A polarizing debate ends with agreement to bring both high-speed standards to fruition.
By ED CADY
In late July in San Francisco, the IEEE 802.3 Ethernet Plenary and Higher Speed Study Group (HSSG) meetings were held in a hotel close to the Embaradero Piers and the World Trade Center. This series of standards meetings was particularly tense due to several months of contention between different industry subgroups that have been voting either in favor of, against, or abstaining from voting on, the 40-Gbit/sec Ethernet proposal that has been going along with the 100-Gbit/sec Project Authorization Request. A 75% majority affirmative vote had seemed an impossible feat.
Committee members went into this meeting with a “We are in this together” agreement, but soon, attendees could see the polarization as the presentations and critiques rolled out. But surprisingly, rather than having several no-majority votes, no new task force, and no new specs, each competing subgroup voted to have both 40-Gbit and 100-Gbit spec developments authorized. Hopefully, the entire Ethernet community will evolve from mutual hostage voting to co-existence mode, and finally on-the-same-page mode by the scheduled release target dates.
Subgroups carry clout
Multi-mega-data center end-user corporations are the new 1,600-pound mammoths that have gotten everyone’s attention. This new and growing Internet-based market segment could end up having its own consortium and standards someday. Its members support 100-Gbit/sec Ethernet going on a faster track using all possible resources.
These members would also like to see the start of 1-Terabit Ethernet work. It appears this group has little, if any, plans for 40-Gbit Ethernet for now. They currently use a lot of 10-Gbit CX4 copper links for intra-data-center ports and considerable long-distance LX optical links.
Using a 4 × 25 = 100-Gbit four-lane solution appeals to these end users. Some of their equipment sources question their forecasted uptake of 100-Gbit products, and the group seems architecturally committed to satisfying with “times 10” performance jumps.
Computer, storage, and datacom OEMs have enjoyed a relative Camelot convergence of 10-Gbit IO interfaces. Now, they and a larger part of their combined customer bases, seem to be looking forward to a 4 × 10 = 40-Gbit convergence with InfiniBand, Ethernet, Fibre Channel, and maybe other standards.
These members support a faster-track development of native 40-Gbit Ethernet specifications and products. They have disdain for the implementation of 40-Gbit Link Aggregation products as increasingly complex and costly versus the proposed native 40-Gbit Ethernet. 100-Gbit Ethernet is on their roadmaps as well as 40-Gbit, 80-Gbit, and 120-Gbit InfiniBand. CX4 copper cables and backplanes will need to go from 3.125-Gbit to 10.5-Gbit per lane.
Telecommunications OEMs and carriers appear to be primarily interested in 100-Gbit Ethernet development. Those that have certain server and storage system elements seem also to support 40-Gbit Ethernet. The ITU’s (International Telecommunications Union’s) OC-768 40-Gbit SONET/ATM standard has been used for some time with all these segment members. The next ITU standard is for 160-Gbit, as they use a “times-4” performance jump. This particular group used CX 4 copper cables for 4 × 3.125 = 12.5-Gbit Ethernet as well as customized stack-plane backbone interconnect links.
Most suppliers of chips and cable connectors and assemblies apparently need to support 40-Gbit and 100-Gbit Ethernet with two new product family developments, or a wider range of multi-data-rate products. Suppliers hope for and promote using the same four-lane technologies, performance range, and products for other industry standards and semi-custom applications.
New connectors, assemblies
Intel has announced a new silicon optical modulator that encodes data at 40-Gbits/sec. Several of these modulators and new 40-Gbit silicon lasers can be integrated on one chip to achieve Terabit-level transmission.
The new 30% shorter LC optical connectors are getting a lot of attention for better cable-management possibilities.
CX4 copper cable assembly suppliers are developing 4 × 6.25, 4 × 10, 4 × 12.5, 4 × 20, and 4 × 25 performance-range products. Some semiconductor manufacturers probably need more 4 × 3.125 product return-on-investment before offering more than a 4 × 10 new product within a year.
10GBase-T Ethernet twisted-pair cable assemblies use Category 6A or Category 7A RJ-45 connectors, with each of the four lanes running at 2.5-Gbits. RJ-45 connector suppliers have been claiming a theoretical 10-Gbit per-pair maximum performance capability. Are they developing such a connector for potential horizontal 4 × 40 = 4-Gbit Ethernet links, and if so, what transmission distance will these links support?
8G iSCSI (SCSI storage over Ethernet) performance products are coming out with 2-Gbits per lane with LC, RJ-45, or CX4 cabling implementations.
Single-lane 10.5-Gbit Ethernet is used with new SFP+ copper cabling, and four-lane 12.5-Gbit Ethernet is sometimes used with new QSFP copper cabling or OE modules for very long distances.
Observations, conclusions
Liaison work with the OIF’s CEI-25G subcommittee is a likely activity for the new Ethernet task force that will be sizing up 4 × 25 = 100-Gbit links. Will 80-Gbit bridge generation or link aggregation products become part of the picture? Some Ethernet liaison work with the InfiniBand and potential Fibre Channel 40-Gbit standards bodies seems to be likely.
It appears that when a very successful, and large, standard expands into many market segments, a growing chance of dramatic schism can occur due to disruptive market trends and new market leadership players. The positioning of competing technologies, design elements, test data, test methodologies, and building consensus will be a key part of the specification development and release.
The nascent PCI-Express Rev. 3 specification is said to be going with 8-Gbit versus 10-Gbit per lane performance, even though there is no real consensus within the polarized standards committee. Because it does not employ the commonly used (InfiniBand and Fibre Channel) 8b10b code, there is concern about how this could affect Ethernet 40- and 100-Gbit implementations in future server-centric applications.
The new PCI-Express architecture is targeted to maintain the same real data payload as current 10-Gbit interfaces, so maybe the proposed handshaking will work. But it might add cost or complexity, or adversely affect reliability.
The somewhat divided Ethernet committee is embarking on a mixed-agenda process; hopefully, members will reconcile the 40- and 100-Gbit specifications, technical issues, and interoperability while continuing to hope that their corporations will successfully achieve their desired share of worldwide trade.
ED CADY is market development director with Meritec (www.meritec.com).
