| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Team, I would like echo Jeff’s concern about market acceptance of FEC latency, for server to server (East-West) traffic in datacenters.
·
Majority of the traffic in the access and aggregation switches is E-W traffic, and increased latency directly affects the “response time”
seen by the applications and users.
·
For 100m fiber links, the RS-FEC latency will add more than 20% to the fiber delay, which will be a noticeable impact. (and worse for shorter links).
·
This increased latency will be noticeable when 100G SR4 / PSM4 modules are used to replace existing 40G SR4 or 100G SR10 modules.
·
Considering the industry trend and investments in lowering the latency of datacenter switches, if customers demand zero latency on PMDs, it will promote non-standard optics (or operating modes) that disable the FEC on SR4 / PSM4 modules. (and then provide extended reach using FEC). Also the assumption that FEC is already available on the host may not be true.
·
We expect 100G “optical” ports to use retimed modules, and support for CR4 passive copper cables may not be required or may be lower priority, so the FEC is not “free” for these hosts. There will be definite cost and power increase going from CAUI-4 / VSR port to CR4 capable port.
·
The use of FEC for .3bm optics (SR4, PSM4) should be treated as additional component being added to the host design, or inside the module.
·
We need to compare the cost / power of “adding” FEC logic into VSR type ports, versus other solutions (eg: more equalization, better optics). --vineet From: Jeffery Maki [mailto:jmaki@xxxxxxxxxxx]
John, The assumption of the use of FEC has to be explicit so all know that it must be implemented and so that they know that the latency of FEC will occur. Is FEC really so important to the technical proposal so at
to justify it? Cost reduction of FEC would be one metric for its justification. The issue then turns to market acceptance of FEC latency. For 500 meters, one might argue 802.3bj FEC latency is not an issue. As the reach goes down, the relative contribution of the FEC latency increases
in terms of the overall link latency (time of flight of light in glass). Please keep in mind also that absolute link latency is a market differentiating variable. I’d like to see market acceptance of FEC latency presented as a function of reach for optical
links. (Just because the 802.3bj task force adopted an FEC with a certain amount of latency does not mean we get to stop thinking in the 802.3bm task force about our own market acceptance of FEC.) Jeff From: John Petrilla [mailto:john.petrilla@xxxxxxxxxxxxx]
Hello Jeff Unfortunately, justification of the 500m SMF objective for the project is explicitly cost, size and power. While I can sympathize with other with other issues, I would need a lot of help explaining why cost,
power or sized was sacrificed. Regards, John From: Jeffery Maki [mailto:jmaki@xxxxxxxxxxx]
All, We have to make sure that the assumption of the use of FEC is explicit. Although we might argue not to count the cost of the FEC encoder/decoder itself, we do need to understand the impact on cost that the use
or lack of use of FEC poses. It is good thus to make analysis of the impact on cost of designs presuming the use of FEC versus designs presumed NOT to use FEC. I see the potential lack of need of FEC as one of the technical advantages of parallel single
mode as well as market-acceptance advantages. Jeff From: Anslow, Peter [mailto:panslow@xxxxxxxxx]
Hi, As previously announced, there is an SMF Ad Hoc meeting starting at 8:00 am Pacific today Tuesday 18 December. I have currently received two requests for presentations, so the draft agenda is: ·
IEEE patent policy reminder o
http://www.ieee802.org/3/patent.html ·
Approval of the draft minutes from 4 December call ·
Presentation o
PSM4 Technology & Relative Cost Analysis Update Jon Anderson, Oclaro o
Basic Study on Receiver Bandwidth Requirement for Discrete Multi-tone Modulation Masato Nishihara, Fujitsu
·
Discussion ·
Future meetings (next opportunity - 8 Jan) I hope to post both presentations on the SMF Ad Hoc
web page just prior to the meeting. Peter Anslow from Ciena has invited you to join a meeting on the Web, using WebEx. Please join the meeting 5-10 minutes early so we may begin on time.
+44-203-4333547 (United Kingdom) 4438636577 (United States)
Regards, Pete Anslow |
Senior Standards Advisor |