Sunday, April 13, 2008

Cisco cool on 10GBase-T

I got an interesting reality check on 10GBase-T from Ethernet giant Cisco Systems last week. The company's hot new data center spin-in, Nuova Systems said it will not use the standard for 10G over 100 meter copper in its latest switches.

"10GBase-T is great in terms of compatibility and simplicity, but the additional power, cost and latency means it is not really feasible for us and I don't think we will use it," said Dante Malagrino, director of product marketing at Nuova for a story that should be posted to EE Times shortly.

Instead Nuova/Cisco will use a hybrid solution based on a new copper cable terminated by SFP+ optical transceivers which it claims has lower power and latency than the 10GBase-T options. Malagrino pegged 10GBase-T at 2-3 microseconds in latency and 4-8W per link in power consumption. The new hybrid option, one of several alternative cables to emerge in the last year, will initially be limited to 1, 3 and 5 meter lengths but could shave 30 percent off the overall costs of an optical fibre link, he said.

SolarFlare Communications, one of three startups working on 10GBase-T transceivers, is announcing a single 65nm CMOS chip that transmits 10G over copper up to 100 meters while consuming 5.5W. But the chip is just back from the fab and has not yet passed testing.

Competitors Aquantia and Teranetics may follow suit with similar products before the end of the year. I've yet to hear anything from established players such as Broadcom and Marvell. (If you know something, drop me a line at rbmerrit@cmp.com)

SolarFlare claims 10GBase-T switches, aggregation boxes and server cards are in the works, some of which will ship before the end of the year. But the Cisco comment makes me think the new crop of transceivers while major accomplishments in design still may not extend very far the reach of this emerging market.

4 comments:

Anonymous said...

I expect better here. The comparison is based on different distances therefore invalidating the conclusions drawn. 100m is certainly a significantly longer distance than 1-5m. Is there any wonder that the power consumption would be different as well as the cost? Cost needs to be examined in terms of $ / Gbps / meter to normalize to a common metric and draw any credible conclusions.

Cisco has an agenda to push their hybrid technology likely because they charge a good deal of money for such cables - 30% savings over existing fiber solutions seems rather abysmal to me but may be to their customers, it is a big price break.

SolarFlare has an agenda to push their copper technology. Why they didn't or perhaps it is in a separate product take advantage of the IEEE specification to support shorter distances less than 30m and therefore cut power significantly is a good question to ask.

You need to dig deeper and pose the tough questions and insist on using the right metrics to cut past the marketing hype and provide your readers with credible information.

Rick Merritt said...

I have to admit I forgot to factor in the distance issue. My bad.

However, please note that in the full EET story, IBM had some good albeit generic things to say for 10GBase-T, and the Cisco/Nuova exec could have said "we love 10GBase-T but just not for this particular app."

Anonymous said...

You should have been to the CAC workshop at the IPDPS conference on Monday.
http://www.c3.lanl.gov/cac2008/

The panel session covered this item (with an Nuova guy) as well as lots of interesting things such as Ethernet/HPC convergence that you often raise on your blog.

Rick Merritt said...

I have been busy with my own conference this week (see http://www.eetimes.com/news/latest/showArticle.jhtml;?articleID=207400128).

Can you say a little bit about what you got out of your event?

 
interconnects