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

Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction



Ali,

 

Mark Nowell corrected my 10G road map 300-pin starting point as not being the first form factor but rather the second, because the first 10Gb/s clients were implemented as entire host cards.

 

This also reminded me that 40Gb/s clients also followed this roadmap; first host card implementations first, next multiple 300-pin versions, followed by CFP, and then QSFP+.

 

So 100G CFP could be argued to have skipped two generations in size and plugability. That does not change the fact that this was the first 100GE SMF PMD implementation so comparisons to XAUI are not helpful.

 

Chris

 

From: Ali Ghiasi [mailto:aghiasi@xxxxxxxxxxxx]
Sent: Friday, February 03, 2012 12:23 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction

 

Chris

 

You have previously said in regard to the 100G alphabet soup we skipped one form factor so the CFP should be 

compared to Xenpak and not 300 pin but the cost is comparable to 300 pin!  If I recall correctly Xenpak was 1/10 the 

cost of 300 pin.  The 300 pin MSA had very short lived life in 10 GbE and in similar time frame Xenpak was driving 

force and worked already started on X2XPAK.  

 

One of the reason for 100GNGOPTX study group is the astronomical cost of 100G-LR4.  All I am saying we should not 

be using something with astronomical cost to justify another PMD 1/2 astronomical cost.  

 

Thanks,

Ali

On Feb 3, 2012, at 11:43 AM, Chris Cole wrote:



Paul,

 

As long as we insist on looking at 1st generation 100G technology cost, we will get reactions like: ”100G-LR4 with its astronomical cost!” This is similar to suggesting we use as baseline 1st generation 10Gb/s modules in the large 300-pin form factor from the late 90’s, instead of 5thgeneration 10Gb/s modules in the SFP+ form factor.  

 

40GE QSFP+ already has a large supply base, ramping volume, and costs estimates which have broad agreement. It is also has a well justified 4x cost ratio between LR4 and SR4 so that we can study SMF and MMF trade-offs.

 

Chris

 

From: Kolesar, Paul [mailto:PKOLESAR@xxxxxxxxxxxxx] 
Sent: Friday, February 03, 2012 10:52 AM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction

 

While I can appreciate the various perspectives on cost comparison baselines, some preferring mature technology, some preferring form factor commonality, here we have a situation that warrants different considerations.  Here we are studying possible additional 100G solutions.  The operative word is “additional”.  This means we already have existing solutions to compare to.  By using these existing solutions as comparatives we can directly tell if we are indeed projecting lower costs.  If we do not get favorable projections, then the justification for a project with an aim of optimization (i.e. doing better than before) would loose a critical support structure.  That loss may not kill the project, because it may still proceed due to density improvements, but it would be a big wound. 

 

While Petar and Chris agree on the use of QSFP+, I think they do so for very different reasons.  I will hazard a guess and say that Petar may want a 4-lane migration path.  For Chris, his past arguments have been based on projections to future generation optics in different packaging.   Both valid, but their convergence is coincidental.

 

We can take any baseline and apply it uniformly.  But as long as the baseline is stable over the timeframe of our studies, why introduce more obscurity with “currency” conversion?  Is stability the problem?  Given what may be reasonably close and respectable volumes of multimode optics units being sold into HPC for 4x10G and 10x(or 12x)10G, they are likely just as good to use as baselines from a stability perspective.  So I’d argue to stay with 10GBASE-SR10 as baseline.

 

Regards,

Paul


From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx] 
Sent: Friday, February 03, 2012 12:05 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction

 

Peter,

 

I fully agree with you.

 

The right comparison is with 40GE-SR4 QSFP+ and 40GE-LR4 QSFP+.

 

Chris

 

From: Petar Pepeljugoski [mailto:petarp@xxxxxxxxxx] 
Sent: Friday, February 03, 2012 9:30 AM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction

 

Hi Ali, 

I agree with you. We need to find the proper comparison. For one, I think the lowest cost at one time was 4GFC, (although today, per gigabit its cost is too high relative to some other solutions that appeared recently). 

I also agree with Dan Dove that the 10x perfromancce 3x cost model does not hold any more, at least in the environment I am interested in. I can't imagine the 4x25G solution to be more expensive than 10x10. If it is, we are doing something wrong. By all accounts, it should cost less (smaller number of components, smaller footprint, smaller number of fibers in the MMF cable). In my opinion (I know some people will disagree), proper comparison is with 4x10. 

If we develop standard around that cost model, I believe people will start developing custom solutions that address their cost targets. This will further fragment the market and increase the cost for the standard. 


Regards, 

Peter 


Petar Pepeljugoski
IBM Research
P.O.Box 218 (mail)
1101 Kitchawan Road, Rte. 134 (shipping)
Yorktown Heights, NY 10598

e-mail: petarp@xxxxxxxxxx
phone: (914)-945-3761
fax:        (914)-945-4134




From:        Ali Ghiasi <aghiasi@xxxxxxxxxxxx> 
To:        STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx 
Date:        02/03/2012 12:10 PM 
Subject:        Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction





Paul/Dan 

My concern by using 100Gbase-SR10 which is high cost we are creating too low a bar.  The situation is even 
worse if you look at 100G-LR4 with its astronomical cost! 

I prefer still for any cost comparison to go to 10Gbase-SR/LR something were its cost close to earth than moon. 

Thanks, 
Ali 


On Feb 3, 2012, at 6:03 AM, Kolesar, Paul wrote: 

All, 
I struggled with the cost basis question while I prepared my contribution to the just-past meeting.  I considered using either 10GBASE-SR or 100GBASE-SR10 as the basis.  In reviewing contributions on transceivers (a.k.a. PMDs) I noticed that there was a tendency to use 100GBASE-SR10.  So to make it straight forward I decided to also use 100GBASE-SR10 as the cost basis for the cabling cost analysis portion of the Solution Set Analyzer called “Kolesar Kalculator 2012_01_25” in the tools folder on our web site.  This spreadsheet analysis tool allows one to get a complete channel cost analysis that includes the transceiver modules and the cabling that connects them.  See 
http://grouper.ieee.org/groups/802/3/100GNGOPTX/public/tools/index.html 
  
Therefore, to use this tool to get a complete cost comparison, the cost basis comparison of transceiver modules must also be entered relative to 100GBASE-SR10.  
  
I offer this tool as a means to get various cost projections on the same playing field.  My rationale is that using a common analysis tool will remove some of the variability factors that cloud the cost picture. This should allow us to make more confident and less contentious decisions regarding reach objectives that often contain significant consideration of the trade-off between reach and cost.   
  
The analysis that I included at the end of my user’s-guide contribution “Solution Set Analyzer Update (revised)” 
http://grouper.ieee.org/groups/802/3/100GNGOPTX/public/jan12/kolesar_01a_0112_NG100GOPTX.pdf 
  
indicates that the optimal reach objective for a possible future 100GBASE-SR4 is heavily dependent on the cost of the single-mode transceivers that will be needed to satisfy channel lengths that exceed -SR4 capability.  I stressed this point at the conclusion of my presentation, and I bring it up here again, because the structure of the ad-hocs mentioned by Dan compartmentalizes and separates the MM from the SM studies.  While this seems like a natural and traditional organization structure, and there is overlap among those involved, I would like all involved to keep this fact in mind:  These are not independent silos.  They are highly interdependent parts of the whole data center ecosystem and must be treated that way in order to arrive at an optimal solution set.   
  
So I encourage folks in both ad-hocs to apply the Solution Set Analyzer to sets of PMDs (both MM and SM) that are needed to provide complete data center channel coverage.  With sufficient cross-pollination between the ad-hocs, each should be able to keep informed of the other’s relevant contributions in a timely way.   
  
Regards, 
Paul 
 

 



From: Daniel Dove [mailto:ddove@xxxxxxx] 
Sent:
 Thursday, February 02, 2012 10:41 PM
To:
 
STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject:
 Re: [802.3_100GNGOPTX] Minutes Uploaded - Forward Direction 
  
Hi Ali,

I don't think the 10x vs 3x model applies to new PMDs for a given speed.

What is important is that we believe we can achieve Broad Market Potential, Economic Feasibility, and Technical Feasibility for any particular objective we produce. (assuming distinct identity and compatibility are met)

First, we need to convince ourselves of this. Then, we need to convince 802.3 and the SEC.

Dan


On 2/2/12 8:33 PM, Ali Ghiasi wrote: 
Dan 
  
We can go with 100GBase-SR10 cost if its cost has reach traditional Ethernet cost which is 10x BW 3X the cost. 
Otherwise it would be better to go back to the basic  100G-SR4 should be 3x the cost of 10GBase-SR 
and the new SMF PMD should be 3x the cost of 10Gbase-LR. 
  
I expect both 100Gbase-SR10 and 100Gbase-LR4 would fail above criteria! 
  
Thanks, 
Ali 
  
On Feb 2, 2012, at 4:49 PM, Daniel Dove wrote: 


Participants,

I should mention that my suggestion below to use 100GBASE-SR10 for relative cost is not mandatory. It was a suggestion and should be considered as one possible approach to relative cost assessment for SR4.

For an SMF objective, the ad hoc should decide what it wishes to use for relative cost assessment if anything.

In theory, having a common point for relative cost assessment is not required. Each presenter can provide their best judgment, explain how they came to it, and let the group determine whether they are accurate. This approach is also acceptable.

Regards,

Dan Dove



On 2/2/12 2:43 PM, Daniel Dove wrote: 
Dear Study Group Participants,

The draft minutes have been uploaded to our website at http://www.ieee802.org/3/100GNGOPTX/public/index.html. Please take a moment to review and feedback any items to Kapil Shrikhande our vice-chair.

For those who attended the Study Group meeting in Newport Beach, you are aware that we ended without a wrap-up due to a call for adjournment, second and lacking opposition, we adjourned. At the time, I felt it would be better to gather our thoughts and focus on how to move forward.

For those who did not attend, we were able to secure another objective, "Define re-timed 4-lane 100G PMA to PMA electrical interfaces for chip to chip and chip to module applications". In addition, we saw many presentations related to study for MMF and SMF objectives.

We ran a few straw-polls and essentially the outcome was that we have additional work to come to consensus on either a SMF or MMF objective. In addition, guidance was provided through the straw-polling, but it may not have been specific to making progress as a team. Rather, it was oriented towards the type of material we would like to see. So, below I will address a proposal for making forward progress using the information gleaned from our meeting.

Adhocs:


Multi-Mode ad hoc led by Jonathan King. 
Single-Mode ad hoc led by Pete Anslow.

These two groups will solicit participation and meet via teleconference to: 
1.        Identify a straw-man objective that can gain consensus 
2.        Identify specific media/reach, market potential related to that reach based on relative cost to SR10, technical feasibility 
3.        Identify presentation material that they believe will convince the SG their objective is valid. 
We need to avoid getting into "baseline proposal mode" where we see our preferred alternative competing with the other alternatives. The goal of SG presentations should not be not to sell a proposal. They should be focused on demonstrating an objective meets the 5 criteria with multiple approaches. 

If these two groups can pull together a compelling set of objectives, and presentations that demonstrate they meet the 5 criteria, we can then come together in March and work to identify any remaining consensus opportunities that can be worked on for the upcoming May meeting.

Presuming we can finalize our objectives, 5 criteria responses, and get our PAR completed in May, we will be ready to pre-submit and move toward a July PAR submittal.

Best Regards, 

Dan Dove
Chair, IEEE 802.3 Next Generation Optical Ethernet Study Group