Using the 40/100G CFP MSA for nx10G Ethernet

When the 40/100G CFP MSA (http://www.cfp-msa.org) first came out in March I initially commented that it seemed optimized for 100G applications, though its size would make it easier to implement initial 40G modules. However after after taking a look at new systems from Extreme Newtorks and Arista Network, plus having some time to digest the finer details of what’s written, it’s now clear that the “size” of the CFP module open up the potential to use the format not only for 40/100G but also for high density 10G ports.

One of the concerns I’ve had about the adoption of 40/100G Ethernet is how easy it will be to build modular systems that allow people to switch from nx10G to 40G and finally on to 100G, since each line rates uses a different mechanical format (SFP+, QSFP and CFP). Having such a broad range of mechanical formats creates big problems in inventory management (you never have the right module) and was probably one of the factors that slowed the adoption of 10G Ethernet down (XENPACK, X2, XPACK, XFP and finally SFP+). It’s interesting to note that the Fibre channel community avoided this as much as possible with 1Gbps upto 8Gbps FC all using the same SFP format.

After discussing this concern with a friend I was pointed to Extreme Networks and Arista Networks to take a look at how these two companies were approaching the problem. The latest system from Extreme (X650) has an uplink module of eight 10GBase-T links using link aggregation, it’s easy to envisage the step to a plug in module using 2 x 40G especially as the internal system interface “could” be identical (8xXFI could also work as 2xXLAUI). This lead me to consider how the CFP format could be used not just as a “plug in” Optical module but as a network interface module covering both Optical and Electrical interfaces, here’s what I mean…

  • 1x40GBase-SR4/CR4
    A single QSFP cage with quad SFI+ retimer, such as those from Gennum, interfacing to the CFP XLAUI system interface. The available QSFP pluggable modules seem to come in three types, optical modules with an MTP/MPO connector (e.g. Avago) , “Active” optical cable (e.g. Finisar ) and direct attach which is for copper (e.g. Quellan).
  • 4x10GBase-SR/LRM SFP+ module based on CFP format
    Using a 4xSFP+ cage (e.g Pulse Engineering), which just about fits into the width and height of the CFP format, along with a Quad SFI+ retimer from Gennum to interface to the XLAUI system interface. This would allow the use of 10G SFP+ modules such as those from Opnext, Excelight, Finisar and Avago.
  • 4x10GBase-T module based on CFP format
    Using a 4xRJ45 ganged connector (e.g Pulse Engineering or Belfuse) with two dual 10GBase-T transceivers from Teranetics which support XFI for the system interface. The Teranetics devices are 6W per channel so the module would consume in the region of 25W (including on board controller) which is just on the border line between the CFP class 3 and class4 rating.

All the above assumes that the MAC can be switched from 4x10G to 1x40G, this is probably not unreasonable and the announced Prestera CX from Marvell is described as having this capability, we’re eagerly waiting further detailed public info on this new switch family to find out more.

Initially I thought that the maximum number of 10G ports one could fit on a CFP would be 4, but then I checked out what Arista were doing. What really caught my eye here was how their “Pass Through” cards (used with IBM BaldeCenterH) use QSFP to 4xSFP+ direct attach cables to consolidate the cabling – now this got me really thinking (or dreaming)….

  • 3x40GBase-SR4/CR4 or 12x10GBase-SR/CR1
    Use the 3 x QSFP option in the CFP MSA and either direct attach QSP:4xSFP+ modules or standard QSFP Optical modules with MTP/MPO to 8xLC connector fiber optic cables from Molex.

To take advantage of the possibility to aggregate 3x40G and 12x10G onto a single CFP would require a very flexible 10/40/100G MAC but just imagine the possibilities : a 24+2 switch (by that I mean 24x10G + 2x100G) on a 1U platform! It will take a while as we’ll need an Ethernet switch chip that is capable of 0.5 Tbyte capacity that supports Data Center Ethernet (DCE) but definitely something to look forward to.

40G Ethernet Resource Center
http://www.40gethernet.com
http://twitter.com/40GEthernet

One Response to “Using the 40/100G CFP MSA for nx10G Ethernet”

  1. 40gethernet Says:

    Oct 21, 2009 12:10 ET : Reflex Photonics to Introduce New Dual 40GBASE-SR4 40G Ethernet CFP Transceivers at Supercomm 2009
    http://www.reflexphotonics.com

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s


%d bloggers like this: