OSFP interconnect: what is it?

OSFP, Octal Small Formfactor Pluggable, is a very new module and interconnect system in development that is targeted to support 400-G optical data links inside datacenters, campuses and external metro long reach. The private consortium OSFP MSA group was founded by Google and is led by Arista Networks. This consortium already has more than 50 member companies, though most member companies are optical component or module manufacturers with much fewer OEMs and end-user companies. It is likely that this group will develop an interconnect spec to a certain degree then release it to the public and have the SNIA SFF committee create a much more detailed specification and an official SFF-xxxx document number as several QSFP, Quad Small Formfactor Pluggable, modules, connectors and cables have been done.

OSFP’s first iteration is an 8-lane times 50G PAM4 = 400G physical link with a possible future 4 x100G PAM4 = 400G and 8 lanes times 100G PAM4 = 800G variants. This physical packaging system is agnostic relative to the different protocol I/O interfaces that will likely use it. The OSFP module or cable plug is a dual paddleboard direct-attachment type connection. The plug’s PCB paddleboard will have an EEPROM chip for memory-mapping and system management functionalities, similar to other established DAC interconnects. The right-angle, receptacle connector and its metal cage and heat-sinking mechanisms are under NDA development. The only image available is of just the module from the OSFP MSA website, to the right.

OSFP’s module size is said to be slightly wider and deeper than the QSFP-DD, module thus taking up more PCB surface area. Several OSFP modules on a line card use much more area and only 32 ports per 1U box faceplate are possible versus the 36 ports of QSFP-DD. The larger size OSFP may not have enough power and cooling advantages versus QSFP-DD’s density and its four extra port capability. Different users with different applications may fervently prefer one connection system versus the other relative to their panel density, cost, performance, power and cooling priorities.

OSFP implementations will likely use mostly modules and passive optical cables versus relatively short active copper cables and even fewer passive, very short reach much, larger diameter and heavy copper 2-3 meter cables with 16 twin-axial pairs. Eight- and 16-lane design development and signal integrity standard measurements take much longer to master and achieve compliance and interoperability versus similar throughput 1-, 2- or 4-lane connections.

Some market forecasting indicates that the rapidly growing 100 G per link usage market take up is helping to further drive the consumption uptake of 400-G interconnects. It seems that different market segments and customers may favor one type of 400-G interconnect but some OEMs and interconnect suppliers will need to develop and offer OSFP, QSFP-DD and standard OBO products. It appears that QSFP-DD may be out into the market before OSFP which is usually an advantage. Custom OBO modules are already being used in some market segments with standard OBO modules may become prevalent in 2018. For mostly telecoms and OIF-based telephony interfaces already using the 400G CDFP and 400G CFP8 modules and optical cables, OSFP uptake may lessen the volume use of these two older interconnect packaging systems, though some have switched to using QSFP-DD also.

As the OSFP interconnect and QSFP-DD are somewhat similar acronyms and competing interconnect packaging systems with fervent evangelists supporting one or the other connectivity, it is important to be very clear on listening and talking about these systems with prospective OEMs and DC end-users. Even so, the developing COBO inside the box module interconnects (OBOs) may become the preferred packaging and module system versus 50G PAM4 per lane OSFP or QSFP-DD or maybe double stacked µQSFP when 100G PAM4 per lane IO cables become mainstream by 2019.

Visit www.osfpmsa.org for the latest information.