Wednesday, August 22, 2007

Happy birthday to meeeeeee!

Well, it's been 357 days, 258 posts and 34,400+ page views since I started this blog a year ago. According to Google AdSense I have earned roughly $40 so far. I haven't actually seen a nickel yet. Google cuts you a check when you hit $100.

While this may be no financial bonanza, I feel proud that I have been able to sustain this venture and grateful to those of you who find it worth surfing on a regular basis. I look forward to more posts, comments, hits—and maybe someday enough cash for a little vacation in Greece—or at least Gilroy.

If you have any comments about what you like, hate or want to see more of in this space rant away with a comment here or at rbmerrit@cmp.com. I looooove your frank feedback. And if you have any ideas about how I can make money on this little adventure, I wouldn't mind hearing about that, too!

Now I actually am going on vacation (on my own dime) for the week of August 27. I may post from the occasional Internet café, but for the most part my situation will be described in the words of the Beach Boys who once said, "I'll see you in September!"

Two leaps for Fibre Channel

Now that Ethernet folks are finally rolling out a variety of commercially interesting 10G adapters and have laid out a road map to 100G, the Fibre Channel folk are scrunching up for a little leapfrog action. At least two companies are sampling 8G FC adapters—presumably being tested with early versions of 8G switches and arrays—and the community is gearing up for work on its next act.

The T11 group is editing technical comments on its 8G spec now with plans for a vote on a final draft expected late this year or early next. The early sampling of adapters is part of an effort to shake the bugs out of the technology

"The challenge with these high speed technologies is in handling the dispersion and signal disruptions all across the system in a cost effective manner," said Robert Snively, a senior Brocade engineer who chairs the T11 group. "Two-thirds of the new work in 8G--and 10G Ethernet for that matter-- is in how you define and measure behavior of the components," he told me in a phone interview the other day.

Next up: 16G Fibre Channel. "16G is just a gleam in our eyes right now," said Snively. "There has been no formal work done on it, but there is a consensus we will start work on it late this year or early next year," he said.

Engineers seem confident the 16G serial spec is do-able in CMOS. The big debates likely will come over what sorts of equalization and pre-emphasis to use, signaling techniques that are still more art than science.

Tuesday, August 21, 2007

Stumping for Ethernet storage

Alacritech said Monday its product revenue from iSCSI sales doubled from the previous fiscal year, and the company expects it to double again in its next fiscal year. No doubt the actual revenue numbers are pretty low, since Alacritech declined to state them. You can assume this is one of those phenomena of huge growth on a tiny base.

The company also claimed sales are on the rise for Microsoft's TCP Chimney offload software. Alacritech has technology inside Chimney that Microsoft licensed after an intellectual property tussle. Again, no hard numbers. I welcome anyone with hard data to share it in a comment here or by dropping me a note at rbmerrit@cmp.com. From what I have heard anecdotally, use of Chimney is still very marginal because few end users are enabling TCP offload in their networks.

Alacritech quoted the Enterprise Strategy Group, a market watcher I don't know, saying more than 20,000 customers have deployed iSCSI. In a recent study, the analyst firm found that of 500 IT managers surveyed 17 percent are using iSCSI in product environments with 20 percent planning to implement the technology. In addition, Alacritech quoted International Data Corp as saying it expects iSCSI to account for 20 percent of external disk storage by 2010.

Lots of interesting factoids, but I take them with a grain of salt. Market research habitually shows everything moving up and to the right, and I suspect even in their optimistic reports there are more wrinkles and qualifiers to the story Alacritech chose not to quote.

Alacritech's "news" may be little more than marketing in an effort to attract a few more deals for its TOE and iSCSI technology. Still, the company should be allowed to crow a little bit. It helped pioneer those technologies and slogged through the desert of the past several years when it was not clear if either would take off.

Frankly we are not out of that desert yet. Alacritech may be seeing no more than a small oasis, or even a mirage in the distance. As I have written here many times, Intel and Sun are trying to obviate the need for TOEs with new processor features and Ethernet designs they are happy to license. And as for iSCSI, it has been the next big thing since about 2000. Many of the startups that raised millions on its promise are no longer here to pop a cork with Larry Boucher and Co.

If someday iSCSI and TOE do go mainstream, Alacritech and other survivors may be too busy really celebrating to issue a press release.

Monday, August 20, 2007

FCoE ready for ASICs

Kudos to Brocade and Cisco for agreeing on a single frame format for Fibre Channel over Ethernet. They struck a compromise that moves the standard forward, although it probably requires both companies to rework any ASICs they have in development.

Now chip development can start in earnest for adapters that will handle both Ethernet networking and Fibre Channel storage. But there is plenty of software work yet to be done to finish the spec, and no final standard is expected until April 2008.

Ah, to live in a simpler world where everything rides over Ethernet. We are stepping closer to such a world—even as the Fibre Channel folks gear up to leapfrog good old Ethernet yet again. More on that later.

Friday, August 17, 2007

Proximity a step nearer

It's still too early to say whether or when Sun could turn Proximity into a useful product, but the company disclosed a milestone in its research today based on a paper for next week's Hot Chips conference. Proximity is Sun's concept for a high bandwidth, ultra dense chip interconnect using capacitive coupling. The latest advance takes the form of a multi-chip module package that solves the problem of aligning as many as 72 signal pads.

Sun's work on a four-port 10 Gbit Ethernet switch was a good vehicle for vaulting some of the semiconductor manufacturing and packaging hurdles in the technology's way. But until the research can be scaled up to a proof of concept that shows promise for commercial products, Proximity remains a fascinating idea. It may take a few more years of work to determine whether it could be a viable alternative to today's copper or tomorrow's optical chip-to-chip alternatives.

It's too bad Sun didn't get the Darpa HPCS contract to pursue Proximity as part of its supercomputer bid. But a separate team at Sun is also looking at how Proximity could be used in tandem with optical interconnects so this soup continues to cook.

Thursday, August 16, 2007

A finger on the problem

For years people have told me fingerprint recognition hardware will be ubiquitous in notebooks, keyboards and mice in a year or so. Thanks to Roger I. Quint, CEO and CTO of software startup 123id, for explaining to me why this is not happening—a lack of standards.

Quint said his startup has some of the great software needed to read and match a fingerprint image quickly. But his company has to support about eight major fingerprint readers, each with its own way of capturing and reporting an image. The lack of standards is a particular problem for IT managers who don't want to have to support multiple conflicting products or risk buying systems that embed what could be the losing recognition hardware.

Keith Horn of Fujitsu said the company's latest chip, in the market about a year, costs just three dollars at the OEM quantities and will appear in some notebooks next year. Still, to penny pinching PC makers $3 is a lot of money for something users happy with their passwords see as a nice-to-have novelty feature. To prod the market, Fujitsu has assembled a whole soup-to-nuts solution of a mouse design and all the software needed to surround the chip. Such efforts are necessary, but not sufficient for chip makers these days.

Everybody in security knows we need to get beyond the no-security level of passwords and move to something like these fingerprint recognizers. But until this sector can set some standards and drive the costs down even lower this will remain a niche option for a single SKU in the broad product portfolios of the big PC players.

Wednesday, August 15, 2007

Be my eyes and ears

I hereby deputize all readers of the Interconnects Blog to be reporters for a week. I'll be covering an out-of-town conference when the Hot Chips and Hot Interconnects conferences are going on at Stanford and a first-ever meeting of a new WiMax broadband wireless special interest group are held in Silicon Valley.

I wrote a preview of the Hot Chips/Interconnect conference, and have a report coming Monday on the biggest new announcement from the event—a massively parallel processor from startup Tilera. Likewise, I wrote extensively about the outlook for WiMax in the new 700 MHz being auctioned off early next year.

Nevertheless, sometimes unexpected things get said at these events and there is often some good scuttlebutt just from hanging out with the community. If you pick up some insights, I invite you to post a comment here or drop me a line at rbmerrit@cmp.com. After all, this is the age of the citizen journalist, and you now have been officially blessed as temporary members of the Fourth Estate. I look forward to hearing from you.

Tuesday, August 14, 2007

The IB beat goes on

Startup Luxtera Inc. announced Blazar, a 40G optical active cable based on its silicon photonics technology. The cables, among other uses, will help pave the way for lower cost links for the upcoming 40G Infiniband.

IB has already helped define and proliferate the short reach CX4 copper cables that are seeing increasingly broad use in the data center for hops of a few meters at 10G. Blazar could extend such links to 300 meters as well as quadrupling the bandwidth.

Yes, this is just a niche interconnect for data centers, but it keeps IB early in that high-end, high margin part of the business where it needs to be. Proprietary technologies are also expected to jump on board Blazar, as eventually will Ethernet given the recently approved plan to define a 40G version of it for the data center.

Blazar uses single-mode fibre attached to optical transceivers that power four transmitters running at 10.5G each with a single hermetically sealed laser. Luxtera claims it is easier to install and maintain and has reliability and cost advantages over VCSEL-based solutions.

The interconnect is packaged with a Quad Small Form-factor Pluggable MSA connector and consumes 2.2W per cable-end. It samples at the end of the year and is in production early next year, about the timeframe for the 40G Infiniband interconnect.

I suspect Mellanox Technologies has its hands full these days getting its ConnectX hybrid IB/Ethernet technology out as well as supporting an increasingly broad line of products. But hey, we have a line out for you now. Bring on the 40G Infiniband briefing.

Luxtera quotes International Data Corp. As saying the Infiniband revenues will grow 45 percent on a compound basis over the next five years. Is it getting time for a second source of silicon?

Monday, August 13, 2007

Banging the Infiniband drum

The folks at Intel continue to promote Infiniband as a high-speed interconnect in the data center. Next month, they will help launch the InfiniBand Low Latency Technical Forum, co-located with the Fall Intel Developer Forum in San Francisco. (Intel provides discounts for engineers attending both events.) The event is co-sponsored by the InfiniBand Trade Association and the Open Fabrics Alliance which maintains a common software stack for Infiniband and high-performance Ethernet.

The Sept. 17 event is really a follow on to last year's IBTA and OFA Joint Developer's Conference that kicked off IDF. The link to last year's presentations is not currently live, but as I recall it was a mix of case studies of Infiniband success in core markets such as technical computing clusters and new territories such as large commercial database systems. I would expect to hear more of the same this year as Intel tries to drive IB beyond its technical niche and more into the enterprise mainstream.

I don't doubt Infiniband will always have a latency edge over any flavor of Ethernet, and often it will also have a throughput advantage. But with all the activity and energy going into 10G Ethernet, backplane Ethernet, the proposed higher-speed standards at 40G and 100G and the work on Fibre Channel over Ethernet, the advantages of Infiniband may narrow over time, making its shift from technical to enterprise markets increasingly difficult.

Friday, August 10, 2007

PC, telecom don't see eye to eye

So is it a bad thing that the PC and telecom worlds are out of step when it comes to things like interconnect rates, serdes and equalization? That's the question some folks have raised in the wake of the news about 8GT/s and linear DFE being chosen for PCI Express 3.0.

Personally, I think this is an opportunity missed (once again) for industry convergence, but not a huge one. And I completely get the rationale behind the PC industry's go-slo strategy.

Still, I'd like to conduct a straw poll from the field to hear what you think and why. While we are at it, chime in on anything you heard in the initial Express 3.0 news that you have questions or comments about. After all, that's the first step in the process of getting to a final spec. As always you can comment here for all the world to see and react to, or drop me a line in privacy to rbmerrit@cmp.com

Wednesday, August 08, 2007

It's lucky 8 day, I guess

For the Chinese 8 is a lucky number, and 888 is even luckier. Perhaps they are the ones behind the 8/8 announcements of not only 8 GT/s being chosen for Express 3.0 (see below) but the first 8G Fibre Chanel products I have heard about. Emulex and QLogic (thank you Mr. Anonymous Commenter below) announced host bus adapters, a mezzanine card for server blades and an embedded I/O controller for 8G FC today. The company said its 8Gb/s products are still in prototype testing phase, but they should sample before the end of September.


Geez, I didn't even know the 8G FC standard was done! Hey, somebody drop me a note at rbmerrit@cmp.com about how this transition is shaking out and what's happening on the 16G standard, will ya. I am waiting to hear from the FC world!

It's 8 GT/s for Express 3.0

Not surprising the PC industry should place cost and compatibility over raw performance. That was the result of a six-month debate on the maximum throughput target for PCI Express 3.0 announced today.

A few years ago, the PCI SIG debated whether Express 2.0 would double the theoretical 2.5 GHz rate of its initial version or go for the gusto and try to match up with the world of 6.5 GHz serdes and interfaces set by the telecom world. Yep, they opted for 5 GT/s. The PC has always been about good enough performance and massive volumes based on keeping a tight lid on costs, so today's decision on 8 over 10 GT/s for the third gen of the interconnect is not a huge surprise.

Perhaps more interesting will be details of Express 3.0's new encoding scheme which will break with 8b/10b encoding and what kind of equalization it will employ. The group did not specifically say 3.0 will be fully backward compatible with both today's 5 and 2.5 GHz products, but I think that level of compatibility was a key factor in deciding on 8 GT/s.

Tuesday, August 07, 2007

Tidbits from LinuxWorld

I've been wearing out the carpets at Moscone Center in San Fran today looking for stories, but only picked up a few scraps, mainly about 10 Gbit Ethernet from the LinuxWorld show floor.

One switch-chip vendor I talked with said some people are now saying Intel will win the controller wars with its IOAT technology, and ultimately they will adopt RDMA in a future generation. Sujith Arramreddy, the chief technology officer at ServerEngines, said that makes no technical sense because the IOAT approach requires multiple data movements inconsistent with the RDMA approach.

I had a comment from one reader who said IBM is doing—or planning—some innovative moves with IOAT. I haven't heard back from Big Blue on that yet, but Bill Ott, a former IBM xSeries server designer now COO at ServerEngines, said there was no such thinking at IBM when he left last year.

Meanwhile, startup Teak Technologies says it is porting its congestion management algorithms to the programmable core in the BladeEngine chip recently announced by Server Engines. It currently runs in an FPGA, which is OK for a switch blade but too expensive for a prototype NIC it showed at Linux World running next to the Neterion 10 Gbit controller.

Separately, the folks at the Storage Networking Industry Association have started a task force on green technology. It aims to act as a liaison with the Green Grid which announced its plans for 2007 today. The SNIA group also hopes to develop its own metrics for power efficiency in the SAN.

Monday, August 06, 2007

Making too many waves?

Two interesting stories on wireless ironically appeared on the same page of my San Jose Merc this morning. In one, Intel crowed because it has finally shipped a notebook chip set for the emerging WiMax network. In the other, metro Wi-Fi operators bewailed the fact casual consumers are not signing up for their services, forcing net operators to require significant contracts with city governments to assure they get paid for building and maintaining the nets.

Intel has been driving WiMax hard for years, claiming it will make it happen just as it did Wi-Fi by embedding it in every notebook. Intel needs the extra bandwidth to drive systems sales. But with broadband at the office, at home and in the coffee shop, there is not a huge pent up desire to fire up devices while walking in the park or driving around town. Someday novel applications and improving devices such as the iPhone may drive a need for broadband everywhere, but it's gonna take awhile.

Thursday, August 02, 2007

Next up for 10GbE: The Squeeze

I got a few minutes with one of the Valley's leading computer designers after a panel discussion today. Since it was just one of casual encounters and he was speaking pretty frankly, I'll let him remain anonymous.

He estimated more than $300 million in venture capital has gone into funding companies designing 10G Ethernet media access controllers. "That's bigger than the whole market for 10G MACs," he quipped.

For all the clever design work, the winners will come down to the two or three companies that have superior price/performance while staying under 5W. And success may depend more on software given the complex iSCSI, RDMA, TOE and etc. stacks these chips may need to run. When it's all done there may only be enough market for three or four large companies who sell in enough volumes to make profit in this cost-squeezed sector, he said.

So, now that the 10GbE controllers are about to ship from Broadcom, Intel and Marvell—as well as many startups, it won't be long before the numbers are out and the shakeout begins. Bob Wheeler over at The Linley Group thinks the big OEM design wins may not come down until next year at this time given the software complexity and the relative lackluster demand for 10GbE.


Got some independent test results on one or more of the new 10GbE controllers? Drop a post here or send me a heads up at rbmerrit@cmp.com

Wednesday, August 01, 2007

Q'comm incubates Wibree killer

Just when you thought the fragmentation in personal area networks was easing a bit, Qualcomm tips its plans to roll out a new PAN technology. I don't have a name or any specs, but Len Lauer, a Qualcomm group president, told me after a panel discussion at the AlwaysOn Stanford Summit that the company will have out in the market next year a new short-range wireless net now in its labs.

The Q'com PAN will compete with Wibree, developed by Nokia. Like Wibree it will have much lower power than Bluetooth, and target applications such as sensor networks. It will also not need line-of-sight to make connections. "It's similar to Wibree, but we think it's better," Lauer told me.

There's a lot at stake. Designers see the cellphone as a natural gateway for a wealth of consumer devices that will have always-on links to PANs.

It's not surprising Qualcomm is going its own way here, especially when Nokia—engaged in an intellectual property battle with Qualcomm--has a claim on the underlying Wibree technology. Qualcomm has a history of ambitiously doing its own thing, for example developing its own mobile TV technology rather than adopting the DVB-H standard.

Still this is a curve ball to the industry. Just weeks ago the Bluetooth SIG struck a deal with Nokia to bring Wibree under its umbrella. So a lot of people will be frustrated to see another potential source of fragmentation and competition in this space.

Know more about the Qualcomm PAN technology? Post a comment or drop me a line at rbmerrit@cmp.com

Can Bluetooth boogie?

I see an increasing number of Bluetooth ear clips these days but few and far between are the stereo headsets that go beyond voice calling and deliver music. Tom Nault is pretty frank about the problem: the headsets either cost too much or deliver poor quality. He hopes his new lossless codec changes that fact.

We shall see. Even the vaunted Apple iPhone lacks support for the Advanced Audio Distribution Profile piece of Bluetooth to enable stereo, but that could change with the next rev.

One big potential gotcha: how to make sure all those wonderful 2.5 GHz devices such as cordless phones, microwaves and Wi-Fi nets that surround the people likely to use such headsets don't interfere with their Bluetooth music. Now there's a design challenge that could push things off to the Bluetooth 3.0 version that leaps to ultrawideband technology.

 
interconnects