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

Re: [802.3_100GNGOPTX] Relative Cost to 100GBASE-SR10 CXP



Chris,

 

Thanks for kicking this off.  I agree that we should come to agreement on the cost ratios – at least have a range. 

 

I had lunch with Brad Smith of Lightcounting and he provided me their latest numbers for the ratio of 100GBASE-LR4 to 100GBASE-SR10.  Here are a couple of data points:

2014 – LR4 = 12 X SR10

2015 – LR4 = 8 X SR10

 

So the forecasts of cost ratios do drop as everyone hopes –from 12 to 8 in one year.  This is mainly due to aggressive cost declines for LR4.  I don’t think Lightcounting has taken an MR4 solution into consideration for their forecasts. Brad is now on this reflector so he could comment on these forecasts but he’s probably on a plane now.

 

I think 5X ratio is too aggressive for the foreseeable future.  I would suggest to use 8X.

 

Thanks,

Scott

 

From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx]
Sent: Wednesday, January 25, 2012 1:49 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Relative Cost to 100GBASE-SR10 CXP

 

It could be helpful to future cost comparison exercises if we had a common set of assumptions on relative cost. Scott Kipp initiated off-line discussion by proposing ratios in his presentation. This gave us the opportunity to look up Lightcounting (LC) numbers for several optical interfaces which I would like to use in a proposal below.

 

For 2012, LC projects the cost ratio of SFP+ 10GE-LR to SFP+ 10GE-SR as ~3x.  The raw LC 2012 cost ratio projection for 40GE-LR4 to 40GE-SR4 is about 6x. Since the 40GE-LR4 cost blends CFP and QSFP+ form factors, our view is that the a more accurate market cost ratio in 2012 for QSFP+ only based modules for 40GE-LR4 to 40GE-SR4 will be about 4x.

 

The raw LC 2012 cost projection for 100GE-LR4 (discrete EML based) to 100GE-SR10 is about 16x. The LR4 cost is primarily CFP based and the SR10 cost is primarily CXP based. If we wanted to factor out differences in form factor cost, 10x would be a more reasonable ratio. If we assume a 2x cost reduction from discrete EML to DFB laser PIC, this leads to a cost ratio of about 5x.

 

Based on this, the proposal is that we use a long term cost ratio of 5x for 100GE-LR4 (Gen2 DFB laser PIC based) to 100GE-SR4. The mid-term cost ratio proposal is also 5x for 100GE-LR4 (Gen2 DFB laser PIC based) to 100GE-SR10 in similar packaging, and 8x for 100GE-LR4 CFP Gen2 to 100GE-SR10 CXP.

 

It would be worthwhile for others to review our interpretation of LC numbers, and to have additional off-line discussions to tweak the above ratios to enable apples to apples cost comparisons in future presentations.


Thank you

 

Chris

 

From: Ali Ghiasi [mailto:aghiasi@xxxxxxxxxxxx]
Sent: Wednesday, January 25, 2012 11:57 AM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] Relative Cost to 100GBASE-SR10 CXP

 

Jeff

 

Some very valid points but regard to your item 7.

When the host ASIC/switch I/O due to capacity moves to 25G then your option are either use an inverse gearbox with 

100G-SR10 or use 100GBase-SR4.  This is the time I expect most will move to 100G-SR4 if it exist.

 

Thanks,

Ali

 

On Jan 25, 2012, at 11:39 AM, Jeffery Maki wrote:

 

All,

 

I think we are nearly a potential epiphany based on a “bottoms up” review of material presented in the study group.  Let’s listen openly to the presentations, and outline encouragingly where we would like to see more material/data.

 

In my opinion,

1.       100GBASE-SR10 CXP should be the basis of all relative cost comparison for any potential SMF Objective or any potential MMF Objective.

2.       The relative cost of 100GBASE-LR4 to 100GBASE-SR10 should be understood for any potential common form factor, and I will note that relative cost does depend upon form factor.

3.       Comparing to 100GBASE-LR4 misses the point of the study point that I believe is best known as “datacenter reach extension.”  “100GBASE-nR4” was not meant to replace 100GBASE-LR4 but fill in where parallel optics either fail to perform or market acceptance of parallel fiber does not exist where I presume duplex SMF for “100GBASE-nR4.”  Low cost, low power, and compactness (CFP4/QSFP) are paramount or the potential standard will not be adopted by the market.

4.       We should realize that 100GBASE-SR10 will live as a deployed, durable, standard for years to come and likely shall see follow on implementations beyond CXP.

5.       Parallel 4x25G SMF optics are showing potential merit at present as a viable proposal for interconnects in cases where the bulk of parallel fiber is acceptable.

6.       “100GBASE-SR4” needs to be proven and at present its potential contender is parallel 4x25G SMF optics.

7.       I would like to learn which markets will stick with 100GBASE-SR10 and which markets will move to or develop around “100GBASE-SR4.”

 

Jeff

Juniper Networks