Two form factors are vying for supremacy in the rapidly developing 400G optical transceiver space, the OSFP and the QSFP-DD. OSFP stands for Octal Small Form Factor, and QSFP-DD stands for Quad Small Form Factor – Double Density. Each of these device categories are defined in Multi-Source Agreements (MSAs), as follows:
|FORM FACTOR||TITLE||ISSUE DATE|
|QSFP-DD||QSFP-DD Hardware Specification Rev. 4.0||September 18, 2018|
|OSFP||Octal Small Form Factor Pluggable Module Rev. 1.12||August 1, 2017|
To predict potential commercial success, it is interesting to note the major switch vendors’ involvement in the development of these MSAs. As indicated in the “Points of Contact” in the OSFP MSA, Arista Networks was heavily involved holding Editor, Associate Editor, Chair and one Co-Chair position. Rounding out the OSFP PoC were Finisar and Amphenol. The QSFP-DD MSA was Co-Chaired by Cisco who was joined by a Co-Chair and Technical Editor from Molex. Cisco demonstrated a QSFP-DD transceiver at the OFC2018 conference while there is no mention at all of the OSFP form factor on Cisco.com. Arista, on the other hand seems to be hedging their bets, implementing switching solutions supporting both OSFP and QSFP-DD devices. Another major vendor, Juniper Networks, was involved in the definition of the OSFP but has, so far, announced plans to design only QSFP-DD ports into its new 400G level switches and routers.
Both form factors primarily utilize eight 50G lanes and Pulse Amplitude Modulation with 4-levels (PAM4). They also both include a full range of Optical Physical Media Dependent (PMD) types, including, DR4, SR8, FR4, etc. (these will be described in future blogs) and several optical connector types, Duplex LC, MPO-12 and the new Dual CS interface (shown above).
While the two form factors look quite similar, the OSFP is substantially larger in all three dimensions resulting in a package consuming a bit over three times the total volume of the QSFP-DD. Notice the QSFP-DD in the figure is followed by “(type 1)”. The QSFP-DD MSA defines 2 packages, the Type 2 is up to an extra 15mm longer but all the length is outside of the cage receptacle so does not affect the comparison from a switch port density. The idea behind the OSFP’s larger package was twofold. One, to allow more internal space for electronic components and, two, to provide an integrated heatsink scheme. While certain high-power transceiver types like long-haul DWDM may well require heatsinking, the OSFP burdens all interface types with its integrated design. The QSFP-DD is more flexible, leaving the choice of heatsink or no to the host device design. Equipment designed for high-power transceivers can integrate additional heatsinks to the QSFP-DD cages as necessary. The QSFP-DD is also more flexible with regard to backward compatibility. The QSFP-DD is identical dimensionally with the QSFP and its edge connector is designed in such a way as to allow QSFPs to be supported in QSFP-DD slots without and adaptor. At the very least that allows vendors the potential to upgrade existing 100G class products to 400G without front panel re-designs.
Both OSFP and the QSFP-DD MSAs define multiple “Power Classes” for their respective form factors. Each class defines the instantaneous, peak and steady state current on the 3.3Vdc inputs. As shown in the following table, the OSFP has five defined power classes and the QSFP-DD has eight.
|Power Class||Curr. (A)||Power (W)||Curr. (A)||Power (W)|
|8||6 (max.)||19.8 (max.)|
The idea of Power Classes is to allow various host equipment vendors (producers of equipment with 400G ports) to specify with which class of transceiver a given switch or module may be equipped. It may also be used to limit the number of a specific class may be equipped. For example, a manufacturer may specify that all 32 QSFP-DD 400G ports may be equipped with fully equipped with Power Class 1-4 but equipping with Power Class 5-8 modules must be limited to 50% of ports and in non-adjacent slots. This allows for equipment targeted for exclusively for short optical reach applications (e.g., TOR switch) to be designed for very high port density but limited heatsinking and simple, low cost air handling.
Many in the industry are skeptical the smaller, non-integrated heatsink QSFP-DD package will be able to support high-power applications such as long-haul DWDM with coherent detection until substantial integration enhancements are achieved. Both power and size constraints make this a reasonable position. In any event, at least as far as the MSA definition, both modules are defined with power options to cover essentially all 400G application classes.
Currently, the QSFP-DD module form factor is the industry’s smallest 400GbE module providing the highest port density per Rack Unit (1RU = 1.75” vertical). It has proven in technology demonstrations to support all the expected high-volume 400G applications. With Cisco fully behind the QSFP-DD and Arista, originally behind the OSFP, with a foot in each camp at this point, market forces are aligning behind the QSFP-DD.
Most of the major switch vendor collaborators participating in the OSFP MSA Group (e.g., Arista, Juniper Huawei) have announced products or plans to support the QSFP-DD. It is starting to appear that most of the 400G market will be in the form of QSFP-DD products. The primary play for the OSFP package may be in long-haul transmission products on the WAN side, where DWDM/Coherent technology may require the size and heat management of the OSFP. In addition, products targeting 600G and 800G may also get behind the OSFP. It is difficult to envision 16 x 50G or 8 x 100G PAM4 being accommodated in a QSFP-DD package within the next few years. Such forward thinking vendors may want to abandon the QSFP backward compatibility for the package that can definitely support 400G through 800G.
A similar trade-off occurred at 10G between the XFP and X2 packages and the SFP+. The X2 and XFP were substantially larger and could dissipate more heat and both achieved substantial deployment. However, the SFP+ was backward compatible with the SFP and, with the help of external heatsinks, could soon support most 10G applications and now totally dominates the 10G space.