x
Optical/IP

10-GigE Vendors Get Cold Feet

In the last couple of weeks, two performance tests for 10-Gigabit Ethernet have been canceled due to a lack of vendor interest.

The Institute of Electrical and Electronics Engineers Inc. (IEEE) finally ratified the 802.3 specification for the technology on June 13, 2002. And even though several companies claim to have 10-Gbit/s products ready, many are balking at the chance to test their products’ performance. In the past few weeks, a live demonstration at the 10 Gig Inaugural Expo has been canceled, and Meir Communications has called off its 10-Gig Ethernet test. Why aren’t vendors interested?

“No one came right and out and said that they didn’t want to participate because their box couldn’t perform at line rate,” says Mike Bennett, senior network engineer at Lawrence Berkeley National Laboratories, which was putting together the 10-Gbit/s demonstration for the Inaugural Expo. “But let’s just say they haven’t been beating down our doors to be included. You can draw your own conclusions.”

OK. We will.

The Inaugural 10 Gig Expo was supposed to get underway next week in Phoenix (see Show to Demo 10-Gig Throughput). Bennett and his team from Berkeley Labs had designed a test bed using two sets of 12 high-end computers to pump traffic through the network at 10 Gbit/s. Some vendors had already signed up to participate, but many at the last minute declined or pulled out. Finally, two weeks ago, the show’s sponsor, Pinnacle Conference Network, was forced to call it off.

Last week, Meir Communications, an independent Gigabit Ethernet testing service, was supposed to begin testing gear for an article that was scheduled to appear in the August edition of Business Communications Review. Once again, a lack of vendor interest killed plans for that test. Sources say that BCR is still expected to run a story, but instead of testing all the 10-Gbit/s players, it is simply asking vendors to fill out a survey of their capabilities.

There are several switch companies already shipping, or at least beta testing, 10-Gbit/s Ethernet products. Many of them participated in the 10 Gigabit Ethernet Alliance (10GEA) interoperability demonstration last month at Network + Interop in Las Vegas (see Vendors Show Off 10-GigE at N+I). These included switch vendors like Cisco Systems Inc. (Nasdaq: CSCO), Enterasys Networks Inc. (NYSE: ETS), Extreme Networks Inc. (Nasdaq: EXTR), Force10 Networks Inc., Foundry Networks Inc. (Nasdaq: FDRY), and Nortel Networks Corp. (NYSE/Toronto: NT). Riverstone Networks Inc. (Nasdaq: RSTN), which has not yet shipped its 10-Gbit/s product, did not participate in the N+I demo.

Some of these companies, like Foundry and Force10, agreed to participate in the 10 Gig Inaugural Expo demonstration. But many, like Cisco and Riverstone, declined. As for the BCR test, Meir Communications would not reveal which vendors declined to be tested. Force10 says it is still willing to have its gear tested by both testing houses; it will be used in a demonstration staged at the Lawrence Berkeley Lab facility later this month. Cisco says it plans to provide BCR with the necessary specification information for the August story.

“The reason we decided not to participate is basically a resource issue,” wrote Larry Yu, Cisco spokesperson, in an email to Light Reading. “We had to choose between an upcoming third-party 10GigE interoperability test at the end of the month and this. We chose to participate in the interoperability test.”

Cisco is planning to participate in an interoperability test run by the Tolly Group.

It seems clear from the choice of tests that vendors are more comfortable in an interoperability demonstration than in a performance test. And who can blame them? According to Bennett, who is familiar with most of the products available today, Force10 Networks is the only company that even comes close to achieving 10-Gbit/s line-rate throughput (see Force10 Shows Off 10-GigE Switch). He says the startup’s switch typically achieves forwarding throughput between 9.0 Gbit/s and 9.5 Gbit/s, depending on packet size.

“It is somewhat misleading to customers, because if you say it is a 10-Gbit/s product then I expect it to forward at line rate,” says Bennett.

Then, of course, there's the issue of customers. Despite the hype surrounding 10-Gbit/s Ethernet, the fact is that most service providers aren’t ready to offer it. Tier 1 service providers in the U.S. aren’t expected to deploy 10 Gbit/s for at least 12 to 18 months, according to Mark Sue, an analyst with Frost Securities Inc. Large interexchange carriers like WorldCom Inc. (Nasdaq: WCOM) and RBOCs like SBC Communications Inc. are just now rolling out Ethernet services (see WorldCom Unveils Metro Ethernet). Meanwhile, Ethernet service providers like Yipes Communications Inc. and Sigma Networks are filing for bankruptcy protection and selling off assets (see Another Metro Provider Fails: Was Vendor Financing the Difference?) and Yipes Joins Chapter 11 Club).

While large carriers typically adopt new technologies at a slower rate than new players, 10 Gbit/s may also be cost prohibitive right now. Initially, 10-Gbit/s Ethernet will be expensive, with per-port costs as high as $80,000. And as T1 costs drop from $1,000 per month to $750 per month, there is little incentive for customers to switch their service to Ethernet.

But analysts expect prices to drop drastically over the next few years, just as 100-Mbit/s Ethernet and 1-Gbit/s Ethernet did. According to IDC, 10-Gig's prices will decline to $7,800 per port by 2005.

After the pain of the initial investment wears off, the technology will likely take off, says Mark Sue of Frost Securities in a research note he published yesterday. He says that with Sonet OC192c interfaces priced at around $300,000, 10 GigE is already a bargain. But many people disagree. They say that Sonet OC192c port prices have also dropped and are now under $100,000.

The true sweet spot for 10-Gbit/s Ethernet, at least in the near term, seems to be in the enterprise. One of the biggest drivers in this market is the sharp decline in pricing of 1000base-T Ethernet interfaces. These network cards are selling for as low as $50 a piece. And with 1000base-T Ethernet running within enterprise networks, 10-Gbit/s Ethernet switches will become necessary for traffic aggregation. Bennett says he also sees supercomputing as another potential market segment.

— Marguerite Reardon, Senior Editor, Light Reading
http://www.lightreading.com
Page 1 / 9   >   >>
Scott Raynovich 12/4/2012 | 10:13:09 PM
re: 10-GigE Vendors Get Cold Feet Thanks Michael, greet feedback!
Mike_Bennett 12/4/2012 | 10:13:21 PM
re: 10-GigE Vendors Get Cold Feet Folks,

I would also like to add a few points for the sake of clarification.

One thing, perhaps I am reading into the responses, is that I am GǣdissingGǥ the efforts and accomplishments of the IEEE and 10GEA. Nothing could be farther from the truth. I have the utmost respect for those who have worked so hard to create a high quality standard as scheduled. The folks who solved the extremely difficult problems under difficult circumstances in a downturn economy where human resources were dwindling as the months went by deserve enormous respect and accolades.

Regarding the Pinnacle event, the context of our demonstration is that the standard is done Gǣjust in timeGǥ to support research efforts that take place in institutions such as the national labs, University of New Hampshire, San Diego Supercomputing Labs, and many others since that is where the Gǣearly adoptersGǥ of the technology are. As an end user of the technology, like Roy, I have also used SpirentGs equipment for both demonstration and testing. Spirent has been at the forefront of the technology in my experience for 1 Gbe as well as 10 Gbe, in many cases enabling the work necessary to complete the standard to be done.. I also agree that objectives are quite different between testing and demonstrating a given technology. There are also different objectives in different technology demonstrations. As Mark Fishburn so clearly stated in his response, GǣGǪputting together a 20 hop demo with data passed successfully from a generating system through every system vendorGs equipment was just an outstanding achievement. It featured full reach of the optics over 200km. at a good rate if not full rate and was a tremendous testimonial for the plug and play nature of Ethernet in a network containing a larger group of vendors than may ever be seen in a real deployed network.Gǥ I agree, this is quite an accomplishment in anyoneGs book.

With that said, the purpose of LBNLGs demo is to show that we do have real applications that can use the BW right now, not 12-18 months from now as some folks would have you believe. Will sales folks make their numbers based on the volume of these applications? No. Do my clients care about that? No. The point is that I have to deliver networks with the capacity to handle their application(s) in a cost effective manner. Part of providing a cost effective network is building the simplest network possible to meet the end userGs requirements. Why? Because simple is easier to deploy and manage and that affects the bottom line. So, I could deliver 10 gig with any vendorGs equipment. I could use link aggregation as someone suggested in one of these threads to aggregate 10 individual 1 gig links. Link aggregation is not the simplest solution. Admittedly, 10 Gigabit Ethernet is not the lowest cost solution right now, thatGs the price you pay to be on the Gǣbleeding edgeGǥ of the technology. As the number of ports shipped goes up, the prices will come down, Ethernet will become more pervasive outside the traditional LAN and when Ethernet in the First Mile is done, we network engineers will not have to deal with as many protocol translations, perhaps none at all. In the long run, that makes things simpler, cheaper, and everyone wins.

Obviously, I am an Ethernet advocate. While the squabbling here and elsewhere will continue regarding the individual vendorGs products being able to deliver line rate, the choice ultimately lies in the hands of the end users. The accomplishment of the IEEE to produce the high quality standard in time to meet our needs transcends all of that. Someone once said, GǣIf you build it, they will come.Gǥ Such is the case for 10 gig and will be the case for the technologies that follow. I, for one, am quite thankful that there are so many brilliant and capable engineers out there willing to take on the task and deliver as promised.

Respectfully,

Michael J. Bennett
Sr. Network Engineer
Lawrence Berkeley National Lab
hoffmane 12/4/2012 | 10:13:32 PM
re: 10-GigE Vendors Get Cold Feet Based on what I have seen of the major vendor 1 Gbe switches, their 10 Gbe will be jitter (latency variation)inducers.

It does not matter if the transmission line is jitterless, going through a layer 2/3/4 switch may produce jitter effects all of its own.

Perhaps Cisco, Foundry, Extreme et.al. should publish figures on multihop latency build up and also transportation of jitter sensitive traffic like video and voice (and not just 1-5 streams, try 80-100+ streams).
hoffmane 12/4/2012 | 10:13:33 PM
re: 10-GigE Vendors Get Cold Feet Probably including the cost of a major OC192 ADM or DCC in this price (expensive, large ADM/DCC).
Roy_Bynum 12/4/2012 | 10:13:33 PM
re: 10-GigE Vendors Get Cold Feet sigint: "Surely, the equipment could derive its timing from the upstream SONET equipment. This could be provided as an option where required."
_________________________________________________

This functionality was specifically excluded from the 10GbE WAN PHY functionality. Ethernet interfaces are peer level respecting transmission clock timing, each transmitter clocked independently, with the far end receiver deriving bit clocking from the embedded data signal. 10GbE was not intended to be a client on a data transmission network, so network source timing was considered to be "out of scope" of this Ethernet development project. If some vendors implement this functionality, it is not part of the 10GbE standard. In the future, there may be a need for 10GbE to be a client to a 40Gb SONET/SDH system. At that time, Ethernet switch vendors may make it a practice to include full support for the ANSI standard for SONET client interfaces.
Roy_Bynum 12/4/2012 | 10:13:33 PM
re: 10-GigE Vendors Get Cold Feet MarkFishburn: "it should be pointed out that putting together a 20 hop demo with data passed successfully from a generating system through every system vendorGs equipment was just an outstanding achievement. It featured full reach of the optics over 200km. at a good rate if not full rate and was a tremendous testimonial for the plug and play nature of Ethernet in a network containing a larger group of vendors than may ever be seen in a real deployed network."
______________________________________________

To those reading this board,

Having put together multi-vendor test beds, I know that it is often more difficult to generate "full rate" data aggregation through multiple vendor and multiple hop/source/sink data links and sources than one might imagine. While it is often easy to "daisy chain" Ethernet switches, making sure that the configurations from the different vendors systems (each with its own set of defaults) do not interfere with other, takes more patience than most realize. Given the nature of many demos, I am surprised that you were able to demonstrate even a Gǣgood rateGǥ.

There is a big difference between demonstrating a technology and stress testing the technology. I am familiar with Spirent test gear as well as other vendors data generation and communications test systems. I have used it to not only demonstrate technology architectures, but also to do stress testing of deployment configurations and new technologies. There are also test system configuration differences between demos, that run for long times, and GǣtestsGǥ that often run for short periods of time.

Running a demo at Gǣa good rateGǥ over a period of time does not mean that either the testing systems or any vendorGs systems that might be put under a bandwidth performance test would not pass a full bandwidth performance test. It means that running the demo over an extended period of time was done at an optimal data transfer rate for the systems and length of time that the demo was running. Actual performance information will have to come from the individual vendors. Even the third party testing labs often do not publish that information because of current NDA practices.

Roy Bynum

sigint 12/4/2012 | 10:13:34 PM
re: 10-GigE Vendors Get Cold Feet Huub_van_Helvoort wrote:

> The whole point of making the 10GigE PHY
> standard was so that> SONET/SDH and 10GigE > would be compatible.

So why did the IEEE make it not fully compatible?
There are enough differences to make it even
incompatible.

At the last ITU-T SG15 meeting operators claimed
that they would NOT transport this signal because
it will not meet the +/- 4.6 ppm clock accuracy.
_________________________________________________

Surely, the equipment could derive its timing from the upstream SONET equipment. This could be provided as an option where required.
Roy_Bynum 12/4/2012 | 10:13:34 PM
re: 10-GigE Vendors Get Cold Feet Hub_van_Helvoort: "So why did the IEEE make it not fully compatible?
There are enough differences to make it even
incompatible.

At the last ITU-T SG15 meeting operators claimed
that they would NOT transport this signal because
it will not meet the +/- 4.6 ppm clock accuracy."
_________________________________________________

The WAN PHY was NOT intended to be compatible with SONET or SDH. It borrowed the overhead of SONET/SDH to provide consistant OAM capability over optical facilities. There is a deliberate need for SONET/SDH translation at the section and line level of the overhead. Only the path overhead and payload was intended to be carried unmodified over service provider optical systems. The translation system is referred to as the "ELTE" by P802.3ae and is referred to as a "Path Relay" by ANSI. The transmission clock tolerance was changed from +/-100PPM to +/-20PPM at the instance of ITU SG15 to support Class B regenerators. (+/-4.6 PPM clock accuracy is for Class A regenerators and Stratum 1 sources.)
st0 12/4/2012 | 10:13:35 PM
re: 10-GigE Vendors Get Cold Feet Markfishburn said:
"It featured full reach of the optics over 200km. at a good
rate if not full rate and was a tremendous testimonial for the plug and
play nature of Ethernet in a network containing a larger group of
vendors than may ever be seen in a real deployed network".
-----------------
Agree about good progress in technology.

I know it is Ethernet. Net has this wonderful rate depression: 14.4 modem usually gives you 2.5, 56 gives a bit of high than 32 and so on. As we talk about killer application, we explore the full rate capability, voice the feature and benefit of the high rate. When it comes to deliveble, we cheer the "near full rate" as "good rate", without even shame about "good rate" is hiding under the name of the "full rate".

Let's just call it as is: 8GbE or 10GbE-(minus). If we can't be honest on the tech issue, may be we should just go to Wall Street join the analysts.

I would prefer someone to state the cause of short coming and the "fixes" implementation timeline in order to achieve the full rate, as we set up to deliver at 1st place. At the moment, just simplely state: we are not there yet (over promise and under deliver).

st
hoffmane 12/4/2012 | 10:13:37 PM
re: 10-GigE Vendors Get Cold Feet 15 switch routers in the same box, connected to a backplane using 8 Gb/s as a link.

So really you have the equiv. of 15 small switches using 802.3AD to link to a backbone switch? Yes?

Must get complicated to setup for a normal routing environment?, Tell the truth is sounds similar to Fore's ASX 1100, smaller switches meshed on the backplane, and and was a nightmare to work with.
Page 1 / 9   >   >>
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE