Re: [802.3_B400G] Task Force

Thread Links	Date Links
Thread Prev	Thread Next	Thread Index	Date Prev	Date Next	Date Index

Chris,

It was great fun when the term “engineered link” was created for 10G Ethernet (802.3ae), and how it has continued to be used for over 20 years of Ethernet specification development.

NOTE: my assumption is that the following text is obvious to the reader and is being stated to make sure we’re operating from the same starting point. The reach is just a number that is used for marketing purposes as it really is the loss that’s the critical number. 802.3 does an excellent job of stating a baseline topology and associated loss, and end users can use the loss target to determine how to build (or engineer) their data centers. So, while FR/FR4 may have a 2 km reach, it’s really the loss that’s important as end users could build a 100 m reach solution that has 0 margin.

My previous reply was with respect to the statement that end users panicking and taking drastic measures. I believe that is an unfair and incorrect representation. If relative cost and performance are equal for two competing solutions but one has slightly greater margin, the one with greater margin would benefit the market as it will permit end users to consider different topologies and use cases.

Thanks,

Brad

From: Chris Cole
Sent: Thursday, November 17, 2022 1:26 AM
To:
Subject: Re: [802.3_B400G] Task Force

Hi Brad

I am glad to see you maintain composure in the face of potential adversity.

I did not do a good job explaining that in the IEEE we almost never design margin into our PMD specs, but rather the link budget exactly meets the objective. Perhaps one of the few exceptions will explain our methodology. Below is table from 802.3ba showing the link budget for 100G 10km, 30km and 40km. For LR4 and ER4 over 10km and 40km, respectively, there is 0 margin, or additional insertion loss allowed. However, over 30km, ER4 has 3dB of margin.

An implicit example is use of FR4 or FR (or CWDM4 for 100G) PMDs over the more typical 500m reaches found inside the data center. Because they are specified for 2km, there is plenty of margin at shorter reaches.

The table separately also shows that there is precedence in 802.3 to use fiber specs other than the worst case specified by the ITU-T, as explained in note a. As a practical matter, I have not heard of end users measuring actual link loss, i.e. engineering the link, to make sure it can carry ER4, but you have better visibility into this.

With respect to cost, the best we can do right now, given the state of 800G Coherent and IMDD proposals, is to compare 400G ZR vs. 400G LR4 (10km) which suggests a ratio of ~2x. If you are interested in additional differences now, perhaps you can check Mike's and Frank's presentations, although I see them as premature.

Thank you

Chris

On Wed, Nov 16, 2022 at 10:01 PM Brad Booth <bbooth@xxxxxxxx> wrote:

Chris,

I believe your statement may be exaggerated:

This may panic end users into shutting down their data centers, ripping out their optics, and replacing them with ones that have link budget margin.

Is it possible to state the facts about differences in margin and relative costs?

Thanks,

Brad

From: Chris Cole <chris@xxxxxxxxxxxxxxx>
Sent: Wednesday, November 16, 2022 3:04:28 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_B400G] Task Force

Dear 802.3df Task Force Participants,

During yesterday's discussion, a number of misconceptions arose that need clearing up.

1. PMD Margin

In his presentation [ieee802.org], Mike Sluyski pointed out that the 800G LR4 IMDD proposal link budget margin is zero, and suggested this is bad. Unfortunately; all IEEE PMDs ever (with a couple of exceptions) have zero link budget margin, or more precisely unallocated loss. This may panic end users into shutting down their data centers, ripping out their optics, and replacing them with ones that have link budget margin. Fortunately, IEEE methodology picks the worst case PMD application corner, and defines a link budget that exactly meets it. The margin comes in other ways. Most optics are not deployed in worst case conditions, for example 10km optics operate over less than 10km. In modern manufacturing, IEEE spec. limits are several sigma out. There is margin for measurement uncertainty and other factors. So datacenter operators can be at ease. Despite all of their optics having zero link budget margin, the IEEE methodology is robust and their optics will continue to work just fine.

Spoiler alert: a few optics fail in the field, which is a good thing, zero failures would be prohibitively costly.

2. Coherent Proposal

For months, we have been following the journey of exploration by the Coherent proponents, even though the 800G LR1 solution is obvious. It uses the ZR DSP and SiPIC, operates in C band, and replaces the tunable DWDM laser with a single lower performance fixed source. Let's see how many more detours will be taken before the return home.

3. WDM Grid CD

The 800G LR4 proposal is on the LWDM grid, which is tightly grouped near ZDW, and gives roughly 1/4 of CWDM4 grid CD. This is offset by doubling of the rate from 100G to 200G per lane. To first order, 400G LR4 (10km) has similar CD penalty as 800G LR4 (10km).

4. Relative Coherent vs. IMDD cost.

The best way to compare cost is using existing shipments. During the meeting, Tom Williams cited >2x cost ratio of ZR to LR4 (10km), which for a variety of reasons is an excellent proxy and a reasonable ratio. Right now, ZR volume is more than a magnitude higher than LR4 (10km) volume. We expect the ratio to fall over time. To first order, we can call the LR1 laser source savings washing out any cost ratio increase due to lower long-term volume ratio, giving a LR1/LR4 cost ratio projection of ~2x. All other relative cost advantage calculations by the Coherent and IMDD proponents are convoluted and unconvincing.

Thank you

Chris

On Wed, Nov 16, 2022 at 7:45 AM John D'Ambrosia <jdambrosia@xxxxxxxxx> wrote:

All,

Comments received during Task Force Review of D1.0 have been posted –

By Clause - https://www.ieee802.org/3/df/comments/D1p0/8023df_D1p0_comments_rx_clause.pdf [ieee802.org]

By ID - https://www.ieee802.org/3/df/comments/D1p0/8023df_D1p0_comments_rx_id.pdf [ieee802.org]

If you submitted comments , please take a moment to review the report and check your comments. If you find any issues please contact Matt Brown.

My thanks to Matt Brown and our entire editorial team for all of their hard work.

Regards,

John D’Ambrosia

Chair, IEEE P802.3df Task Force

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1 [listserv.ieee.org]

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1 [listserv.ieee.org]

To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1