Tech Brief: HP ProCurve 10-Gigabit Media Alternatives

2
10GBASE-CX4
Because 10-Gigabit Ethernet (10-GbE) had only optical
solutions, the cost remained very high and the
adoption low. To ease this situation, in 2002 ProCurve
Networking drove the development of 10GBASE-CX4,
an extension to the existing XAUI interface, through
the IEEE standards process in record time. “CX4”
provides a very low-cost interconnection for wiring
closet switches and aggregation within a data center.
While very inexpensive, it has a limited range of 15 m
and requires prefabricated cables, which will limit its
application to stacking and aggregation within short
distances.
10GBASE-T
The IEEE is made up of engineers, and everyone knows
that a group of engineers cannot be stopped by mere
physics alone. As a result, the 10GBASE-T project to
provide a solution over unshielded twisted pair (UTP)
was also born in 2002.
While originally touting the ability of 10GBASE-T to
support Category 5 (Cat 5) cable, the IEEE has been
forced to face reality. The IEEE realized that only
limited distances on Category 6 (Cat 6) cable can be
supported, and a new Cat 6A cable has been defined to
ease the effects of “Alien Crosstalk,” a form of
interference that comes from cables that are located
near the link being used.
The IEEE was sold on 10GBASE-T, and the project began
with a lot of talk about scaling CMOS process
geometries down and the ability to use DSP to lower
the cost and power by leveraging “Moore’s Law”—
which says the cost of complexity in CMOS devices will
drop rapidly as time goes on. Unfortunately, Moore’s
Law, like most elements of this physical universe, has a
limit. CMOS transistors require around 1 V to operate
properly. Because CMOS processes at 0.13u already
operate on 1.2 V supplies, 10GBASE-T won’t be able to
reduce its power projections as rapidly as 1000BASE-T
did, because it can’t plan on reducing power supply
voltages by a factor of four. 10GBASE-T is looking at a
brick wall in power reduction, and, if so, its ability to
obtain port density and thus reduce cost-per-port is
going to be limited.
The figure below demonstrates that 10GBASE-T is
unlikely to achieve Moore’s Law with regard to the
benefit of geometry reduction. While more gates can fit
into a given package, the power dissipated by that
package will drive its cost, size, and, ultimately, the
number of ports in a switch that supports it.
The 10GBASE-T specification was finalized in June
2006, but the technology is difficult to implement.
Early products with power requirements on the order
of 10 W per port appeared in late 2007.
Enterprise data center with 10-GbE backbone
15 W
6 W
0 W
Power
1000BASE-T power
10GBASE-T power
20%
10%
10%
20%
50%
33%
Rate of power reduction from geometr y shrink
.35u/3.3 V .18u/1.8 V .13u/1.2 V .09u/1 V .05u/.9 V
Geometry/Voltage