Design Con 2015
Breaking News
Slideshow

Ethernet Dike Sprouts Cracks

10/17/2014 01:41 PM EDT
15 comments
NO RATINGS
1 saves
Page 1 / 10 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
jdambrosia951
User Rank
Rookie
Re: Micro vs macro
jdambrosia951   10/23/2014 7:13:42 PM
NO RATINGS
Bert,

No i don't like to argue, but I am very concerned regarding much of the "information" i have seen in your posts.

Seriously - you can't even follow what i say to you.  I am not arguing for a 4 pair 40G port - what i am saying is that it has found acceptance is that it can serve as 4 10G pairs (which drives cost down for 10G) plus the added benefit of providing a 40G port if needed.

In regards to 2x25G, yes it will find some use.  THis is already common knowledge, due to the 25GbE Consortium.  However, you are once again missing my point - either 2x25 or 2x20 - both have the same issue - not the optimum configuratin, but 2x25 will find some use.  It is being driven from a device that most likely has serdes that runs from 10G to 25G - and this will provide the highest throughput.  So i'm not going to take a sucker bet.  However, does this mean we will see high volume deployment - most likely no, until we get to the serial lane 50G solution, which you seem to think will be just around the corner for servers.  I take exception to that perspective.  It will take time to develop that technology and is just getting started.

Good night.

Bert22306
User Rank
CEO
Re: Micro vs macro
Bert22306   10/23/2014 6:59:17 PM
NO RATINGS
John, I think you just like to argue. One of the very first strong disagreements you seemed to have was when I said that the sequence single serial interface, to multilane parallel, and then back to single, is an ongoing sequence for all interface types, as speeds go up. And that the attraction of the single serial interface is simplicity of connector design and no need for synchronizing the multiple lanes. For some odd reason, you felt obliged to disagree with those (also) very obvious comments, citing instead a particular technique used to synchronize one of the multilane Ethernets.

As if that invalidates the point.

Since you won't tell me exactly what you are referring to it is hard to fully answer you.  Dropping out of the development of 40G serial?

A couple of these EE Times articles have mentioned some vendors putting their 40G developments on hold, as they pursue the 50G. Did you miss that? Are you surprised?

For example, if you would look at the slides you would see how the number of cables or switches can be reduced by a factor of 2 or 4 depending on whether you use a 2x or 4x wide lane rate for the same number of servers.  The use of a 40G port configured to support 4 10G ports is driving the cost of 10G down.

Yes, and is it not totally predictable that the use of two 25G lanes, for 50G, will drive down the price of 25G, reduce the number of connector pins required compared to 4-lane 40G, and soon become the preferred solution over 4-lane 40G? Seems pretty clear to me.

How much do you want to bet that this is "where things are going," in the not too distant future?

 

jdambrosia951
User Rank
Rookie
Re: Micro vs macro
jdambrosia951   10/23/2014 6:17:45 PM
NO RATINGS
 

This has been explained to you so many times and you still don't seem to get it.  What I said was that the 40G port can be used as 4 ports of 10GbE – this allows it to be used in a high density mode to support ever lower costs of 10GbE, and despite all of the focus on the higher speeds, the largest volume to date is still at GbE and is moving to 10GbE. Cloud scale data centers are moving beyond 10GbE.  Therefore, 40GbE ports will be around for a long time yet, as they are capable of supporting 10GbE, and then the same equipment can provide a fat pipe.

Okay, so 50G can be supported with either 5 of these same 10G lanes, or with only two of the newer 25G lanes, or with a new single-lane 50G standard. I don't understand your fascination with sticking with the 4-lanes at 10G each, that's all. In the future, prices for the 25G will come down, as prices always do.

 

>> My fascination?  Seriously?  Have you bothered to read any of the article that you are commenting on?  Volume will drive cost, but so does integration.  For example, if you would look at the slides you would see how the number of cables or switches can be reduced by a factor of 2 or 4 depending on whether you use a 2x or 4x wide lane rate for the same number of servers.  The use of a 40G port configured to support 4 10G ports is driving the cost of 10G down.  Your major cloud scale data centers are using this today.  What I pointed out to you is that this port can then also support 40G as well.  The bi-mode usage is very helpful to many customers.


First of all 50G signaling doesn't exist yet electrically, and the 2x25G mode doesn't allow the highest utilization of the switch device.

And yet, in spite of this temporary issue, a few vendors have opted out of 40G. Does that tell us something, John?

 

>> Since you won't tell me exactly what you are referring to it is hard to fully answer you.  Dropping out of the development of 40G serial? That doesn't surprise me at all, though a number of individuals are still calling for its development.  Perhaps you are referring to 40G ports based on QSFP going away?  Provide me some data sources on that.  I don't see it.  Perhaps you are referring to OC-768 – yes I can believe that.

 

And by the way we were discussing 50G signaling versus 2x25.  50G will use 2x25 in the near term, because it is available, and there are some applications that can use it.  However, my point is still dead-on.  It is not the cost optimized solution, because it requires the two lanes.  Go look at the math and maybe you can figure that out.

You have just parroted speeds that the Ethernet community is working on.

Exactly! That's why I don't understand your resistance to the obvious. The Ethernet community is "working on" getting back to 10/25/50 multiples, and not the weird-o 40 multiples, but for some reason, you are more focused on retaining 4 lanes than on the aggregated interface speeds these vendors and users are aiming for in the future.

>> My resistance to the obvious?  It is very obvious that you really don't understand the underlying applications and factors that are driving all of these different rates.

Your comment "So this ends up being more a question of Ethernet tradition than anything technically limiting" shows a complete disregard for the challenges of the physical channels that the Ethernet community is now being asked to solve

Come now. These challenges are solved one way or another, but there is always a target on the horizon to shoot for. And as I said, several vendors, who evidently are aiming at just such targets out there on THEIR OWN horizons, have dropped work on 40G. Surprised? Not me!

 

>> Come now?  You must be a software engineer to think this way. These are not trivial problems to be solved.  Let me ask you this – if these problems are so easy to solve – why isn't the industry working on Terabit Ethernet?  Oh I know your answer will be "Exactly!"  We answered this question multiple times and ultimately people got it – no real technically or economically feasible solutions on the horizon.  We have to worry about power, complexity, latency.  So while you sit there and continue to not understand where things are going, I will continue to work with the industry to make further higher speeds, like the work being performed by IEEE P802.3bs 400GbE Task Force, happen.   By the way the pages are public, and I would invite you to review them, and perhaps gain a better understanding.

 

Bert22306
User Rank
CEO
Re: Micro vs macro
Bert22306   10/23/2014 4:46:00 PM
NO RATINGS
This has been explained to you so many times and you still don't seem to get it.  What I said was that the 40G port can be used as 4 ports of 10GbE – this allows it to be used in a high density mode to support ever lower costs of 10GbE, and despite all of the focus on the higher speeds, the largest volume to date is still at GbE and is moving to 10GbE. Cloud scale data centers are moving beyond 10GbE.  Therefore, 40GbE ports will be around for a long time yet, as they are capable of supporting 10GbE, and then the same equipment can provide a fat pipe.

Okay, so 50G can be supported with either 5 of these same 10G lanes, or with only two of the newer 25G lanes, or with a new single-lane 50G standard. I don't understand your fascination with sticking with the 4-lanes at 10G each, that's all. In the future, prices for the 25G will come down, as prices always do.

First of all 50G signaling doesn't exist yet electrically, and the 2x25G mode doesn't allow the highest utilization of the switch device.

And yet, in spite of this temporary issue, a few vendors have opted out of 40G. Does that tell us something, John?


You have just parroted speeds that the Ethernet community is working on.

Exactly! That's why I don't understand your resistance to the obvious. The Ethernet community is "working on" getting back to 10/25/50 multiples, and not the weird-o 40 multiples, but for some reason, you are more focused on retaining 4 lanes than on the aggregated interface speeds these vendors and users are aiming for in the future.

Your comment "So this ends up being more a question of Ethernet tradition than anything technically limiting" shows a complete disregard for the challenges of the physical channels that the Ethernet community is now being asked to solve

Come now. These challenges are solved one way or another, but there is always a target on the horizon to shoot for. And as I said, several vendors, who evidently are aiming at just such targets out there on THEIR OWN horizons, have dropped work on 40G. Surprised? Not me!


jdambrosia951
User Rank
Rookie
Re: Micro vs macro
jdambrosia951   10/23/2014 4:29:00 PM
NO RATINGS
Wow– not really sure where you are getting this from? 

But you are missing a key point here.  40 GbE can be broke out to 4 ports of 10GbE, as well as the higher port 50GbE.  50GbE is being discussed more in terms of single lane rates, therefore it won't serve this application space.

How so, John? If "this application space" consists of data centers or ISPs, and/ore anyone who prefers twisted pair to fiber, why wouldn't either a single lane 50G or a two-lane 25G server that space just fine? Cost is all that matters, ultimately. I doubt any of these folk would insist on 4-lane interfaces or that 40G speed (unless, of course, they are a metro area Ethernet still dependent on SONET/SDH speeds).

Are you surprised, for instance, that several companies have dropped development of 40G? I'm not at all surprised. I would have done exactly the same thing, with 50G becoming available.

>>>  This has been explained to you so many times and you still don't seem to get it.  What I said was that the 40G port can be used as 4 ports of 10GbE – this allows it to be used in a high density mode to support ever lower costs of 10GbE, and despite all of the focus on the higher speeds, the largest volume to date is still at GbE and is moving to 10GbE. Cloud scale data centers are moving beyond 10GbE.  Therefore, 40GbE ports will be around for a long time yet, as they are capable of supporting 10GbE, and then the same equipment can provide a fat pipe.

So your question about why not do a single lane 50G or a 2x25G port.  First of all 50G signaling doesn't exist yet electrically, and the 2x25G mode doesn't allow the highest utilization of the switch device.  This is the same reason why people aren't using 2x20G.  Here's a hint – go look at the CFI for the 25GbE single lane server connection, and you may understand this stuff better.

Finally – your comment about twisted pair to fiber – 40G on twisted pair doesn't exist yet.  It is in development.  People are using DACs and in some instances, active cable assemblies or MMF. 


While no one is 100% certain on the future, the reality is that 500GbE does not fit into any emerging trends.

Sorry, John, but if 400G is even close to "an emerging trend," then 500G is more so. The "emerging trend" that created 40G fizzled out, or is in the process of doing so. Emerging trends are 2.5G, 5G, 25G, 50G, 100G, so 500G seems a lot more likely to me than 200G or 400G.

>> Wrong again in so many ways again!  The emerging trend of 1x / 4x implementations has not died out, and continues.  2.5 and 5G is really related to wireless WAPs and the fact that the Cat5E isn't supported by 10GBASE-T.  10G / 40G far from dying.  25G is ¼ of 100G, and as the next signaling rate leads to a cost optimized system.  And what exactly is your trend – you forgot 250G.  The emerging trend I spoke of is the next signaling rate (assuming a continuation of doubling with each generation) then multiply by 4 : 10/40, 25/100, 50/ 200, 100/400.  This of course may change.  Please explain what you see as your trend.  You have just parroted speeds that the Ethernet community is working on. 
The only additional comment I would add here, based on the 25GbE story is that people need to consider the TDC (total deployed cost), and not just the rate of a single link.  Perhaps you meant this, but it was not clearly stated.

Yes, I totally agree with that comment. And I also believe that hardware can be designed to operate at 10, 25, and 50 multiples about as easily as it can at 40 multiples. So this ends up being more a question of Ethernet tradition than anything technically limiting.

>> Never said anything about 40 multiples.  I have spoke of 10x multiples (Ethernet rate jumps), 4x multiples from an Ethernet architecture that is now being reflected in rate jumps, and then 2x multiples in terms of signaling rate jumps.

Your comment "So this ends up being more a question of Ethernet tradition than anything technically limiting" shows a complete disregard for the challenges of the physical channels that the Ethernet community is now being asked to solve.  I have often said that the pesky laws of physics are getting in the way.  Perhaps you want to go look at all the channel analysis that got done as part of the 100GbE backplane project?  It's amazing to think that at one point the electrical length of a backplane connector used to be part of a bit interval, because nowadays- there are multiple bits in a connector at a given point.

Bert22306
User Rank
CEO
Re: Micro vs macro
Bert22306   10/23/2014 3:28:01 PM
NO RATINGS
But you are missing a key point here.  40 GbE can be broke out to 4 ports of 10GbE, as well as the higher port 50GbE.  50GbE is being discussed more in terms of single lane rates, therefore it won't serve this application space.


How so, John? If "this application space" consists of data centers or ISPs, and/ore anyone who prefers twisted pair to fiber, why wouldn't either a single lane 50G or a two-lane 25G server that space just fine? Cost is all that matters, ultimately. I doubt any of these folk would insist on 4-lane interfaces or that 40G speed (unless, of course, they are a metro area Ethernet still dependent on SONET/SDH speeds).

Are you surprised, for instance, that several companies have dropped development of 40G? I'm not at all surprised. I would have done exactly the same thing, with 50G becoming available.

While no one is 100% certain on the future, the reality is that 500GbE does not fit into any emerging trends.

Sorry, John, but if 400G is even close to "an emerging trend," then 500G is more so. The "emerging trend" that created 40G fizzled out, or is in the process of doing so. Emerging trends are 2.5G, 5G, 25G, 50G, 100G, so 500G seems a lot more likely to me than 200G or 400G.

The only additional comment I would add here, based on the 25GbE story is that people need to consider the TDC (total deployed cost), and not just the rate of a single link.  Perhaps you meant this, but it was not clearly stated.


Yes, I totally agree with that comment. And I also believe that hardware can be designed to operate at 10, 25, and 50 multiples about as easily as it can at 40 multiples. So this ends up being more a question of Ethernet tradition than anything technically limiting.

jdambrosia951
User Rank
Rookie
Re: Your feedback
jdambrosia951   10/23/2014 5:31:51 AM
NO RATINGS
Rick,

I only have so many fingers :)  Been getting teased a lot about being the little dutch boy.


John

jdambrosia951
User Rank
Rookie
Re: Your feedback
jdambrosia951   10/23/2014 5:30:55 AM
NO RATINGS
Larry,

When you say "these standards" you need to be a little more specific, especially in light of the various efforts that are currently being tackled in the IEEE 802.3 Ethernet Working Group.  Individuals targeting data centers are indeed having a big push on 25GbE efforts.  If you were to look at 400GbE, you would see an interest mix, as there are people targeting the Router / client connections, as well as looking to the future of 100GbE / 400GbE in the data center.   If you look at the CFI to do the next gen Access BASE-T Phy effort, you would see an effort driven by anticipated growth in BW from WAP points.  Note this is cost driven, as in many instances, people don't want to upgrade their existing UTP cable structure.  And if you were to look at the single pair efforts and PoDL- you would see a strong base from the automotive and industry communities. 

So you see it all depends on which standards you are referring to....

jdambrosia951
User Rank
Rookie
Re: Micro vs macro
jdambrosia951   10/23/2014 5:24:18 AM
NO RATINGS
Bert,

Please see my comments below.

I think that there's more apparent confusion than actual confusion.

>> Really?  Yes, I remember seeing you at these meetings all the time, as we discussed the various rates of Ethernet – 100 Mb/s to 400 Gb/s.  Oh wait, no I don't.  The reality is that there are a lot of these discussions going on, and have been going on all year.  It appears that the Ethernet community has now had a chance to digest and embrace these new applications and needs, and we have moved forward on various development efforts.  The industry, however, will now get hit with a multitude of similar solutions that vary in terms of rate and supported solutions, that it will need to deal with.  The Ethernet Alliance will be ramping up its efforts to highlight the interoperability between the various solutions, but it will still be a lot.

First thing is, I'll bet a truckload of cash that many Ethernet customers do not care about a specific speed, as much as they care about a price point at a significantly faster speed than they use now. For example, I'm pretty positive that no data center customer requires 40G, as opposed to 50G, aside from cost and availability. So again, more apparent confusion than actual confusion.

>> yes, a rather simplistic statement, don't you think?  But you are missing a key point here.  40 GbE can be broke out to 4 ports of 10GbE, as well as the higher port 50GbE.  50GbE is being discussed more in terms of single lane rates, therefore it won't serve this application space.  And I am pretty sure no one wants to go back and change ASICs and modules to then support 4 lanes of 12.5Gb/s to get to 50Gb/s.  So the application space will also play a determining role. 

And yes indeed, it should be expected that 40G and 50G do not need to coexist, if the costs end up being similar. The 40G is bound to get left behind. By extension, same will happen to 400G, when the predictable 500G comes to be.

>> 500GbE predictable?  Let's see, this must be based on your belief that Ethernet will continue its 10x progress.  While no one is 100% certain on the future, the reality is that 500GbE does not fit into any emerging trends.  Will there be 10x leaps – well I doubt it will be single 10x leaps between generations, but it is very likely that you will sometimes see this, such as between 40GbE and 400GbE.  But look at 25GbE – no one is talking about 250GbE – we are talking about 200GbE – why – the 1x/4x trend that has been repeatedly pointed out to you.  And why would we consider 500GbE?  ¼ of that would be 125GbE – once again not the lower 100GbE rate.  So I highly doubt 500GbE, based on today's trends, and not based on where Ethernet was 10 plus years or so ago.

Then there are the implementation details for each speed, the media options. So, if data centers or ISPs want a much faster copper standard, the choice of, say, 25G vs 40G vs 50G will certainly not be made based on the exact speed, but rather based on the availability of a twisted pair standard in that general speed range, at an affordable price.

>> The only additional comment I would add here, based on the 25GbE story is that people need to consider the TDC (total deployed cost), and not just the rate of a single link.  Perhaps you meant this, but it was not clearly stated.

So I don't think reality is (will be) as chaotic as it's being made out to be. And as some of the viewgraphs showed, accurately is my bet, the popular interface standards will revert back to those that easily aggregate to powers of 10. Any takers for my bet?

>> You need to better define your bet.  Are you talking about 10x leaps between generations – or whether there will ever be 10x variations between rates.  I already answered that above.

PFM_#1
User Rank
Rookie
Re: Your feedback
PFM_#1   10/20/2014 11:40:45 AM
NO RATINGS
Perhaps there is confusion or maelstrom because silicon is no longer scaling, thus designers, having never faced this situation, are unsure what to do. Or, perhaps it is the data content since WiMAX is going away and all the videos are going to be streaming over the network from wireless, thru wired, and back to wireless. Maybe it is people doing standards too long and are unable to get out of the box to reinvent or reinvigorate Ethernet to serve the new type of users. Is it possible that MSA's are the new standards path since this type of effort is more dynamic and able to provide the desired Quality of Service. Today a gap, tomorrow chasms.

Page 1 / 2   >   >>
Flash Poll
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Top Comments of the Week