Re: [802.3BA] Discussion on 40G for => 10 km SMF

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

Hi Gary,

Thank you for the insightful questions in your Sept. 10 email; they are helpful in clarifying the data.

In my Sept. 8 email, I plotted OC-192 SR-1 only until 2003 because I was making apples to apples cost comparison of the first three years of OC-192 (2000 to 2003) and OC-768 (2005 to 2008) shipments. In the initial years, OC-192 was based on 300-pin form factor using 16:1 SerDes technology. However around 2003, OC-192 rapidly transitioned to XFP form factor using CDR ICs (no SerDes, i.e. true 10G I/O serial technology.) Since OC-768 is still based on 300-pin form factor using 16:1 SerDes technology, we wanted to look only at comparable OC-192 data.

As per your request, the Ovum RHK OC-192 SR-1 data is extended below to 2007. In later year reports, Ovum does not separate the cost of 300-pin and XFP based modules; they just give the aggregate cost of all OC-192 units. XFP represented a dramatic technology shift in 2003 for OC-192 by eliminating 16:1 SerDes ICs and fiber pig tail OSAs in a much smaller form factor. XFP is a much more dramatic technology change then the proposed 40GE-Serial modules which will use both 4:1 SerDes ICs and fiber pig-tailed OSAs (internally.) That is why the graph in my Sept. 8 email showing OC-192 300-pin 16:1 SerDes technology followed by XAUI 4:1 SerDes technology is a better analogy to what we can expect for OC-768 transition to 40GE-Serial.

However, the extended Ovum data offers other insights. It shows what kind of cost drop we can expect from the introduction of a dramatic new technology. The transition to XFP dropped OC-192 SR-1 cost by an additional factor of 2x in 2003. This significantly increased the aggregate cost drop to 6x from 2000 (i.e. in the first three years.) Afterwards, OC-192 SR-1 continued on a typical high volume optics cost reduction curve.

Also as per your request, the Lightcounting OC-768 data is extrapolated with the exact Ovum OC-192 relative cost data. This is done in two ways; one with 2007 as starting point and second with 2008 as starting point. In the graph, all 10G data is normalized to 2000 OC-192 cost, and all 40G data is normalized to 2005 OC-768 cost.

Here are the conclusions from the OC-192 SR-1 cost data:

The cost decline of OC-192 and OC-768 SerDes based 300-pin modules in the first three years of shipment are similar, >3x cost drop for OC-192 and 2.5x cost drop for OC-768. This makes OC-192 a good historical analogy to use in 40GE discussion.
A dramatic shift from SerDes and fiber pig-tailed OSA based 300-pin technology to CDR based receptacle OSA XFP technology gave OC-192 modules an additional factor of 2x cost drop in 2003. We should expect similar ~2x cost drop boosts from future dramatic technology shifts; NOT 10x.
The cost projections by 40GE-Serial proponents are a factor of ~3x more aggressive then predicted by even the best case (i.e. transitioned to XFP) OC-192 SR-1 historical cost data. This makes these cost projections a poor basis for making standards decisions.

(The historical comparison of 1GE to 10GE data you requested can be found on page 7 of cole_04_0708. Also, if you are wondering about the time stamp of this email, I am not writing this on a Saturday night in California, but rather on Sunday while stuck in a typhoon in Taiwan.)

Chris

From: Gary Nicholl (gnicholl) [mailto:gnicholl@xxxxxxxxx]
Sent: Wednesday, September 10, 2008 12:26 PM
To: Chris Cole; STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3BA] Discussion on 40G for ="" 10 km SMF

Chris,

Very good data points and an interesting analysis. However I am not sure why you stopped the OC192 curve in 2003 (when people are still shipping OC192 today). Is the assumption that the pink curves tends towards the yellow curve (joins ?) after 2004 (and is this why you didn't continue the pink curve ? ). I also get confused when people start mixing sonet and ethernet when comparing costs (given that the technologies have very different cost structures).

Also at a quick glance, the pink to yellow transition doesn't look that different to the green to blue transition ... but maybe the difference does not come across very clearly because of the scale of the graph ? Maybe if you had plotted how the green curve would track if it followed the same path as the pink to yellow transition, and then compared that to the blue curve it might make it easier to see the differences (and challenges) for 40GE serial.

Another analysis which I would find very useful would be to look at the cost of 10G LR from 2003 (introduction using 300pin module) to 2008, compared to the cost of 1G LR over the same period, i.e. I would like to see how the multiplier for 10G versus 1G changed over the period from 2003 to 2008.

Gary .......................

From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx]
Sent: Monday, September 08, 2008 9:37 AM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF

In his Aug. 7 reflector email below, Takai-san asked about 10G cost during the early stage of deployment. We realized that such data could offer excellent insight into 40G optics costs so we dug up a 2003 Ovum RHK report with early OC-192 costs.

In the enclosed graph, we plotted OC-192 SR-1 300-pin 10G data from Ovum RHK together with 10GBASE-LR XAUI data from Lightcounting (both relative to 2000 OC-192 SR-1 cost.) To the same graph, we added OC-768 VSR 300-pin 40G data from Lightcounting and 40GE-Serial cost projections from traverso_02_0708 (both relative to 2005 OC-768 VSR cost.)

Takai-san’s intuition that OC-192 10G cost dropped dramatically in the early stage turned out to be correct; the drop was 3x in 3 years. Interestingly, OC-768 40G cost in its early stage mirrored this dramatic drop, falling by 2.5x in 3 years.

The data further shows that the first 10GBASE-LR XAUI modules in 2004 were a substantial >2x cost reduction in two years, from the cost of OC-192 SR-1 10G modules in 2002.

However, this historical 10G >2x cost drop is dwarfed by the enormous >10x cost drop projected in traverso_02_0708 for the first 40GE-Serial modules in 2010 from the cost of OC-768 VSR modules in 2008.

In fact this projected initial 40G >10x cost drop in 2 years is so dramatic that today 10GBASE-LR XAUI modules have not yet dropped this much from the cost of OC-192 SR-1 10G modules in 2002; the 10G cost drop has been about 6x in the last 6 years.

802.3ba Task Force would be well served next week not to rely on such hugely optimistic 40GE-Serial cost projections when considering proposals for the 40GE 10km SMF PMD baseline.

Chris