UXPI, the Unified 10Gbit Physical Layer Initiative, aims to produce a common electrical specification for the physical layer of 10Gbit/s networking. Publicly launched in October 2003 by Applied Micro Circuits, IBM, Infineon, Texas Instruments and Xilinx, the goal of this group is to advocate a common 10 Gbit/s physical-layer standard across multiple markets (telecommunications, data communications, computational, storage, etc.) to simplify and accelerate the implementation of next-generation 10 Gbit/s serial networking systems. Agilent, the testing company, has just joined.
Why does the industry need UXPi? The organisation says that the trend toward serial connectivity is being driven by the need to reduce system costs, simplify system design and provide scalability to meet new bandwidth requirements. UXPi members have been strong supporters of the work within the OIF (Optical Internetworking Forum) on the development of the CEI (Common Electrical Interface), which includes 6 Gbit/s and CEI 11 Gbit/s high-speed backplane standards. UXPi's vision is to see such development occur in parallel coordination in numerous standards venues. UXPi interfaces will provide a basis for future high-bandwidth connections in chip-to-chip, chip-to-optical module and backplane applications.
High speed serial I/O is an inevitable event. If there is no single unifying physical standard specification, then interoperability between various semiconductors will become a major problem and system designers will face a huge dilemma when specifying components for a system implementation. If a single unifying standard exists, the physical layer problem is solved and IC and system manufacturers can concentrate their efforts on adding their particular value add rather than worrying about which serial standard to support.
No storage vendors in UXPI
Despite the applicability of this to storage networking there are no storage networking companies listed on the UXPI membership roster, not Brocade, not McData, nor Emulex, Adaptec, QLogic, etc. Perhaps the timing is too early?
Brandon Hof, senior advanced technology marketing manager for McData says that, "It will take about two years for the standard to be ratified. Then ASIC vendors need about two years to integrate it into their products. Then systems vendors need about two years again." So we might not see product using this standard for 4 - 6 years.
Is it a worthwhile standard?
It is a worthwhile standard according to Hof, "a serious and substantial effort. It will enable vendors to use different suppliers' chips to build their products. This is a very worthwhile exercise; 10Gbit/s is so fast. But just running ten gig connections across a server board is very difficult in itself."
"Take HBAs operating at 10Gbit/s. These connect to a server motherboard at 2-4Gbit/s with PCI today. We only get a percentage of their capability. In the future they'll be able to connect through at 10Gbit/s. It will be very useful for things like Infiniband also. We will see lower complexity and lower chip counts running across the backplane. Customers will see lower cost and higher performance."
Here is the perspective of an HBA vendor; Chitra Vivek, director of product marketing for Adaptec, commented: “UXPi looks like an interesting development and one we would certainly, albeit indirectly, support if it did become a standard. However, it’s not something we’d get directly involved in. This is because Adaptec supports TCP/IP offload which is a Layer 2 and above activity; Adaptec is not at the physical layer that the UXPi organisation is focused on.”
Is UXPI really needed?
Perhaps UXPI is surplus to requirements? Jag Bolaria, an analyst for the Linley Group, writes, "Because XFI (another standards initiative) and the CEI working group at OIF specify 10Gbit/s interfaces, there does not appear to be a need for another 10Gbit/s initiative. Rather than develop another new interface, UXPi intends to unify the different 10Gbit/s interfaces and ensure that the resulting interface accommodates future requirements. If successful, the UXPi group will prevent fragmentation of 10Gbit/s interfaces. A common interface enables vendors to develop one piece of silicon, reducing development costs. For example, UXPi could influence adoption of a more-efficient encoding scheme."
"In an ideal scenario, UXPi would serve as the steering committee for all 10Gbps physical-layer interfaces. It remains to be seen, however, if UXPi can effectively steer these various specification efforts or if it is merely a backseat driver."
What is being specified is the physical/electrical layer. This means voltage levels, impedances, transmission distance, jitter etc. Everything it takes for ICs to be able to send signals that can be read and understood by an IC on the other end. In specifying the electrical layer, the actual channel will end up being specified as well. There will be actual technical specifications that dictate what channels can be transmitted across.
The UXPI initiative specifies NRZ (non-return to zero) signaling. Companies such as Accelerant Networks and Rambus already have 10Gbit/s product up and running using both NRZ and an alternative four-layer pulse amplitude modulation (PAM4) approach. They are not members of UXPI.
If the UXPI initiative succeeds then we will see working 10Gbit/s product sooner and can be more confident that different vendors' 10Gbit/s products will interoperate properly.