With 400G switches just under release, the industry is already seeing two form factors in competition. What does this mean for the network operators? Why are these two form factors in competition? Are they really in competition?
While there are multiple form-factors for 400G bitrates, the QSFP56-DD and OSFP are clearly in competition for all “access” and “data-centre” applications.
QSFP-DD and OSFP Specifications
QSFP56-DD (QSFP56 Double Density – 4x (quad) 50G x2 (double density) = 400G)
The QSFP56-DD (also called 400G QSFP-DD), is standardised by the QSFP-DD MSA Group. It has a similar size compared to the QSFP transceiver (QSFP+/QSFP28/QSFP56) but it has eight data lanes instead of four. The QSFP-DD port (the cage) is backward compatible with the QSFP transceiver and is compatible with a QSFP28 (if the switch supports it).
While there is no direct thermal management mechanism (no integrated heat-sink), the QSFP-DD allows a maximal power consumption of 12 watts (initial spec), challenging the transceiver industry to release a 400G technology not exceeding that. Recently, the QSFP-DD MSA Group has published an application note confirming the 15W feasibility, while the last specification (in its version 4.0) defines a Power Class 7 (up to 14W) and a Power Class 8 (>14Watts).
The OSFP (Octal SFP – 8x 50G =400G) is standardised by the OSFP MSA. Its size is slightly bigger than a QSFP module and is designed for 8 data lanes, reaching 400G bitrates (with 50G PAM4). It has a direct thermal management (integrated heat-sink) and allows a maximum power consumption of 16 Watts, relaxing the requirements for the transceiver industry. The OSFP cage backward-compatibility capabilities toward QSFP modules is only feasible with an adapter.
Arista Networks, a promoter of the OSFP, also believes that OSFP can go up to 20 Watts thermal capacity, opening the door for coherent DWDM transmission.
The key names behind the two form factors
Interestingly, key companies are behind the QSFP-DD MSA Group and OSFP MSA, while some are supporting both standards:
At the release date of this article, some switch vendors have announced/released “400G ready” switches with either QSFP-DD or OSFP ports*:
- Arista – 7060DX4-32 supporting OSFP and 7060PX4-32 supporting QSFP-DD
- Cisco – Nexus 3K and 9K supporting QSFP-DD
- Edgecore – AS9716-32X supporting QSFP-DD
- Huawei – NetEngine 9000 – 8*400G-QSFP-DD supporting QSFP-DD
- Juniper – PTX10003, QFX 5220 and QFX10003 supporting QSFP-DD
- Nokia – 7750 SR supporting QSFP-DD
* the list is not exhaustive and is likely to change rapidly and frequently.
1U Rack – 32 or 36 ports?
Both QSFP-DD and OSFP remain “compact” for access and data-centre applications. However, a 1U rack can handle maximum 32x OSFP ports, while the same rack can handle 36x QSFP-DD ports. More density with QSFP-DD is available, but also brings more challenges in term of power dissipation.
The 800G bitrate (8x 100Gbps PAM4) will be available on the OSFP form factor as it has been designed for. A QSFP-DD, called QSFP112-DD, will potentially be released for 800G application. It will have the same dimensions as QSFP56-DD but the electrical interface will be optimised to support 112Gbps bitrate (100G PAM4) per-lane.
Trends vs reality
While a trend is clearly visible, it is still too early to define a winner. The QSFP-DD has a real advantage of being natively backward compatible with the QSFP form factor, while the OSFP seems to have more robustness (in terms of heat dissipation, space for components and 800G ready).
Except for customers of Arista, the network operators will decide the form factor when selecting the switch vendor, and it will be QSFP-DD. At this time, only Arista has released a switch with OSFP ports. However, this situation can drastically change. As switch vendors are in the early stage of releasing the 400G switches, we have seen only few demos and limited announcements. The first half of 2019 will be key to observe the industry to confirm or infirm the actual trend.
Do you have doubts about moving to OSFP or QSFP-DD? As the OSFP has more space and less power restriction, the transceiver industry will use this form factor for the first instances of complex technology (more than 2km, DCI application, etc.) but will rapidly support the same technology on a QSFP-DD. The early-adopters might prefer the OSFP for this reason.
If we follow the trend, it seems that switch vendors are going for QSFP-DD, even Arista is coming up with a dedicated version; set to include key features such as backward compatibility, lowest power consumption and smallest form factors; the key drivers for a long-term adoption.