Friday, June 29, 2007

It's all about Ethernet

I love the little reality checks that bubble up from the field. Here's one in my email box this morning from a source who has been around the block a few times but asked not to be named:

"In 2001 the 'fabric wars' were going full blast. By 2007, it’s clear that Ethernet is the overwhelming winner, with Rapid IO being used for a few specialized, DSP-heavy applications as Rapid IO is popular as a DSP interface."

A crisp synopsis of this sector in the last six years. Of course even a little success can be a lot for a small movement, as I noted in an article for the RapidIO newsletter out this week.

Meanwhile the big action is in Ethernet. The growing crowd looking to hammer out the next big step in data rates appears to have hit a stalemate, but the chairman of the IEEE High Speed Study Group is optimistic of a breakthrough when the group meets again in July. Stay tuned.

Thursday, June 28, 2007

ATCA backers sound off

My email has been humming with chorus of ATCA proponents weighing in on last week's EE Times story (and my post below referring to it) that said support for ATCA is cooling, especially among big telecom systems companies. I invite anyone with specific facts on what is happening inside board, system and end-user companies to make a comment on this post.

A manager from Nokia Siemens said his company is actively behind ATCA with products and work on the standards, a comment I have already put in an update version of the EE Times story and blog postings.

Dick Somes, technical officer of the PICMG group that defines ATCA, took issue with the claim the standard has too many flavors. That's not fragmentation, but flexibility, Somes said in an email.

Lance Leventhal who helps edit the ATCA Newsletter said that outlet has published several interviews with big telecom systems companies speaking favorably about ATCA, including a Nortel interview a year ago.

One source who asked not to be named said the sources dis-ing ATCA in the EE Times story are focused on the North American market which has been fairly flat. Most of the ATCA growth these days is in Asia, Europe and South America, he added.

Others chiming include a manager from Kontron who said "there have been several public statements of Telecom Equipment Manufacturers such as Alcatel-Lucent, Nortel Networks or NEC who have endorsed the ATCA standard." However, "this industry thinks in 5- to 10-year cycles rather than in quarters" so actual products are just now coming up for release in some cases, he said.

The rise of new groups such as the Communications Platform Trade Association trying to drive standards and working with ATCA is another sign the standard will gain traction he said. The group includes Hewlett-Packard, Motorola, Nortel, Sun and China's ZTE.

Thanks to all for your emails and, please, make a comment here with all the specifics you have to offer. They may help drive future stories.

Wednesday, June 27, 2007

Like fries with that?

So now Intel sells optical cables. Oh, the things you have to do to enable the market for x86 processors!

It turns out most end users buy 24-gauge copper cables that reach just 10 meters for data center computer clusters. But the new Intel Connects Cables with optical transceivers built into their connectors have an 100-meter reach. That enables multi-floor clusters extending today's 1,000 CPU systems to clusters with potentially tens of thousands of x86 chips.

When you are the world's largest semiconductor maker and have essentially one cash cow product, you wind up offering some strange fish to sell your chips.

Turning on 60 GHz radios

The WirelessHD group has its specification for 60 GHz radios for short range, high bandwidth home networking out for final review. It should be released within 60 days along with the names of new semiconductor companies now members of the group.

That was the news from an interview Tuesday with SiBeam, a startup formed to deliver CMOS chips based on the technology. SiBeam helped start the WirelessHD group which was announced last November.

The group was supposed to announce its spec in the spring, but the effort was delayed a few months. "All these standards efforts are like herding cats but the spec is essentially complete," said John LeMoncheck, chief executive of SiBeam.

The startup has a demonstration of a working 60 GHZ radio transmitting 1080-progressive video wirelessly over about 10 meters. It is based on an FPGA since the spec was too recently finished to support an ASIC implementation.

The 60 GHz spec will define two networks that work in tandem. A primary beam-forming 60 GHz network will deliver multiple Gbits/second of bandwidth for transmitting uncompressed high definition video between devices within a room.

A secondary omni-directional network capable of "tens of Mbits/second" will continuously look for and negotiate with devices within its 10 meter range. It will use device discovery techniques particular to the 60 GHz spec, but apparently will not interfere with any existing mechanisms supporting other device discovery techniques such as those defined by the Universal Plug and Play Forum.

LeMoncheck said SiBeam will be a fabless semiconductor company, focused on generating revenues from sales of its own 60 GHz CMOS chips. Any intellectual property royalties it generates will be modest or negligible, he said.

LeMoncheck was formerly a senior executive from Silicon Image which holds IP rights on the High Definition Multimedia Interface. HDMI requires royalties of about four cents per port, a charge some users have protested.

"It is not part of our business model to make significant money on IP. We want to compete in chip implementation," he said.

The startup clearly has the firepower to focus on chip design. It has hired two senior technical executives from Wi-Fi chip designer Atheros as well as another from Silicon Image. "I am really proud of the team we have assembled," said LeMoncheck.

The team includes George Palmer, former vice president of operations for PortalPlayer which made chips that powered many of the early models of the iPod. "He knows how to turn on a million units a month for big customers such as Apple," said LeMoncheck.

Other members of the WirelessHD group will deliver their own merchant silicon for 60 Hz radios. They aim to compete in bandwidth, ease of use, cost and power favorably with ultrawideband companies such as Pulse~Link, Tzero and Alereon which also aim at high-bandwidth wireless home networking applications.

SiBeam expects to announce its OmniLink60 products at or before the Consumer Electronics Show in January.

Tuesday, June 26, 2007

SGI carves Infiniband into blades

Struggling server vendor SGI hopes to distinguish itself by using Infiniband natively in its Carlsbad server blades released today and geared for high-performance clusters. The move makes sense because Infiniband is growing in such technical applications, but given SGI's modest role the move will not likely have much market impact.

Still it is interesting to note that Intel engineers who co-develop some of the SGI boards were the ones who influenced the move to Infiniband. IB was a big Intel initiative many years ago until it became clear the market for the interconnect would be relatively limited and Intel dumped plans to make the chips. Nevertheless it still clearly promotes the technology, perhaps out of love for this orphaned child and its evergreen belief in making everything in computers a commodity standard—except its high-priced CPUs.

Monday, June 25, 2007

ATCA's failure points to blades

It looks like the big telecom OEMs are not interested in a move to standard chassis and backplanes, according to an excellent report in the EE Times print edition today from my colleague Loring Wirbel. "Full-sized ATCA may end up being a highly customized boutique business, while high-volume manufacturing moves to the smaller MicroTCA standard," Loring wrote from NxtComm in Chicago.

Board makers told Loring the handful of big telecom OEMs seem to be interested in all sorts of open interfaces and software tools, but not the ATCA standard. Now they are hoping the larger group of OEMs making a more diverse set of edge-networking systems adopt the MicroTCA spec, but that may be more hope than reality.

This sounds to me like a replay of what's going on in server blades. Everyone recognizes it would serve the end user best to have a standard size for boards that could fit into any vendor's chassis. But the four top players who control the brunt of the server market have been able so far to keep shipping an ever-changing mix of non-standard boards and boxes to lock users into their proprietary approaches and avoid doing what's right for their customers.

One embedded board vendor told Loring the ATCA failure became clear when Nokia Siemens Networks backed off ATCA following its merger. However, Kai Sjoblom, a technology manager at Nokia Siemens Networks (Espoo, Finland) said in an email that Nokia Siemens Networks "still plans to ship systems based on ATCA."

The company "continues to keep ATCA as a strong part in its hardware platform offering and the merger has not caused this basic plan to change," said Sjoblom who is also a member of the PICMG group that defines ATCA.

Another board maker at NxtComm told Loring the fact there is an ATCA standard for every flavor of backplane interconnect confused and drove away others. One board maker was so glum on ATCA, he said the company decided it will license its ATCA designs and do custom development for others. That's a huge change from just five years ago when the PICMG group behind ATCA was gearing up all its engines to drive ATCA.

Though things look dark today I would not give up on either ATCA or a server blade standard. We are in an era of radical cost savings when no one can afford to re-invent the wheel. Those dynamics could force even markets controlled by a handful of giants to bow to market pressures and common sense.

Interestingly, Intel was a big promoter both of ATCA and a server blade standard. It just wanted to see as many boards and systems sold as possible, no matter what the implications were for OEMs.

Thursday, June 21, 2007

The HP way to 10GE

I've been getting bombarded lately with announcements from one 10 Gbit Ethernet chip or board supplier after another, each claiming some unique edge in this emerging market. To get some perspective on this issue from a major server maker, I talked with John Gromala, a product marketing manager for Hewlett-Packard's x86 servers.

The HP Proliant group is exclusively using startup Netxen's 10GE chips now. I asked John why Netxen, what are HP's criteria and what's the outlook given a dozen large and small companies have production 10GE chips rolling out shortly.

"You have to find a balance between the cost, latency and overall throughput. We don’t have a specific spec or number each of these vendors has to meet," Gromala said.

Specifically, HP likes the fact Netxen supports TCP offload (aka TOE), and does it in a stable way on both Windows and Linux. The two companies have had a long development relationship, he added.

Gromala would not say anything about HP's future plans except that it is evaluating 10G chips from multiple companies. He did say HP has decided not to use the so-called I/O Acceleration Technology that Intel has been putting on some of its Xeon chips as a substitute for TOE.

"IOAT is an example of an approach that has been used before and really just moves the processing from one part of the CPU complex to another. It doesn’t completely move the processing off the CPU and on to the [networking] card," Gromala said. "We don't want [the job of processing] the TCP stack using up all the CPU's horsepower," he added.

HP has been a big backer of iWarp that defines for Ethernet the remote direct memory access (RDMA) capability already in Infiniband. HP is working with Netxen to get RDMA into products, but so far the work is still somewhere in the development pipeline.

"What Netxen offers today is good enough for now but eventually we will bring a full RDMA function [to the products]," he said.

The next big goal for 10GE is to get to affordable options for copper media, Gromala said. HP is tracking the progress of 10GBaseT as well as backplane Ethernet and other options, he added.

My take: Intel needs to revise its 10GE story if giants like HP aren't buying into it. TOE and RDMA will be requirements with backing from HP.

Not incidentally, Sun's Neptune is looking better all the time as a way of carving out a unique approach that is apparently giving it an edge over Intel and its top OEMs like HP. Hey, maybe a big system house still can differentiate its products with some smart silicon development. Imagine that!

Digging deeper into Wibree

Jack Shandle's Wireless Net DesignLine does a good job this week digging deeper into the details of last week's news that Wibree is merging into the Bluetooth group as its new ultra low power (ULP) offering. Jack's article contains links to deep dives on the basics of Bluetooth and Wibree.

The article also rightly points out that ULP is not limited to personal area networks. ULP will be the first link of choice for many sensor network nodes which ultimately will be part of the wide area net. The site also provides a good article on how WiMax and Wi-Fi can coexist with BT in the broader wireless soup.

Wednesday, June 20, 2007

UWB's rainbow maker?

What's it take to get ultrawideband going? Most OEMs and chipmakers I have talked to suggest it is cost, cost and cost. Thus they are choosing CMOS and aiming at an integrated transceiver and media access controller. Their goal is to get below $10 fast.

Startup Alereon is taking a different path to a similar goal. It has moved its silicon germanium transceiver from Jazz Semi to IBM to craft a design that spans frequencies from 3 to 10 GHz, covering the waterfront for global UWB use. Most of the CMOS chips are attacking just the 3-4 GHz range used for wireless USB chips in the US with plans to add on 6 GHz for future Bluetooth sometime late next year.

It will be interesting to see who gets to the goal first. The Alereon AL5000 won't itself be in production until early 2008. Even then, the startup won't comment on what its initial costs will be, but it knows it has to sink below $10 fast.

Tuesday, June 19, 2007

RIO gains traction in the shadow of Express

No big surprise here, but Crystal Cube Consulting released its projections for RapidIO adoption yesterday. The group estimates RIO was used in about ten percent of all DSPs in 2006, a figure that will rise to about 34 percent by the end 2011. A report from In-Stat earlier this year provides more details and a broader view of this sector.

The interface is mainly catching on as a way to link DSP farms in wireless base stations as well as gear aimed at video applications, CCC said. Serial RapidIO also is being used for embedded backplane systems in high-end defense, medical and telecom systems. That's pretty much the story as the RapidIO Trade Association tells it. In fact, I suspect the association may have commissioned the report to give itself credibility.

According to CCC, the upside for RIO is that it is gaining design wins in the wake of the demise of Advanced Switching that was supposed to smooth some of the bumps of migrating PCI Express from computers into multi-host embedded systems. The downside is Express still has plenty of momentum thanks to the hundreds of millions of computers using it.

Thus "some companies are recommending that people put native PCI Express interfaces on their communications chips. If that were to become quite common it could have an impact on the RapidIO market’s continued strong growth, but that remains to be seen," according to the executive summary of the report.

Monday, June 18, 2007

Hard slog in the last mile

Passive optical networks, the next big thing in broadband last mile technology, are under siege, according to my colleague Loring Wirbel in this week's EE Times. PONs have the potential to replace high-speed cooper DSL links, but Loring is skeptical about how many people will reap the rewards of this historic shift from Mbit to Gbit technologies.

Loring notes that PONs are becoming as hotly competitive as Ethernet switches were a few years back. Suppliers and service providers are consolidating, even though the PON technology is still fragmented into multiple camps. The story is being played out where all the parties are gathering at the NxtComm show I Chicago this week.

Friday, June 15, 2007

Nudging HyperTransport

In an effort to keep its Itanium processor in the news, Intel engaged in the all-too-common practice of a little spec strip tease yesterday, disclosing three or four new factoids about the road map of the server CPU. The big deal to me was a little more detail about its plan to replace its CPU front-side bus with a HyperTransport competitor called the Common Systems Interface.

Intel is still not giving details about CSI yet, not even its name. But yesterday it publicly said it will use the interconnect on both a 2008-class Itanium 2 chip called Tukwilla, and at least one Xeon and common south bridge I/O chip in the same time frame. Execs also said the link will have unique reliability and scalability features, probably alluding to the support both for SMP and NUMA systems an Intel server technologist told me about earlier this year.

I'm hoping for a briefing with Intel before the summer is out to fully take the wraps off CSI and see how it compares with HyperTransport defined by archrival AMD. The other big question is whether Intel will license the interconnect to third parties, a move that would make a lot of sense to me but create some real costs and management issues for Intel.

If anyone else has more G2 on this situation, feel free to post what you know here or drop me a line at

Thursday, June 14, 2007

Cisco's $10,000 question

Got a future networking fantasy, a wet broadband dream as it were? Cisco wants to hear about it. In fact they will pay $10,000 for what the company considers the coolest idea for a new network-enabled application or device. For more information, open a hailing frequency to the Cisco Connected Life contest.

Fun with Jack and Chumby

Jack Shandle, veteran tech journalist and editor of Wireless Net DesignLine, is encouraging engineers to have some fun playing with Chumby. In his blog, Shandle calls the device "open, extensible and eminently hackable…a stem cell of an electronic appliance."

Powered by a 266 MHz ARM processor from Freescale, the device sports an LCD, Wi-Fi, audio and USB—even an accelerometer. The collie is apparently optional. The company behind the concept provides all the design resources anyone needs on their web site. It's like being back in the Apple II days, Shandle says. Woof!

Wednesday, June 13, 2007

Intel scales low power I/O

In the limbo dance of the lowest mW per Gbit/second, Intel Corp. hasn't dipped as low as Rambus did earlier this year. However, it has gotten to a significant low with a chip-to-chip interconnect that scales better.

Intel presented at the VLSI Symposium this week the I/O links it uses in its Terascale research processor. The big figure of merit is 2.7 mW/Gbit/s, just slightly north of Rambus at 2.2. But the Rambus technology is fixed for use at 6 Gbits/s while the Terascale I/O can scale from 5-15 Gbits/s. For more see the EE Times story.

Searching for Google

I had an interesting chance encounter with a senior server designer at Google yesterday. He said the company was buying off the shelf boards for its systems up until 2003, but since then has been "designing everything but the chips." Asked specifically about interconnects for big data center clusters, he said Google definitely wants to have in-house expertise in that key area. I hope to delve more into this subject with a future interview to get Google's views on servers, networking and data center technology issues, so stay tuned.

Tuesday, June 12, 2007

Wireless nets get in phase

Two stories of consolidation in consumer wireless networks hit today. Tzero debuts a new ultrawideband chip set geared both for wireless and coax, and Wibree backers join the Bluetooth SIG.

On the UWB side I was surprised to find the WiMedia Alliance has set up a work group to determine if it should set a standard for UWB over coax. The group is still just a few weeks old, but has the potential to move quickly if this path set by Pulse~Link looks as fruitful as some think.

On the Bluetooth side, its champagne corks all around for a deal that gives the SIG a broader market to address, Wibree backers an instant infrastructure for managing their spec an OEMs one less meeting and compliance process to staff. Proponents of the low-power, short range Wibree faces a review from SIG members but should still be able to ship products before the end of 2008.

There's still more digital home nets than you can shake a stick at with Wi-Fi, MoCA, HPNA, HomePlug, Zigbee and near-field all in play. But today things got just a tad more sane.

Sunday, June 10, 2007

Waiting on 6 GHz radios

That's the story for the Bluetooth SIG. The move to put the Bluetooth protocols over ultrawideband is making good progress as a spec, but the group needs prototype hardware to validate its work.

In a short chat Friday afternoon, Bluetooth SIG executive director Mike Foley said the group expects to see its first samples of 6 GHz UWB radios before the end of the year. That would enable it to finish its spec by this time next year. The SIG opted to skip the UWB Band Group 1 silicon operating at 3-4 GHz that's coming out now because some planned 4G cellular offerings will occupy that space, and cellphone makers were concerned about interference.

In the short term, the SIG is about to roll out its version 2.1 update with a number of adds. The chief addition is a simpler method for "pairing," the initial secure configuration of two Bluetooth devices so they can work together. This feature has been fairly user unfriendly to date.

Watch this space for more Bluetooth news Tuesday.

Friday, June 08, 2007

Flash slowly steps ahead

I hear Marvell is planning to deliver a flash controller for the emerging market for solid state drives, but I am having problems getting any of the Marvell-ous ones to talk to me about it. If you have any details about this or other flash controllers drop me a note at or leave a post here.

I am especially interested in controllers that try to solve some of the problems that are keeping multi-level cell flash out of computer systems. MLC flash has a tenth the number of rewrite cycles and perhaps ten times the bit error rates of its lower capacity cousin the single-level cell flash. I know of at least one company addressing some of these issues but would like to hear from others as I research this sector for a story to come at EE Times.

Thursday, June 07, 2007

Circling the wagons around Ethernet

It's looking pretty good for an early move to an enhanced Ethernet standard carrying Fibre Channel protocols based on the news from this week's T11 meeting. The proposal Cisco put together had co-authors from a whopping 16 supporting companies. The only other proposal, from Brocade, had backing from HP and IBM who were also behind the Cisco work.

There are a ton of details to sort through here—-defining Ethernet frames, Fibre Channel mapping, understanding all the various interconnect models and ensuring all existing functions and management tools work well together. Nevertheless, the spirit of these meetings seems very congenial, people are rolling up their sleeves and getting to work.

One wrinkle: EMC, HP and IBM are suggesting as part of a requirements document they put together the industry decouple this work from the IEEE congestion management standard which they think could take awhile. The congestion features could be added later, they suggest. It's also interesting to see Broadcom is staying in the background on this effort so far. The competitive friction between Cisco and Brocade could still spark debates, but so far it looks like smooth sailing.

Wednesday, June 06, 2007

Wireless USB's math: 99+150

A tip of the hat to my colleague Mike Clendenin hammering away at Computex in Taipei. His article this morning nails the math on what it will take to get wireless USB out the door: a $99 retail price tag on a device sporting 200 Mbit/second application layer throughput.

Today's notebook modules for wireless USB cost at least $25 to $40 to the OEM, according to a top Lenovo engineering manager I talked to recently. He said that needs to descend to less than $10. Meanwhile throughput for many of the devices is 50 Mbits/s or less, according to Mike's story today.

Originally, backers hoped wireless USB products would hit in a big way for the 2006 holiday season, but being just a year off for a big industry initiative is not a huge miss—if they hit this Xmas. Last I checked at WinHEC in mid May, the WiMedia Alliance was just starting its certification process, so there could be some news on this front, probably next week.

Incidentally, here's a little bit of my own math: This is my 200th post since beginning this blog last August. Perhaps it is sacrilegious to say this on Google's blog site, but...Yahoooooooo!

Speck Trek

The future of computing is all about getting everything on the Net to create an era when computers can automatically monitor and control all sorts phenomena once beyond the reach of digital, networked technology. Today a spinoff from the University of Spain is taking one small step toward that future of ubiquitous computing with the release of Squidbee, an open source hardware mote for sensor networks.

Libelium consists of about seven researchers who have developed a sensor network control point and a gateway suitable for Zigbee and Wi-Fi mesh nets. The products are available as schematic circuit designs and source code for the programs that are running inside the mote. As far as I can tell the design is available for free under a Creative Commons license. You can also buy the products from Libelium if you prefer.

There are plenty of other companies making sensor network systems and software, but none I am aware of providing the designs free of charge. This is an important direction that I have been watching out of the corner of one eye. Please chime in if you have any insights to add about Libelium's move or what's going on in sensor nets generally.

Tuesday, June 05, 2007

Frothing up the MoCA

Entropic, the first mover behind the multimedia over coax spec is cranking up the volume now that ultrawideband chip companies such as Pulse~Link are bringing their ultrafast technology to coax. Pulse~Link aims to deliver one chip set optimized both for coax and wireless, hitting data rates far beyond the 135 Mbits/s promised by MoCA.

To keep the pace, Entropic will ratchet its data rate up to 180 Mbits/s later this month, thanks to some new software on its existing chip. It also plans to add parameterized quality-of-service to its chip set so cable companies can guarantee no PC on a home network will bring down their digital TV streams. Entropic, the only company currently shipping MOCA chips, hopes the new speeds and feeds and their extensive field trials is enough to keep them ahead of the UWB pack.

Do you know of any other UWB chip makers planning to ride both coax and wireless? Drop me a line at or publish a post here.

BTW, I never did hear any word from Comcast that was supposedly going to announce at the Cable Show its plans to use MoCA. However, Charles Cerino, vice president of new media development at Comcast, was elected president of the MoCA board in May, a pretty telling endorsement.

PMC stays SASsy

PMC-Sierra continues its early-to-market drumbeat for the next-generation of the Serial Attached SCSI specification today, rolling out a RAID controller for SAS 2.0. To approach the throughput needed for the 6 Gbit/second SAS spec, the PMC chip also moved to the 5 GHz PCI Express 5.0.

As I noted earlier, PMC is using the 6.25 Gbit/s serdes from its telecom heritage combined with the SAS design capability it acquired from Agilent to leap ahead of others who may face bigger struggles with the relatively fast designs. The downside is the mainstream server and storage markets PMC targets probably won't move into SAS 2.0 for another 12-18 months. Even the new PMC controller is only sampling now.

Monday, June 04, 2007

The multi-radio world

Well, it's finally happened, the predictions a few years back that cellphones would incorporate four or more radios has come true. I sat next to a Nokia marketing guy at a conference lunch a few weeks ago who was regaling the table with his latest device that builds in cellular, Bluetooth, Wi-Fi, GPS and FM.

Just a couple years ago at a conference a Motorola manager was predicting this day, as if he were conjuring up a scene from Star Trek. I am reminded by this science fiction become reality by a note in my email box about a new cellular report from market watcher Forward Concepts. The reports predicts strong growth in multiple-mode wireless peripheral chips supporting WLAN, Bluetooth, GPS, FM radio and mobile TV.

Although troubled, it seems like mobile TV will gain traction as the next radio in the device. The report even projects the rise of 16:9 aspect ratio displays on cellphones in tandem with the growth of mobile TV. HDTV in your pocket? C'mon!

Of course it's not all about the glitz of the high-end handset. "The demand for voice-only low-cost and ultra-low-cost cellphones will
constitute the dominant market volume for the next three years," the report concludes. Thank God, I think I am already getting a headache from all these radio waves! Excuse me, my mobile instant messenger is ringing!

Friday, June 01, 2007

No habla Esperanto

Michael Feldman gives his reasonable assessment of the data center and the state of Infiniband over at HPCwire today. He and I agree that despite all the talk about data center convergence, the truth is we will see a polyglot world where Infiniband, Ethernet, Fibre Channel--and a few interesting options and variants lurking around the fringes-- will increasingly learn to co-exist.

Ethernet is the granddaddy taking an increasingly dominant slice of the market. Very slowly over time, Fibre Channel will become increasingly niched, I believe. But don't expect convergence.

The capitalist drive of markets to create proprietary value argues against technological homogeneity. (Whew, translate that into Esperanto!) The latest example is FastSoft Inc. which my colleague R. Colin Johnson writes today is rolling out a new appliance with yet a new take on how to process the TCP stack.