Talk:100 Gigabit Ethernet
![]() | The contents of the IEEE 802.3ba page were merged into 100 Gigabit Ethernet on 2013-01-13. For the contribution history and old versions of the redirected page, please see its history; for the discussion at that location, see its talk page. |
This is the talk page for discussing improvements to the 100 Gigabit Ethernet article. This is not a forum for general discussion of the article's subject. |
Article policies
|
Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL |
Archives: 1 |
![]() | This article is rated C-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | |||||||||||||
|
Confusing interface types
[edit]It should be possible to find from the article how many fiber strands the individual 40G/100G interface types need to be connected with for bidirectional operation. As a bonus it would also be nice to know how many wavelengths per fiber are used.
In the table we have the colums: Media, OFC/RFC, #Media, Lanes.
E.g. for 100GBASE-SR10 according to the article Fibre 850nm, MPO-24, 2 # media and 10 lanes.
What does #media? and lanes mean? Does this mean 100GBASE-SR10 needs 10 fibers times 2 (media?) per direction = 20 fibers in an MPO-24 connector?
But then 100GBASE-LR4 has Fibre 4x~1300nm, LC, 2 # media and 4 lanes.
Is it 4 (lanes?) x 2 (#media?) per connection divided by 4 wavelengths = 2 fibers?
But then 100GBASE-ZR according to the article is Fibre 1546.119 nm, LC 2 media, 2 lanes.
That can't be 2 (lanes?) x 2(media?) divided by 1 wavelength because that would be 4 fibers and that does not fit in LC connector. Or does this use multiple LC connectors per connection?
And then 100GBASE-SR2 has 2 media and 2 lanes.
But 100GBASE-DR has 1 media and 2 lanes on a single wavelength?? How many fibers is that then?
My point is the article is completely not understandable for the average reader right now:
- It should be clearly explained what number of media, number of lanes mean, because right now that's not in the article.
- How many fibers are required for a connection. Is this the same number as lanes?
- And how the lanes relate to the required connector and the number of wavelengths.
- And if the requirements are per bidirectional connection or per direction.
Kwinzman (talk) 00:19, 19 December 2019 (UTC)
I agree - and have for a long time. I just reworked the 100G table to hopefully make it clearer. If there are no objections I'll do the same for the 40G table. JeffMorriss (talk) 19:25, 21 April 2021 (UTC)
- JeffMorriss, Is the Lambdas column per media, per direction or total? ~Kvng (talk) 13:47, 24 April 2021 (UTC)
- Kwinzman, both Lambdas and Lanes are per direction (the single arrow is supposed to denote this but the tooltip/hover text tries to make it especially clear). Suggestions for improvement are welcome. :-) JeffMorriss (talk) 12:48, 26 April 2021 (UTC)
- JeffMorriss, The reason I asked is there seems to be some inconsistency. For instance, the 100GBASE-SR10 entry indicates 1 lambda and 10 lanes. I believe it is 1 lambda per lane so lambdas should be 10 there. ~Kvng (talk) 15:55, 1 May 2021 (UTC)
- Kwinzman, 100G-SR10 uses the same lambda on 10 different fibers (i.e., 850nm on fiber 1, 850nm on fiber 2, ...), thus 1 lambda but 10 lanes and 10 fibers (in one direction). Contrast this to 100G-LR4 (and other standards using WDM) which uses 4 lambdas on 1 fiber (thus multiplying out to the 4 lanes used). JeffMorriss (talk) 17:39, 1 May 2021 (UTC)
- JeffMorriss, OK then the Lambdas column should be labeled "Lambdas per media" or somesuch. ~Kvng (talk) 15:15, 4 May 2021 (UTC)
- @Kvng: I don't think "Lambdas per media" would be too useful. A wavelength is a wavelength, no matter how many strands it is used in. "Lambdas per media" sounds like wavelengths in different strands need to be different, too. --Zac67 (talk) 05:20, 5 May 2021 (UTC)
- Zac67, we should try to find some way to clarify that lambdas * media = lanes / 2 ~Kvng (talk) 13:34, 6 May 2021 (UTC)
- @Kvng: I don't think "Lambdas per media" would be too useful. A wavelength is a wavelength, no matter how many strands it is used in. "Lambdas per media" sounds like wavelengths in different strands need to be different, too. --Zac67 (talk) 05:20, 5 May 2021 (UTC)
- JeffMorriss, OK then the Lambdas column should be labeled "Lambdas per media" or somesuch. ~Kvng (talk) 15:15, 4 May 2021 (UTC)
- Kwinzman, 100G-SR10 uses the same lambda on 10 different fibers (i.e., 850nm on fiber 1, 850nm on fiber 2, ...), thus 1 lambda but 10 lanes and 10 fibers (in one direction). Contrast this to 100G-LR4 (and other standards using WDM) which uses 4 lambdas on 1 fiber (thus multiplying out to the 4 lanes used). JeffMorriss (talk) 17:39, 1 May 2021 (UTC)
- JeffMorriss, The reason I asked is there seems to be some inconsistency. For instance, the 100GBASE-SR10 entry indicates 1 lambda and 10 lanes. I believe it is 1 lambda per lane so lambdas should be 10 there. ~Kvng (talk) 15:55, 1 May 2021 (UTC)
- Kwinzman, both Lambdas and Lanes are per direction (the single arrow is supposed to denote this but the tooltip/hover text tries to make it especially clear). Suggestions for improvement are welcome. :-) JeffMorriss (talk) 12:48, 26 April 2021 (UTC)
OK, I've applied these changes to the 40G table too. Is it time to close/delete this issue? JeffMorriss (talk) 14:26, 28 April 2021 (UTC)
Can you clarify 100GBASE-SR2-BiDi as well? It states its a (25g x 2)x2, but its really a 50g x 2. (if it was 25g lanes it'd need 4 fibers or 2 more lambdas). 76.25.36.110 (talk) 15:02, 10 May 2022 (UTC)
- I'm not too happy with the table either – 100G-SR2-BiDi uses 26.5625 GBd x PAM4 x 2 lambdas. --Zac67 (talk) 17:06, 10 May 2022 (UTC)
ZR fiber type
[edit]Hi Zac67, after some rechecking I'm sure ZR works on regular plain SMF, PMF is not required. Please note that the PMF reference was only added by myself in this edit https://wiki.riteme.site/w/index.php?title=100_Gigabit_Ethernet&diff=next&oldid=845141188 a few days ago after I read and misunderstood the limited available specs for ZR. I have since looked through some more documentation and identified my misunderstanding. --Eqvinox (talk) 20:01, 22 June 2018 (UTC)
Article should be about 100GbE or 100/50/25GbE, not 100/40GbE.
[edit]The article should be about 100/50/25GbE, not 100/40. 50/25 are based on the electrical/optical specs that were developed for 100Gbps Ethernet, and emerged in products at the same time. Therefore 100Gbps devices generally support 50/25 if they support 40. 100/50/25Gbps are the only volume technologies above 10Gbps. On the other hand, 40Gbps is a niche based on 10Gbps, and not even supported by most 10Gbps devices. Neither 10Gbps nor 40Gbps were able to beat 25Gbps to the punch of being lower cost per Gbps than 1Gbps, but were useful in system design before there was 100/50/25Gbps. Furthermore, at what point in time was 40Gbps related to 100Gbps? (I mean, sure a 100Gbps device can run at 40Gbps, and the standards were written that way as a curiosity, and the 50/25 standards were late... But the 50/25 *products* were not at all late; they coincided with 100G. So who in their right mind would use an actual 100Gbps product as 40Gbps rather than the lower-cost options of 1x50G or 2x25G?)— Preceding unsigned comment added by 108.185.180.195 (talk) 19:01, 7 September 2018 (UTC)
- Reasonable point. Perhaps we should WP:SPLIT this article into 100 Gigabit Ethernet and 40 Gigabit Ethernet. We have a separate article on 25 Gigabit Ethernet which also covers 50 Gigabit Ethernet. I'd leave that separate for now but we could consider merging it into 100 Gigabit Ethernet if the spit works out well.. ~Kvng (talk) 13:29, 10 September 2018 (UTC)
- The background is the first-generation co-development of 100 and 40 GbE. 40 GbE was based on four 10 Gbit lanes and 100 GbE on four sped-up 25 Gbit lanes (which led to first-generation 50 GbE and 25 GbE). On the long run, this should be split. --Zac67 (talk) 14:37, 10 September 2018 (UTC)
- Support: Makes sense and good argument. 40GBE is mostly QSFP+ (4x10GBE SFP+). And 100GBE QSFP28 (4x25GBE SFP28). 10/40 belong together as does 25/100. The spit makes sense to me.
- The only odd one out is from Twisted Pair Ethernet land where recently with 802.3bq-2016 40GBASE-T and 25GBASE-T were developed alongside. Kwinzman (talk) 10:33, 20 December 2019 (UTC)
- Support: Further, a seperate article on 10 MBit/s (Classic Ethernet) is also missing. Currently everything is redirected to 10BASE-T which is not correct. Fibre-based varieties thereof are misplaced in the latter article. I also have some more TP-PHY tables with the new design left to be added to this topic. Nightwalker-87 (talk) 14:51, 10 September 2018 (UTC)
- 10BASE-T redirects to Ethernet over twisted pair. There are also 10BASE-F, 10BASE2 and 10BASE5. Ethernet and Ethernet physical layer ties this all together pretty well but maybe another hub at 10 megabit Ethernet would help. ~Kvng (talk) 13:15, 13 September 2018 (UTC)
- Classic Ethernet is now available as the new hub article for the generation of Ethernet standards. It is likely useful to merge-in the content of 10BASE-F (article stub), while 10BASE2 and 10BASE5 should remain seperate. Nightwalker-87 (talk) 22:23, 13 September 2018 (UTC)
Remove the list of companies
[edit]Or maybe short the list, because I do not think it does add a lot of encyclopedic value. — Preceding unsigned comment added by Initramfs (talk • contribs) 10:55, 12 December 2018 (UTC)
- There are companies listed in Early products and Commercial trials and deployments sections. Which are you proposing to remove? ~Kvng (talk) 14:05, 5 November 2019 (UTC)
CFP-100GBASE-ZR
[edit]Dear Zac67, I don't want to enter a revert war with you or anyone else.. We have information from Juniper Networks Inc., which indicates that their proprietary standard is not licensed to any other MFR. If there is/are other(s) with different wavelength/line code, then you should add a row in the table. --2001:678:A08:0:54AB:F35B:B339:6163 (talk) 18:35, 2 November 2019 (UTC)
- BrightStar is just selling Refurbished/Used Networking Equipment. They are not MFR. Specifications section indicate: Juniper Networks, Inc as the Manufacturer. FYI. --2001:678:A08:0:54AB:F35B:B339:6163 (talk) 18:49, 2 November 2019 (UTC)
100GBASE-SR2 & 100GBASE-SWDM2
[edit]I fixed an error in the 100G table, which listed 100GBASE-SR2 as having two lambdas, when it should've said one lambda. The numbers didn't match the source, and also didn't add up. I presume it was either a typo, or a misread of the second source, which talks about 100GBASE-SWDM2.
On that front, I'm trying to figure out what standard, if any, does exist that matches the mistaken listing, i.e. 100G duplex over 2 fibres, 2 lambdas, PAM4. Here's my research so far:
- 26.5625GBd with PAM4 gives you 50G per lane after line coding and FEC overheads. For 100GBASE-SR2 that's 2 lanes, 4 fibres. If you used 2 lambdas that'd be 2 lanes, 2 fibres.
- 100GBASE-SWDM2 appears to be the gen3 equivalent of 100GBASE-SR2-BiDi, i.e. 26.5625GBd * PAM4 * 2 lambdas, giving you a 100G link over a single fibre pair.
- 100GBASE-SWDM2 is mentioned in the second source but that seems to be a proposal against 100GBASE-SWDM2 as a standard.
- Another IEEE source mentions 100G-SWDM2 in passing but not as a ratified standard.
- This presentation from 2017 suggests that 100G-SWDM2 is a proprietary solution, using gen3 signalling (25G with PAM4 for 50G) over two lambdas.
- This publication makes reference to 100G-SWDM2 having "already been demonstrated by a few vendors", but doesn't source that claim.
- This IEEE presentation has a timeline on slide 2 that suggests that 100G-SR1.2 (which would be 25G * PAM4 * 2 lambdas = 100G on one pair) may have been rejected as a standard.
Searching online I can only find talk of these standards being proposed or upcoming. As far as I can tell the standard doesn't yet exist, nor do any modules.
My familiarity with fibre standards is superficial, so I'm erring on the side of caution for now and have not included any of this in the page. If someone else has a greater familiarity of the subject and wants to integrate this info in, please do. Gsuberland (talk) 18:12, 26 September 2021 (UTC)