....
My guess is it was hard enough to handle 10 km chromatic dispersion - without FEC, remember - that people already moved to tighter spacings for 100G-LR4. But then they could relax it in the 400G era because now host KP4 FEC is mandatory.
I think the LWDM spacing of 100G-LR4 is also what makes the existence of so called "LR4 ext" or "20km LR4" possible as a by-product.
Yes, you are right. It can be quite cobbled up. But to summarize it, there are 3 applications: * 100G-LR4 (LWDM): o 4×25G NRZ in O-band with the LAN-WDM 800GHz spacing, 10km class. o Wavelengths: 1295.56/1300.05/1304.58/1309.14nm. * 400G-LR4 (CWDM O-band): o 4×100G PAM4 on CWDM4 wavelengths (wide 20nm spacing), 6km – 10km class. o Wavelengths: 1271/1291/1311/1331nm. * 400G-ER4 (nLWDM): o 4×100G PAM4 with tighter 400GHz spacing clustered near 1310nm to keep dispersion low/flat, 30km - 40km class. o Wavelengths: 1304.58/1306.85/1309.14/1311.43nm
Yeah around 30%, the lite makes do with an APD receiver and no SOA, according to the spec sheet. Maybe that's where the savings come from. But I didn't ask about that either.
That is correct. APD's are cheaper but are more susceptible to noise, especially at higher gains.
Ah sorry I could have mentioned this ahead of time, since I ran into it too. The D.164HG.30 is actually an older version of the ER4 lite.
They said it was cancelled, they can't sell it anymore. D.164HG.30.E and D.164HG.40.E are the current ones I was referring to. That's also what you and I both linked to. There is no shop page for the D.164HG.30 anymore.
Thank you :-). Mark.
On 28/11/2025 04:40:40, "Mark Tinka via NANOG" <nanog@lists.nanog.org> wrote:
While I like those the most because data centre x-connects will generally be under 2km long, experience is showing that operators are carrying their "LR4" habit into 100G-LR1 world :-).
Distance is a bogus metric, most of the time it's about the loss. See shrinking link budget for 10km as you go from 1 to 10 to 100G.
100G-LR1 plugs are about 12% more expensive than 100G-FR1, on average, but it seems old habits die hard :-).
Having an extra dB or two when equinix are messing in the MMR is worth it. Also one less type to carry spares for. brandon
On 28/11/2025 10:44, Brandon Butterworth wrote:
Distance is a bogus metric, most of the time it's about the loss.
Outside the data centre, distance is a useless metric. Inside the data centre, it's generally reliable (unless the data centre are not telling you everything).
See shrinking link budget for 10km as you go from 1 to 10 to 100G.
It's no secret that OSNR budgets become more problematic as you increase capacity and reach. That's the nature of the beast. But like I said, most intra-DC cabling won't exceed 2km, and this isn't fibre that is being re-spliced every 6 months due to a cut. Of course, if you are dealing with a large data centre campus with different buildings, that is a whole other conversation.
Having an extra dB or two when equinix are messing in the MMR is worth it.
I find it more useful to clean up bad x-connects, because even if you can deal with a bad connection by having more optical power, fibre like that generally tends to degrade over time and get worse. Eventually, you will lose that signal, or worse, raise BER.
 Also one less type to carry spares for.
Whether you choose FR1 or LR1 really comes down to what others are doing. If you are going to interconnect to other operators, your decision may be driven by what most of them choose. And from what I'm currently seeing, most people appear to be choosing LR1 for intra-DC cabling. Like we did with multi-mode, you could lower your costs by using DR1 or FR1 for intra-rack cabling. That said, optically, FR1 will gladly talk to LR1... you'll just need to attenuate on your FR1 Rx, if you have a very short fibre span. We have tested a 100G-LR1 customer coming into our 400G-FR4 aggregation, without attenuation on our end. No drama, works fine. Mark.
On 28.11.2025 05:40, Mark Tinka wrote:
On the single-lambda side, the 2km plug has been defined under 100G-FR1. While I like those the most because data centre x-connects will generally be under 2km long, experience is showing that operators are carrying their "LR4" habit into 100G-LR1 world :-).
Yes we're seeing the same. And when I ask a customer if they would like their uplinks to be 100G-LR4 or 100G-LR they ask me what 100G-LR is :-)
Yes, you are right. It can be quite cobbled up. But to summarize it, there are 3 applications:
* 100G-LR4 (LWDM): o 4×25G NRZ in O-band with the LAN-WDM 800GHz spacing, 10km class. o Wavelengths: 1295.56/1300.05/1304.58/1309.14nm.
* 400G-LR4 (CWDM O-band): o 4×100G PAM4 on CWDM4 wavelengths (wide 20nm spacing), 6km – 10km class. o Wavelengths: 1271/1291/1311/1331nm.
* 400G-ER4 (nLWDM): o 4×100G PAM4 with tighter 400GHz spacing clustered near 1310nm to keep dispersion low/flat, 30km - 40km class. o Wavelengths: 1304.58/1306.85/1309.14/1311.43nm
Good summary. Just don't forget to check if the 400G-ER4 you're buying really uses the 400GHz spacing, everyone. Remember that FiberMall listing from my earlier email: https://www.fibermall.com/sale-458209-qsfp-dd-400g-er4-lwdm4-40km.htm And with ER4-lite all bets are off. It's anyone's guess which shop will sell which spacing, just make sure your inventory stays consistent for your own sanity. Thanks for bouncing back and forth with me on this topic, Mark! Good weekend to everyone, sooner or later, depending on timezone :-) Joel -- Joel Busch, Network Engineer Switch, AS559
On 28/11/2025 11:34, Joel Busch wrote:
Yes we're seeing the same. And when I ask a customer if they would like their uplinks to be 100G-LR4 or 100G-LR they ask me what 100G-LR is :-)
Wait when SFP-DD starts to become a thing :-). We have started shipping these to some customers in Africa. Great for density when you can slot 44x 100G ports into a 1U line card. Single-lambda 100G is definitely the way to go. LR4 just eats up too much space and pays the data centres a lot more in x-connects than I'd like.
And with ER4-lite all bets are off. It's anyone's guess which shop will sell which spacing, just make sure your inventory stays consistent for your own sanity.
I think the commercial gap between 400G-ER4/ER4-Lite is closing quickly enough that for most people green-fielding short-haul 400G above 10km, 400G-ZR+ coherent will look a lot more attractive. Especially because you can bolt on an amplifier and extend that 400G beyond 60km, or reduce the modulation and increase reach for slightly lower speeds (100G, 200G or 300G). 400G-ER4/ER4-Lite being O-band does not lend itself to being amplified. My advice to anyone thinking of 400G-ER4/ER4-Lite would be to consider 400G-ZR+ instead. 800G-ZR+ will help to lower 400G-ZR+ pricing even further, by the end of 2026. My prediction is that 400G-ER4/ER4-Lite fades into the darkness like 100G-CWDM4 did. Mark.
On 11/28/25 08:36, Mark Tinka via NANOG wrote:
Yes we're seeing the same. And when I ask a customer if they would like their uplinks to be 100G-LR4 or 100G-LR they ask me what 100G-LR is 🙂 Wait when SFP-DD starts to become a thing 🙂. We have started shipping
On 28/11/2025 11:34, Joel Busch wrote: these to some customers in Africa. Great for density when you can slot 44x 100G ports into a 1U line card.
Single-lambda 100G is definitely the way to go. LR4 just eats up too much space and pays the data centres a lot more in x-connects than I'd like.
This ^ The other major advantage to the PAM4 is that it all has Forward Error Correction (FEC). At 100g, a 1e-12 Bit Error Rate (BER) is one bit error every 10 seconds. This is simply unacceptable for most applications. Also we're in the analog domain, so BER can go up as things start aging, but still "work". I think RAS gave a talk about this at the last NANOG.
Here's the fun thing about the 100g standards, IEEE made -DR (500m), then the 100gLambda trade group made the -FR (2km) and -LR (10km) standards. DR and FR is exactly the same in terms of link budget, so DR is FR and FR is DR. From FR to LR is 2.3 dB difference; there's no optic-electronic component (Laser/PIN Diode/etc) that has such a small delta. This means the DR and FR are built with the same underlying parts, and DR will have no issues pushing 10km.
https://100glambda.com/specifications/send/2-specifications/9-100g-fr-and-10...
If you look at the 2.1.3 table, you'll see how much signal loss is given to cable, connector, and splice losses. In most cases, this is 5-7 times more than what is achievable in the real world. Unless you have a circuit going through 10 patch panels with saggy fiber, it's a non issue. This does mean you should demand each cross connect is scoped and cleaned with wet/dry process if needed. tl;dr - Use 100g-DR for everything and monitor your pre-FEC error rates. -- Bryan Fields 727-409-1194 - Voice http://bryanfields.net
On 28/11/2025 19:07, Bryan Fields via NANOG wrote:
The other major advantage to the PAM4 is that it all has Forward Error Correction (FEC).
Well, FEC is a necessity for PAM4 to work, as it is more susceptible to noise than NRZ.
Here's the fun thing about the 100g standards, IEEE made -DR (500m), then the 100gLambda trade group made the -FR (2km) and -LR (10km) standards. DR and FR is exactly the same in terms of link budget, so DR is FR and FR is DR. From FR to LR is 2.3 dB difference; there's no optic-electronic component (Laser/PIN Diode/etc) that has such a small delta. This means the DR and FR are built with the same underlying parts, and DR will have no issues pushing 10km.
https://100glambda.com/specifications/send/2-specifications/9-100g-fr-and-10...
If you look at the 2.1.3 table, you'll see how much signal loss is given to cable, connector, and splice losses. In most cases, this is 5-7 times more than what is achievable in the real world. Unless you have a circuit going through 10 patch panels with saggy fiber, it's a non issue. This does mean you should demand each cross connect is scoped and cleaned with wet/dry process if needed.
tl;dr - Use 100g-DR for everything and monitor your pre-FEC error rates.
This is not really good advice. Optically, yes, you can mix-and-match DR1, FR1 and LR1 on the same link in any combination, because they all use a single optical wavelength. However, there is no guarantee that DR1 and FR1 will have the same performance as LR1, especially when the link is longer than 500m (DR) and 2km (FR) and/or has varying levels of OSNR quality. The fundamental building blocks for modern pluggables generally tend to be the same. Because of this, including DR and FR specs. alongside LR makes sense for manufacturers and suppliers because it increases their revenue yield. In production, these are the link budget differences between all 3 variants: * DR @ 3.0dB * FR @ 4.0dB * LR @ 6.3dB If you also consider the minimum Rx power for each variant, you will see the cheaper specs. are also the least sensitive: * DR @ -5.9dB * FR @ -6.4dB * LR @ -7.7dB Yes, each of them can Tx at between +4.0dB - +4.5dB, but the Rx sensitivity will come into question if the link is too long and/or has high loss. Like I mentioned before, we are aggregating 100G-LR1 EoDWDM customers into 400G-FR4 plugs with no problem, inside the data centre. I would not necessarily try doing this between LR1 <=> DR4, because you seriously limit your optical margin. It may work, but you can't have your glass of wine at the bar completely stress-free, ever. Mark.
On 26/11/2025 15:40, Joel Busch via NANOG wrote:
It seems that the wider spaced set is based on IEEE802.3 clause 122 (added in IEEE802.3bs), more specifically in Table 122–6 "200GBASE-LR4 and 200GBASE-ER4 wavelength-division-multiplexed lane assignments". (Note that it doesn't officially specify 400GBASE-ER4, this was apparently added in the 50G-per-lambda-time.)
Just to add for more context, you are right about the tighter spacing of the 400G-ER4 and ER4-Lite plugs being due to chromatic dispersion (CD). Even in the O-band, CD and the dispersion slope start to matter for high-baud PAM4 modulation.
According to a contact at one of the shops the wide spacing is also called LWDM and the tight one is called nLWDM. Both now exist and are out there, so caveat emptor, I guess...
The LWDM (LAN-WDM) is an older spacing technique that was first used in 100G-LR4. nLWDM (narrow LWDM) still runs in the O-band, but with the tighter wavelength spacing, primarily for 400G-ER4 (4x100G PAM4) to get to between 30km - 40km on duplex single-mode fibre. If the older LWDM was used for 400G-ER4 or ER4-Lite, the outer lanes would sit too far away from the 1310nm zero-dispersion point, making it easy for them to pick up noticeably different dispersion penalties and skew. The narrower spacing keeps all lanes closer to 1310nm where the dispersion is lowest and most uniform. Mark.
participants (4)
-
Brandon Butterworth -
Bryan Fields -
Joel Busch -
Mark Tinka