RE: [EFM] 1 Gbps != 999.9 Mbps
Vadimir,
I disagree with your assessment of that Ethernet is not a "clean" data
communications technology. My experience with Ethernet is evidently very
different from yours.
In many cases, it has been found that the "SLA" has little to do with
customer perception and satisfaction. The SLA just determines when he gets
to bill service provider instead of the service provider billing him. If
he wants a "clean" service, he will find it, regardless of the "SLA". In
tern, service providers that want to market a "clean" service will find
vendors that will provide the infrastructure that will provide it.
Thank you,
Roy Bynum
Thank you,
Roy Bynum
At 10:05 AM 9/28/01 -0400, Vladimir Senkov wrote:
>Roy,
>
> >While I would not expect to be able to provide a "circuit emulation type of
>
> >service" that could perform as a "clean" facility from port A to port B. I
>
> >would however expect to be able to provide a true circuit type of service
> >that would perform as a "clean" facility from port A to port B.
>
>I totally understand your concern as a service provider.
>However, Ethernet is frame based technology.
>When you provide customer with a line that has V.35 on one side and V.35 on
>the other . . . then it is a true clean circuit.
>Customers can loop it back on one side, send a byte of data from the other
>and expect it to come back within a certain timeframe. They can also expect
>to be able to pump certain amounts of bytes per second over that line.
>
>However, when a service provider is providing a customer with two Ethernet
>ports and claims that it is a "clean" facility it is a different story.
>Due to the very nature of Ethernet, even If i give you a cat5 crossover
>cable you will still not get the same "clean" facility over it if you use
>Ethernet. There is no way to tell you what the delay is and throughput is
>going to be unless I know details about the traffic.
>For example, if you want to send one byte over the Ethernet . . . what's
>really going to happen is: this byte of yours will be encapsulated into a
>frame. this frame is going to be at least 64 bytes. before transmitting the
>frame (let's say you have full duplex so there are no collisions to worry
>about) a preamble has to go out. then the frame. then the interframe gap. So
>the number of bytes actually sent out to the wire is not exactly 1 . . .
>Even if we have a full duplex 1G link, even if we do NOT calculate frame
>header and CRC overhead, even if we are only using the largest frames
>possible (1518), even if we are working the line rate as we should, the max
>we can possibly get on a 1G is about 81200 frames per second. If we consider
>the entire frame to be accountable as "user traffic" (not exactly true
>because of the frame header and CRC overhead) then we get 81200*1518*8=
>986092800 bits per second.
> 986092800/(1024*1024*1024)=0.9183704853058Gbps.
>You can get much much poorer numbers . . .
>And since we don't know what customer's traffic is going to look like we
>have no way of telling what the thruput in Gbps is going to look like. But
>to just give you an idea . . .
>Imagine a hypothetical situation . . . customer's traffic is very strange.
>He sends one byte at a time. He wants every byte to be in a separate packet.
>We are not putting more than one packet in a frame. Now how many bytes per
>second can he send?
>the smallest frame we can send is 64 bytes. In other words, we have to do
>some padding.
>The amount of 64 byte frames we can send on 1G full duplex "clean" link is
>1488000. So that is the exact amount of bytes per second that we are looking
>for. In gigabits per second this is: 1488000*8/(1024*1024*1024)=0.01Gbps
>
>Now if the customer gets himself a 2Mbps circuit with two V.35s hanging on
>both sides, hooks up some equipment and sends those same bytes over the wire
>he can send at a rate of about 0.002Gbps.
>
>So for this customer we are offering a whopping 5 times increase! That's
>going from 2M to 1G.
>
> >In many
> >cases, if we, the service providers, do not provide that "clean" facility,
> >we wind up paying the customer. It is called "liquidated damages". You
> >can see my reluctance to consider a standard that is based on providing
> >only a "circuit emulation type of service".
>
>I can see that, but i can also see that service providers want the cheap
>stuff. And that's frame stuff, not circuit emulation :)
>After all, frames are not bad or anything . . . they are just frames. So i
>guess if customers really understand what exactly they are buying we are all
>going to be OK.
>
>Regards,
>Vladimir.
>
>At 11:50 PM 9/27/01 -0400, Vladimir Senkov wrote:
> >Roy,
> >
> >So if you wanted to test that hub . . . you'd expect it to be tested
> >according to Ethernet testing methodologies . . . right? I'm saying . . .
> >you would not expect it to provide a circuit emulation type of service with
> >a clean 10 million bits per second from port A to port B.
> >So how is that different from a customer buying EFM line?
> >
> >Regards,
> >Vladimir.
> >
> >-----Original Message-----
> >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
> >Sent: Thursday, September 27, 2001 11:39 PM
> >To: Senkov, Vladimir [BL60:T855:EXCH]; 'ah_smith'
> >Cc: 'stds-802-3-efm'
> >Subject: RE: [EFM] 1 Gbps != 999.9 Mbps
> >
> >
> >Vadimir,
> >
> >I am just an ignorant customer. When I go to a computer store and buy a
> >"hub", what I get is 802.1d bridge that has half duplex Ethernet ports. I
> >have an old "hub" that I got from a just such a store. The box claims
> >802.1d bridge support. Perhaps I am guilty of using the vernacular
> >meanings of some terms. But then, I am just a customer.
> >
> >Thank you,
> >Roy Bynum
> >
> >At 10:51 PM 9/27/01 -0400, Vladimir Senkov wrote:
> >
> > >Absolutely agree.
> > >I didn't want to repeat myself (here and was trying to do that via
> > >personal e-mails), but i guess I just have to say that those who test
>they:
> > >
> > >1) Test equipment vs. technology
> > >2) Know what they are testing and what they are expecting as a result
> > >Those who sell testing equipment:
> > >1) Also sell methodologies, training and certification.
> > >2) Some of those methodologies are in the RFCs. There are other standards
> > >for that as well.
> > >Those who sell equipment:
> > >1) Tell customers exactly what they are selling
> > >2) No matter how much they want otherwise, they are going to sell the
>EFM,
> > >but not the "pure 1G story"
> > >EFM needs to address specific needs of those who are going to buy it.
> > >Those needs may include: delivering data, video, voice, whatever . . . to
> > >whatever distance, concentration of users, etc. Those needs may include:
> > >security, pricing, etc, etc.
> > >
> > >but not just "1G". "1G" is not a need. it is more like a sign on a
>freeway
> > >or something . . .
> > >let's say "upto 1G". even with OAM in-band, it will still be "upto 1G".
> > >
> > >I'll also repeat that Ethernet performance (to the end user anyway) is
>NOT
> > >measured in bits per second. And nobody is selling 10Mbps repeaters/hubs
> > >(yes hub IS a repeater :) for example. They sell Ethernet hubs instead.
> > >
> > >Therefore, customer who buys such a hub knows that he is not going to sue
> > >the seller for not delivering 10Mbps. Customer will only expect
> > >performance characteristics of that device to be tested against Ethernet
> > >performance measurement methodologies.
> > >
> > >Regards,
> > >Vladimir.
> > >
> > >-----Original Message-----
> > >From: Andrew Smith
> >[<mailto:ah_smith@xxxxxxxxxxx>mailto:ah_smith@xxxxxxxxxxx]
> > >Sent: Thursday, September 27, 2001 10:21 PM
> > >To: Roy Bynum
> > >Cc: 'stds-802-3-efm'
> > >Subject: RE: [EFM] 1 Gbps != 999.9 Mbps
> > >
> > >
> > >Roy,
> > >
> > >I'd be interested to see how you propose measuring one of your
> > >802.3x-pause-rate-limited services against one of these "certification"
> > >testers. But seriously, folks, to paraphrase that doctor-patient story
> > >"Patient: Doctor, it hurts when I sell my customers 1000000000.0000 bps
> > >service. Doctor: then don't sell them that, sell them what you can
> > >guarantee".
> > >
> > >Andrew Smith
> > >
> > >
> > >-----Original Message-----
> > >From: owner-stds-802-3-efm@majordomo.ieee.org
> >
> >[<mailto:owner-stds-802-3-efm@majordomo.ieee.org>mailto:owner-stds-802-3-ef
> >m@xxxxxxxxxxxxxxxxxx]On
> > >Behalf Of Roy Bynum
> > >Sent: Thursday, September 27, 2001 4:36 PM
> > >To: bob.barrett@xxxxxxxxxxxxxxx; Harry Hvostov; fmenard@xxxxxxxxxxxxxxx;
> > >'Denton Gentry'
> > >Cc: 'stds-802-3-efm'
> > >Subject: RE: [EFM] 1 Gbps != 999.9 Mbps
> > >
> > >
> > >Bob,
> > >
> > >You would be surprised at how little bandwidth loss it takes for
>equipment
> > >to fail certification in a services infrastructure deployment approval
> > >process. If we tell our customers that we are delivering a GbE, then we
> > >deliver a GbE that will pass their most rigorous performance test,
> > >including throughput.
> > >
> > >Thank you,
> > >Roy Bynum
> > >
> > >At 12:11 AM 9/28/01 +0100, Bob Barrett wrote:
> > >
> > > >Harry et al
> > > >
> > > >yup, all the IP 'stuff' is payload as far as the demarcation point is
> > > >concerned.
> > > >
> > > >The demarc is a PHY that carries packets at the end of the day. Some
> > >demarcs
> > > >may be buried inside a bigger system, however, the standard must also
> >cater
> > > >for stand alone demarc devices. My expectation as a user would be that
>at
> > > >the demarc the bandwidth was the same capacity as my enterprise MAC and
> >PHY
> > > >of the same spec.
> > > >
> > > >Would I miss 10k per second on a 1GE, I doubt it.
> > > >
> > > >Would my test gear pick it up on an end to end private circuit test, I
> > >don't
> > > >know, anyone on the reflector tried this?
> > > >
> > > >Bob
> > > >
> > > > > -----Original Message-----
> > > > > From: Harry Hvostov
> > > [<mailto:HHvostov@xxxxxxxxxxxx>mailto:HHvostov@xxxxxxxxxxxx]
> > > > > Sent: 27 September 2001 17:41
> > > > > To: 'fmenard@xxxxxxxxxxxxxxx'; 'Denton Gentry';
> > > > > bob.barrett@xxxxxxxxxxxxxxx
> > > > > Cc: 'stds-802-3-efm'
> > > > > Subject: RE: [EFM] 1 Gbps != 999.9 Mbps
> > > > >
> > > > >
> > > > > And how about the ICMP and IGMP traffic from the same CPE devices?
> > > > >
> > > > > Harry
> > > > >
> > > > > -----Original Message-----
> > > > > From: Francois Menard
> > > [<mailto:fmenard@xxxxxxxxxxxxxxx>mailto:fmenard@xxxxxxxxxxxxxxx]
> > > > > Sent: Thursday, September 27, 2001 6:05 AM
> > > > > To: 'Denton Gentry'; bob.barrett@xxxxxxxxxxxxxxx
> > > > > Cc: 'stds-802-3-efm'
> > > > > Subject: RE: [EFM] 1 Gbps != 999.9 Mbps
> > > > >
> > > > >
> > > > >
> > > > > Or for that matter, what about ARP traffic unsolicited from my CPE
> > > > > devices ?
> > > > >
> > > > > -=Francois=-
> > > > >
> > > > > -----Original Message-----
> > > > > From: owner-stds-802-3-efm@majordomo.ieee.org
> > > > >
> > >
> >[<mailto:owner-stds-802-3-efm@majordomo.ieee.org>mailto:owner-stds-802-3-ef
>m
> >@majordomo.ieee.org]
> > > On Behalf Of Denton
> > > > > Gentry
> > > > > Sent: September 26, 2001 3:12 PM
> > > > > To: bob.barrett@xxxxxxxxxxxxxxx
> > > > > Cc: stds-802-3-efm
> > > > > Subject: [EFM] 1 Gbps != 999.9 Mbps
> > > > >
> > > > >
> > > > >
> > > > > > Service providers have a desire to offer a full 1GE service and
>not
> > > > > > use any of it's bandwidth for OAM. The rule of conservation of
> > > > > > bandwidth means the OAM needs to go somewhere other then in the
> > > > > > bandwidth reserved for the 1GE payload. I take it as read that
>100%
> > > > > > utilisation of a 1GE is unlikely, but that is not the point. The
> >point
> > > > >
> > > > > > is that service providers want to offer 1GE service period, not a
> > > > > > 999.9Mbit service.
> > > > >
> > > > > Does the existence of the Mac Control PAUSE frame therefore make
> > > > > Ethernet unsuitable for service providers?
> > > > >
> > > > > Denton Gentry
> > > > > Dominet Systems