About 10 Gbit/s class

10Gig Ethernet is heavily used both in the enterprise field and by carriers. There are some differences in how the technology is typically used - e.g. there are probably more carriers than enterprises using WAN-PHY, and fewer carriers than enterprises using 10Gig Ethernet in combination with STP (carriers are more likely to use it as a point to point technology).

So why does it have to be one or the other?

Steinar Haug, Nethelp consulting, snipped-for-privacy@nethelp.no

Bonjour Steinar,

Thanks for your comments.

I'm sure we will have one day to define a complete Ethernet typology.

For example, 10Base-T is, globally, a technology from enterprise, like

100Base-TX/FX.

It is clear that 1000Base-LX10 is only used in access, e.g. for point- to-point FTTH.

Now, as you say, the field for 10G is not so obvious, because we have LAN-PHY and WAN-PHY.

And for 40G/100G? will we say that it can be a LAN technology for entreprise?

Perhaps, we have to consider a thing like this:

(1) EFM class Ethernet (2) Enterprise class Ethernet (3) Carrier class Ethernet that includes (2) and EoT (commonly MAC over TDM).

Best regards, Michelot

if you need enough bandwidth and the cost per Gbps works, why not?

i think you are mixing up design criteria for the original interfaces, and design usage.

FWIW i work for a carrier and we use internally and rent to others at least

3 flavours of 10G:

STM64 for SDH, (and WAN PHY Ethernet which is basically the same format) -

9.8 Gbps 10G LAN PHY - 10 and a bit Gbps, so preferred if it is the same price as you get a few more bits (and the ports on switches and routers tend to be cheaper). Ethernet only. "flexi" ports on modern metro and long haul DWDM for lambda interface cards that can do both, as well as the 10G fibre channel varient and various other formats i dont get involved in. where it exists, pluggable client side optics mean you also get to choose laser power / distance independently.

WAN PHY tends to be more expensive, and a lot of kit interfaces more efficiently if you use it as SDH - but the ports cost more. since 10G SDH / SONET came 1st this stuff works on more or less any 10G transport, so in practice, you can get it anywhere 10G is supported. LAN PHY has lower cost interface on many bits of equipment, and is supported on "stuff" that doesnt have alternatives - LAN switches for example.

ironically sometimes the one you pick is based on subtleties - such as which reliably detects a link failure most quickly at the physical level....

So - what makes you think that such differences make any sense in the 1st place, and that your categories exist anywhere except on a marketing slide?

Bonjour Stephen,

Thanks for that good sense!

Probably, and if we can get together "design criteria" and "design usage" we can also get together technicians, experts and commercial managers. We are in the way of the story.

It has a lot of worth, thanks very much for your expression here.

For me, your statement already refers to 3 flavours of 10G Ethernet. Which one is not used in your organization?

(1) Mapping of 10 Gbit/s Ethernet (MAC frames) into VC-4-64c using 64B/

66B coding. This is specified in ITU-T G.707. Other mappings than MAC can be done into a VC-4-64c container, e.g. HDLC/PPP, ATM.

(2) 10GBase-W also called 10G WAN PHY. It is mapped into a STM-64. This interface is not specified by ITU-T, but by IEEE in 802.3 (formely 802.3ae) so, the clock accuracy is different (20 ppm instead of 4.6 ppm), the use of the overhead is also different.

(2) Mapping of 10 Gbit/s Ethernet into VC-4-64c using the GFP adaptation. This is specified in ITU-T G.7041.

For these 3 flavours of 10G, the rate is 9,95 Gbit/s, the rate of the STM-64.

OK, that's 10Gbase-R with the clock accuracy 100 ppm and the rate 10 Gbit/s before the line coding. It doesn't use the WIS interface for STM-64 IEEE flavour.

In DWDM we can have different rates and different protocols on the wavelengths, e.g. 2.5 / 10 / 40 Gbit/s and SDH / Ethernet / FC / OTN.

I don't know if the wavelength switching is solved, that is changing remotely the wavelength without electrical decoding and without swaping locally the lens in the equipments (issue of wavelength blocking).

Probably depending on the fiber type. Normally, the distances are lower for IEEE than for ITU-T.

That is changing every month, ITU-T is being standardized the failure interlayer relations, e.g. between GFP / MPLS / ETH.

To bring together IEEE works and ITU-T works.

EFM Ethernet is only IEEE (although for the copper, the VDSL and SHDSL physical layers are from ITU-T). Enterprise Ethernet is only IEEE. Carrier class Ethernet is mainly ITU-T (and even if we are some specific IEEE interfaces, like WAN PHY, the specification of OAM is done by ITU-T).

In the future, 40G and 100G will use OTN interfaces specified by ITU-T (G.709).

Best regards, Michelot

If you check out the websites for NetEffect and Chelsio you will see that you can buy 10G NIC cards that will slot into a standard server using PCIe. So 10G is very much available for the enterprise today.

In fact I see Chelsio has for sale two 10GBASE-CX4 evaluation NIC cards and

3m metres of cable for 995 USD. So for less than the price of a high end PC you can hook two computers together in your bedroom using 10G Ethernet.

we use WAN PHY.

clocking accuracy has never come up in practice - possibly as the lambda transport tends to be more forgiving than a formally compliant STM-64.

sort of - again what tends to happen is you get 1 or another 10G-Base interface between the lambda and the Ethernet equipment - although the 2 bits of kit may be in the same site, we have occasions when we drive a fibre link between the 2.

yes - and since DWDM manufacturers want fewer types of card to improve costs just like everyone else, they tend to have a single card that can do different encapsulations at around 10G.

The lower speed services need some sort of TDM muxing built into the cards - those are getting more flexible as well.

We use at least 3 different "metro" DWDM products, and 3 different recent long haul DWDM systems as well as various legacy ones - so interoperability is a big issue.

The TDM on DWDM stuff doesnt seem as good as "raw lambda" for plugging the kit together.

long haul DWDM does opto regen but switching wavelengths at a PoP under remote control makes the network a lot more flexible, and we have some kit that can do that (although it doesnt switch so often).

altering frequency using a tunable card is more about not needing dozens of different card type in a DWDM mux - 1 card may cover a subband for the mux.

And ""colour" on the client side if the last hop needs to go on a metro DWDM system.

yes, but real life isnt just about what you might buy now (once you prove it works) - interworking with the installed base is probably more important.

we probably spend more adding lambdas to existing systems than on new transports

there has to be a big delta in cost / performance before it is worth all the pain and disruption of changing out something that already works unless you have no choice.

No - i dont care where the standard comes from - just that it works, and when i plug it into another type of kit that claims to meet the same standard, no issues come up.

Again - Ethernet is probably easier, more predictable and better at interoperability than many other standards in the telco and IT world.

Try plugging a laptop power supply into another make of laptop........

The troubles i have had often come down to tinkering with working standards to give us "improvements" - who came up with jumbo frames and forgot to standardise the number of bytes in the frame?

(snip)

Sounds good, but how many can actual process data that fast?

-- glen

I don't know what NetEffect has for offload support, but the Chelsio cards support full TCP offload, which will certainly help the processor(s) get more done. Of course, the OS needs to be able to take advantage of that offload.

Also bear in mind, some people aren't looking for 10Gbps so much as they're looking for a lot better than 1Gbps. Even if you can only realize 3 or 4 Gpbs out of the 10G link, you're getting much more throughput than you'd otherwise have with a 1G link (and a lot less complication than trying to do some form of link aggregation?).

Patrick ============================================================================ Patrick Klos Email: snipped-for-privacy@klos.com Klos Technologies, Inc. Web:

Bonjour Stephen,

The reason is not the lambda transport. If we can have a band of 8 x

10 Gbit/s of STM-64 (SDH directly in WDM), the clock accuracy must stay at 4.6 ppm for each of the 8 wavelengths. But it is not necessary that they are all synchrone : lambda #3 could be + 2 ppm, lamba #4 could be - 3 ppm...

Good

Could you give the names ? Is it ADM or OADM / ROADM equipments, like the 1626 LM from Alcatel which can be installed in metropolitan and long haul networks.

It might depend on the grid that is used, and perhaps the gain equalizer

OK, and which equiment are you using for that?

Is your carrier in the long haul part? or in a metro part face to ISPs. Are the differences very clear for you?

Good

Best regards, Michelot

Bonjour Marris,

Sorry, I make wrong click previously!

...and thanks for the information.

It's a little bit cold, I like better my wife!

Best regards, Michelot