Global TMW:
Login  |  Register          Free Newsletter Subscription
Subscribe
Email
Print
Reprint
Learn RSS

Tests validate 100-m InfiniBand cables

Martin Rowe, Senior Technical Editor -- Test & Measurement World, 10/1/2007

Computing clusters need the highest possible data-transmission speeds between them in order to keep up with the speed of their processors. But computers in clusters, even those that use the InfiniBand I/O standard to communicate, have been limited by the transmission rates in copper cables, which are also heavy and block air flow in computer rooms.

A 100-m fiber-optic cable weighs only 180 g, considerably less than its copper counterpart (1135 g). Courtesy of Intel.
To reduce the bottleneck, engineers at Intel developed fiber-optic cables that are direct replacements for copper cables because they include optical-to-electrical and electrical-to-optical converters in their connectors. The figure shows the difference in size and weight between the red optical cable (180 g) and the copper cable (1135 g).

A demonstration of the cables took place at the International Supercomputing Conference on June 27 in Dresden, Germany. The cluster demonstration included computers from 18 companies such as Dell, Hewlett-Packard, and IBM.

Intel engineers tested the fiber-optic cables with Tektronix equipment to lengths of 100 m—a 10X increase over copper cables—at data rates up to 20 Gbps using InfiniBand technology. Each connection uses four 5-Gbps data lanes in each direction. Each lane moves in a separate optical cable. The cable designers wanted to achieve a 10–15 bit-error rate (BER) with the optical cable. The best BER of the copper cable is 10–12.

To prove the capability of the cables, engineers performed BER tests. An arbitrary waveform generator generated transmission streams and added impairments. “With copper cables running at 5 Gbps, the eye diagram was closed after 5 m,” said Tom Willis, general manager of Intel Connects Cables, in a telephone interview with T&MW. “With the optical cable, we observed almost the same eye opening at 100 m as at 1 m.”

During a test, which lasted 5 min, a sampling oscilloscope measured jitter and calculated BER. Using a “bathtub” curve, engineers could verify that the BER met the specification. Engineers calculated that at 10–15 BER, a 100-m cable would produce 1728 errors per day. At 10–12 BER, copper cables would produce 1,728,000 errors per day.


Equipment used:

Tektronix AWG7102 Arbitrary Waveform Generator
Tektronix DSA8200 Sampling Digital Signal Analyzer
Tektronix Advanced Jitter, Noise and BER Analysis Software

Email
Print
Reprint
Learn RSS

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

Related Content

By This Author

Sponsored Links



 
Advertisement
SPONSORED LINKS

More Content

  • Blogs
  • Podcasts

Blogs

Podcasts

Advertisements





NEWSLETTERS

Click on a title below to learn more.

Test Industry News (3 Times Per Month)
Machine-Vision & Inspection (Monthly)
Communications Test (Monthly)
Design, Test & Yield (Monthly)
Automotive, Aerospace & Defense (Monthly)
Instrumentation (Monthly)
Resource Center E-Alert (Monthly)
©2008 Reed Business Information, a division of Reed Elsevier Inc. All rights reserved.
Use of this Web site is subject to its Terms of Use | Privacy Policy
Please visit these other Reed Business sites