Breaking Ground On 25GBASE-T
Late last year we saw a Call for Interest (CFI) take place on 25GBASE-T at the IEEE 802.3 Ethernet Working Group meeting in San Antonio, TX. With 48 contributors and supporters, including representatives of Microsoft, Intel, Cisco, and Siemon, the objective was to gauge the interest in forming a study group to investigate a 25GBASE-T project. IEEE 802.3 member and study group contributor, Valerie Maguire of Siemon, shares her insight on how this potential new Ethernet standard was considered for study and how it is progressing now.
In San Antonio, IEEE 802.3 Ethernet Working Group members considered the reasons for 25GBASE-T most carefully, from understanding the market drivers and answering the question “Why now?” to evaluating the technical feasibility. This was a thorough examination and it resulted in attendees voting overwhelmingly in favor of the formation of a 25 Gb/s Study Group, with 37 individuals, representing 25 companies, volunteering to participate and pledging support. So what led to this green light?
The deployment opportunity for 25GBASE-T was first clearly defined as being in the 30m reach zone, sitting between 10GBASE-T as the top end of the typical enterprise requirement and pending 40GBASE-T typical in the data center. The market drivers considered were trends for cloud vs. enterprise servers, the latest update on server port speed forecasts, the media mix, potential topologies, and cost optimization.
In predicting the future trend towards cloud server growth, the Dell-Oro Controller and Adapter Report of July 2014 was shown to forecast that cloud and enterprise server shipments will equalize in volume by 2018, with a quarter of server units shipping in 2014 already destined for cloud deployment. The belief is that cloud computing will grow fast before stabilizing, with a small number of entities comprising the majority of this market.
Market analysis of the speed of application and storage migration to cloud servers shows that transition to 10 Gigabit Ethernet speeds supported by optical fiber, direct attached twinaxial, and balanced twisted-pair copper cabling is nearly complete throughout data centers and 40 Gig and 100G deployments are becoming more common. 25GBASE-T is predicted to be an attractive and cost optimized step on the speed migration path to 40GBASE-T for support of data center edge connections.
Meanwhile, on premise servers, 1000GBASE-T is still predominant, with 10GBASE-T growing and 40GBASE-T predicted for significant future growth. Here, 25GBASE-T would be considered for deployment only once standardized and multi-vendor, off-the-shelf hardware is available and throughput and reach needs are assessed.
The server port type mix was predicted by Dell’Oro to show continued growth for 10GBASE-T as transition progresses from 1000BASE-T in the enterprise data center. While direct attach QSFP+ and SFP+ copper and optical fiber connections continue to have a place, connection growth forecasts suggest that balanced twisted-pair copper 10GBASE-T will continue to dominate overall.
In considering data center topologies, it was acknowledged that there is no single end-all cabling configuration for every facility and the “right” choice should be dictated by architectural choices. While top of rack (ToR) switch to server connections may be right for some users, for many applications and facilities middle of row (MoR) and end of row (EoR) switch to server connections enable improved cost efficiency, better space utilization, and support for mixed applications. A 25GBASE-T PHY will have the reach to support a much broader range of architectures than direct attach twinaxial connections to easily facilitate all types of cabinet to cabinet, MoR, and EoR switch to server structures.
Cost optimization was a key heading in the CFI discussion and 25 Gb/s transmission was identified as a “sweet spot” to optimise both port count and total bandwidth for server interconnects. Furthermore, it was noted that existing single 25 Gb/s technology for other media types can easily be optimized and re-purposed for 25GBASE-T support within transceiver equipment.
Summarizing the market drivers, the conclusion put forth at the IEEE 802.3 Closing Plenary meeting was that 25GBASE-T provides a useful roadmap from 1 Gig to 10 Gig and ultimately 40 Gig on a compatible infrastructure. It could enable optimized switch port usage over a broad range of server to switch architectures, enable large, modular switches and higher port counts, plus allow ports to be utilized that might otherwise be constrained by the 5m reach limited of direct attach twinaxial solutions.
In considering the technical feasibility of 25GBASE-T, it was observed that the application could build on the well-established technology for 10GBASE-T that exists today and that is evolving to support 40GBASE-T. A comparison of 10GBASE-T, 40GBASE-T, and likely 25GBASE-T transmission parameters, including baud rate, cancellation (echo, NEXT loss, and FEXT loss), and relative signal-to-noise (SNR) margins demonstrated that the complexity of 25GBASE-T over 30m of cabling is roughly the same as 10GBASE-T over 100m of cabling using known capabilities.
Following this convincing comparison, it was stated that the 25GBASE-T application would be targeted for operation over the same 2-connector ISO/IEC Class I/Class II and TIA Category 8 channels currently defined in Draft 1.1 of the IEEE P802.3bq 40GBASE-T amendment. Presentations in IEEE P802.3bq have shown technical feasibility of 40GBASE-T using 10GBASE-T based signaling at a 3.2GHz symbol rate and, therefore, it is likely that 25GBASE-T will be supported using the same 10GBASE-T signalling at a symbol rate of approximately 2.0GHz. In addition, the existing circuit to physical channel (i.e., PHY) models developed in IEEE P803.3bq for 40GBASE-T can be applied to estimate 25GBASE-T PHY transmitter, receiver and cancellation parameters, SNR margins, and power management.
This question is worthy of careful consideration and responses to it heard at the CFI emphasized that 25GBASE-T is a critical and heretofore lacking point on the migration roadmap to 40GBASE-T as enterprise data centers begin transition from 10GBASE-T to higher speeds. It would allow users to leverage capital and R&D investment in 10GBASE-T and 40GBASE-T technology, while optimizing deployment costs as servers and switches incrementally increase data speeds.
A 25GBASE-T standard is seen to positively fit within the very successful balanced twisted-pair copper Ethernet ecosystem. It shares open and common specifications, ensures interoperability, and provides security for those investing in development. Given the market drivers, technical feasibility, and added value that 25GBASE-T could offer to the industry, it was perhaps unsurprising that the motion to form a 25 Gb/s Study Group was approved without objection at the IEEE 802.3 Working Group Closing Plenary meeting.
THE FUTURE OF 25GBASE-T
The 25Gb/s Study Group completed their task to determine objectives for a 25GBASE-T Ethernet application, which were approved at the IEEE 802.3 Closing Plenary meeting in March of this year. The IEEE P802.3bq 40GBASE-T Task Force has been officially renamed the IEEE P802.3bq 25G/40GBASE-T Task Force and the first revision of the IEEE 802.3bq amendment containing draft specifications for both 25GBASE-T and 40GBASE-T has been created and completed its first Task Force ballot review cycle at the end of October.
It’s worth noting that both 25GBASE-T and 40GBASE-T are planned for operation over TIA Category 8 cabling; the specifications of which are under development and expected to publish as ANSI/TIA-568-C.2-1. Work so far suggests that 25 and 40GBASE-T will work anywhere in the data center over Category 8 cabling at up to 30m, including fat tree, leaf and spine, interconnected fat tree fabrics, anywhere in a classic level 3 hierarchy, for full mesh, interconnected meshes, and centralized switching. Adding 25GBASE-T as an alternative to 40GBASE-T will provide even greater choice and flexibility for future data center deployments. One to watch!