Active cables have a future in data centers
by Stephen Hardy
Among the servers and software on display at last month's Supercomputing '08 Conference in Austin, TX, several optical component vendors added to the expanding portfolio of active optical cable products. The venue was certainly appropriate; high–performance computing environments promise to provide many of the first applications for this nascent optical communications technology. Yet whether those applications represent enough of a market to support more than one or two vendors remains a matter of question.
As reported previously (see "Active Optical Cable Options Expand with Market," April 2008, p. 1), active optical cables are touted as a better alternative to copper interconnect in data center applications where links exceed 10 m and speeds reach 10 Gbits/sec or greater. Current copper and coax links suffer from greater weight and more heat dissipation than active optical cables while failing to match the link distances the optical links can support.
Active optical cables combine optical transceivers with either ribbon or single–fiber cable; the transceivers are mounted on each end of the cable and can be inserted where pluggable copper transceivers (usually either a CX–4 or QSFP) usually reside. To the line card, the active optical cable end looks and acts just like a copper–based transmission device.
Because of the initial focus on data centers, the cables usually come in either Ethernet or InfiniBand formats. The latter proved most popular among the new offerings at Supercomputing '08. For example, Zarlink Semiconductor Inc. (www.zarlink.com) announced commercial availability of its ZLynx dual data rate (DDR) InfiniBand active optical cables with integrated QSFP and hybrid QSFP–to–CX4 terminations.
DDR supports 20–Gbit/sec connections. The next step up, quad data rate (QDR), provides 40–Gbit/sec transmission via four–fiber multimode ribbon cable, with 10 Gbits/sec traveling over each, terminated with a QSFP transceiver at each end. Avago Technologies (www.avagotech.com) lent its QDR active optical cables to the show's Scinet multivendor high–performance network. Meanwhile, Finisar (www.finisar.com) introduced its QDR offering, called Quadwire, which also found a place within Scinet. Finisar has delivered samples to lead customers and plans to be in production during the first half of 2009. Finally, Luxtera (www.luxtera.com) also offered the QDR version of its Blazar Optical Active Cable to the Scinet.
As these companies and others, including Tyco Electronics (www.tycoelectronics.com) and Emcore (www.emcore.com), roll out active optical cables, the question of the potential market size remains.
Jag Bolaria, senior analyst at The Linley Group (www.linleygroup.com), has begun to examine active optical cables as part of his firm's overall study of 10–Gigabit Ethernet. Bolaria points out that in this niche, the new cables face a variety of competitors. Copper alternatives remain a strong choice for links of less than 10 m. Beyond this distance, optical alternatives begin to make sense, he believes. But the cables aren't the only optical option. Short–reach transceivers work well for 30 m, Bolaria says; four–wavelength 10GBase–LX4 and serial 10GBase–LRM devices also are available.
But LX4 devices can be expensive and LRM devices haven't achieved the traction originally envisioned, in Bolaria's opinion. "LRM was supposed to give you this 100 m plus [reach] at a low cost, but there seems to be lots of issues with it," he offers.
This leaves the door open for active optical cabling. "There seems to be this hole, which is if you want to go more than 20 m or so, what's the best way to go do it," Bolaria explains. "And that's where some of these active optical cables come into the picture. Because they can get you up to 100 m without doing anything different on the line cards and the boards that folks have."
However, the advent of 10GBase–T copper–based devices may shrink this hole, Bolaria feels. This means that the real opportunity may come at the next highest data rate, 40 Gbits/sec in the Ethernet space and 20 Gbits/sec in InfiniBand. "Once you start talking about 40–gig, there aren't going to be copper solutions for 40–gig that go beyond 10 m or so. Even the standards guys aren't looking at that. So at that point then active cables become an interesting solution," Bolaria says.
Since the existing DDR and QDR InfiniBand specifications already exceed 10 Gbits/sec, it's not surprising that companies have targeted high–performance computing as a point of entry. Ethernet will follow when the 40–Gigabit Ethernet standard arrives. However, the Ethernet market at this speed will probably take several years after that to ramp, Bolaria estimates. While The Linley Group hasn't yet issued a report on 40–Gigabit Ethernet, Bolaria doubts 40–Gbit/sec port shipment numbers will exceed 10% of those of 10–Gigabit Ethernet by 2012.
Depending upon the pricing, this level of business may be attractive to some companies. "I think there's a space for it," Bolaria concludes regarding active optical cables. "As far as how big that is going to be, the jury's probably out at this point."
Stephen Hardy is the editorial director and associate publisher of Lightwave.