Ultra Ethernet Testing-Webinar-Recording-Files原创

Testing.We will start in just a few minutes um but we actually very curiousious there there and and I about a how much you know about art for Ethan it so um and as well.During the presentation you're welcome to ask questions and in your top right corner there's a question mark.And during the presentation you're welcome to ah to ask questions and then in the end of the presentation we will follow up with the with the questions here.But in the meantime it would be great if you can test out this question mark and um if you would share your knowledge about ultra isn't it.On a s scale from one to tent tent being very experienced and one being kind of ah a new be in ah in ultra Ethan it that would be great then um it gives us a feeling for.

For your skills and and we can maybe adjust our our speech a little bit to to that so um I can already now see that some have very little some have some experience and.We also have people that have.Long experience with art for Ethan it so a good mix here.SOI'M.So that that's great that's great so um and hopefully when um the presentation have ended you will be on on maybe a level.Six or seven if you come from a low number so I'M.But I can see a lot of people have joined now so um I would like to say um welcome good morning good afternoon and good evening and I'm very happy to be here together with the Leo and and Lee will take us through some interesting information about art for Ethan it.Linked layer and also how you tested etca so I'm over to youu.

Thank you Martin.Hi everybody um welcome'LL come here to the web are my name's Leo and we're um I'm going to walk you through the archeryathern Internet link layer testing today.So just'again.So my agenda will be the first ah we're going to take a recapb on the critical role of the artery and linked layer in losses networking.And then followed by that we're going to diveing into l law in cb cbfc how they work the mechanisms and how they operates.And followught by that we're going to talk about some key tested scenarios for Ella n CBC and at the end we're going to some test solutions followed by a life DEMO.All right so the critical ROM of um arttraritherite linked layer in lases networking.

So.We have a problem in AI h PC workloads in data center that is packag losss because those two workloads are extremely sensitive to packs.Um.Because paal laws will forces their the the re transmission to ORS stas the um in the communication pipelines.And you would do re transmission on the transfer layer that will be too slow um for AI in h bc training.Which disgraced the performance of the collective operations.So there are actually ah common like two common costs for paty los um the first one is about INC cast.It's actually about the AI HPC workload those collective operation um causing in cast which means multiin notes trying to send to a single notes um resulting in receiver congestion.

Offer over overflow.The second one is about physical corruption like hardware failure or capeing that you have a packet corruption.So these two causes are the main causes that we see um in pack a loss which we which everybody hates.So a little bit on this um in cast cost pack a lots.So um collective operations like are reduce um many meter ranks into one rank and of course when these traffic are are um going through um a receiver Switch.To Switch receiver buffer might get overflow.If not properly managed.And that would drop packets.So this is what we call in cast or congestion at at the receiver site.The other one is hardware and capling for example um in the image you can see maybe you have a faulty or aging concecr that's been there for many many years right on laser power degradation and maybe you have a bent fiber optics somewhere um you know not very carefully managed.

Now all of these they will.Cost some signal degradation and then on the package level zero becomes one one becomes zero and then you have a corruption in your packet.And that is packet corruption and that will cost a packet drop as well.Fortunately in the ethernn link layer of athernet.Maybe probably have already read this back and those two features are optional but optional doesn't mean not important.

And actually those two are very very important for losses network because ellaar linkedla are rettro it's solving the package corruption um based pack lotse.The other one c b f c is preventing the receiver buffer overflow.So yeah.So linked these two features are actually the the two.Major pillars that can guarantee a network to be lossless.Let's dig in into lar and c b f c.So first let's look at how Allah works.So annaar ISS in the link layer and it sits between the ah Mac control and the logic link control layer as you can see there.And um.There'is a sequence number because as you can see is doing the the re transmission that requires a sequence number.

So if two lar frames are categorized into two types the first type is is called lar ineligible.They're just like standard ethernet you know regular ethernnet frames.The other type is called elart eligible that means these frames are protected in case of errors um these frames will be RET transmitted right.So how do they do that they actually put the sequence number in the preamble which is a pretty smart way demonstrated here as you can see.That put the the twenty bitch sequence number in the preamble here.Yes.And then?You know in emalla there's a sender and there's a receiver the sender sends packet receiver will'tell the ser hey I got it or oh no on there's something wrong so they need to exchange some schooling between the sender and the receiver.

So in inalard they um we have four types of um signplent messages they are control orders set they're not um layer to packets they requires some modification to the physical layer um yeah for new of controller set and a LA ack neckck in it and in it an echo.And two of them are used for initialization and the other two are used for flow control.In it in it in it echo for initialization a neck or flow control.So in act and act you can you can actually see the sequence number that is the sequence number receiver sent back to the center saying hey this is the most recent um frame successfully processed by me.And that sequence number is carried by both act and act the only difference here is upon receiving an actck the sender will start to do the replay or re transmission.

For a act we just confirmation.And actually the essence of Ella LA it's about replay or re transmission so let's just diing into this replay mechanism.So basically it's actually very simple the sender will always store a copy of an unan acknowledged anal eligible frame in the replay buffer.For possible re transmission because if it don't do that there's nothing to re transmit right.So we can see here there's a buffer.And we'put some packages in it.And on the receiver side the receiver will pre periodically send theckn acknowledgement back to the send say a I got this I got this but in case of something wrong it will send a neckck negativenowledment and back to the sre.

So when the when the sender receives a nack um it will just.Send all these buffer in from the replay all these packets from the replay buffer.Which is called ref transmission operation.嗯。Yeah and um this as I mentioned previously the sequence number carrieded in the in the act in the act that will be used by the sender to free up.The space in the replay buffer.So if you are familiar with um queuing theory you can easily probably already see that this can.Probably it is an m one model.We'LL come back to that later which is pretty interesting scenario.

So.But what triggers a knack to be as sent at the receiver side.So actually the receiver uses um two variables um to determine the nap.So the first one is about um.A missing sequence.A a a sickness number is a missing that's no brain that's that means there's a package loss.And a second variable is the fcs status bad or good.Why it's because if we look at this packet here um a corruption can happen either in this area.That will um cause f CS status to be bad.But if the corruption happens in the preamble or or s fiftyd you know.Um the f CS is not covering that that part meaning that you don't have an f CS bad.

But that will cause um a failed recognition of this packet of the receiver site meaning that hate a packet you know um a sequence number is'missing.So both these cases the receiver will send a nack.I guess you you'actually um have a question in your in your head like um but in case of a bad f CS um isn't the pack is supposed to be dropped.But um not add the Mac layer anymore not at the Mac sub layer anymore.That means in utratherate blank layer the Mac ah layer is is actually modified.Um the media axis control sub layer is not dropping package anymore in case of f CS error.Is actually a gift that to the EL alarm it will deliver the package all the way to the EL alarm layer.

Together with the um f CS status so that's why lr will receive that otherwise EL would never be able to receive any packets with a bad f CS.Because those would be dropped.Yeah.Um so to put everything into um together um I have this time um ah prot diagram for you.I hope this can give you a holistic view of how in lord operates so let's take a quick look at this.So at at the very beginning um the sender will send um frame you know EL elar eligible frame one two three four five right.And at this point the receiver sentence sentence a one.But you know it takes some time for the a one to to to arrive at the at the s side.

And that is actually what we called the round trip time you know the time it takes for a signal to to to go back and forth like this.Um so before receiving that act the syndrome knows nothing it just keeps sending.And guess what one of the packet number three is is somehow corrupted.And is is triggered the receiver to send a nack two right back to the ser.And bearram mind this is a sequence number and that is the most recent successfully received a packet um by the receiver so it got to be knocked to.Sent back to the center.And still before this the sender knows nothing it just keep standing so as you can see all these packages are are actually put into the buffer.So that is why if you have respect that is why there'there's a requirement on the replay offer size.

At least it needs to take um the bandwith delay product um um so that no packages are are getting.Left out out of the replay buffer.All right so upon receiving the act the rece the sender will remove that um that copy number one from que from the buffer um which is like this.When it receive a nag things are getting interesting.That would trigger a replay that means tos sender will Switch to its replay buffer and start sending all the packages from its replay buffer which is three four five six seven eight just keeps sending.And this is actually called replay window it is the time between um the first nonnacpack.Um to the the last EL EL elar eligible packagecent before the reception of an act these yellow ones.

On the receiver side um as soon as a neckck is generated the receiver is entering to this um silent mode um so is'it's not going to send any ack orna.Um until the expected sequence number is received.So so this period is actually called in a large discard window.So all the packages arrived at this window would be discarded by the receiver.Unless that sequence number is what the receiver is expecting.Look at here number three number three is.Is expected so the receiverrs sentsiment act three and then at five seven nine and ten so I'm so on.Yeah so basically this is a holistic view of how anal law operates.

Yeah.Let's move on to c BFC.So we know lard it's trying to um prevent you know um solving the link reliability issue and c BFC on the other hand is is preventing congestion um cost packal laws.Um if you if you know rocky v two you that you will know that in rocky v two priority flow control is used.Um to do the full control um to attempt to do make make the network lossless.Um but p FC is not good enough p FC has that that is because p FC belongs to um.One of the families um a flow control which is called rate limiting um the principle is very simple the sender will send um data rate is trying to match where the receiverg um can accept.

So this.From this you can you can actually see this when the sender needs to adjust rate and needs to receive a signal from the receiver.But it needs to receive a signal from the receiver.So this doram WSH actually shows you there's a signal coming from the receiver but it takes some time for the sender to get it right and that is an round of time the rt t.Which is why pfc is sensitive to round trip time.When the round trip time is large it would take longer time for the p f c to to react which means guess what there's a lot of in flight um a packet.Which actually means pfc can still cost of overflow at the receiver side.Um if you don't want to have to get overflow um and at the receiver side you will have um a larger buffer headroom um prepared for for this.

Otherwise yeah you will have a problem.Which means if in a in a data center if you change your cable length from one meter to two meter three meter r t is changed that means you need to re tu your your PC is all because.It is recimity.So cbfc on their other hand is um it belongs to another families is called authorization authorizationization sending.It's very simple the sender will never send packages without a permission from the receiver.Which means if the send is allowed to send that means that the receiver is ready to receive so there's no way for the receiver to get overflow.So that's why c bc is um is better than um pfc in in terms of overflowing um prevention.

Um it it is not as sensitive to rtt pfc.But it's complicated to implement.So this is how c receive actually works as a mentioned previously the sender and the receiver they need to um you know exchange this permission the receiver give give the sender permission the sender sent right and that permission in c bc it is called credit.So credit is um a certain number of bytees that the sender will use to calculate am I allowed to send this packet it before sending a packet it would do the calculation do I have the enough token or ticket or credit to send a packet.So after sending this package the standard consumed some credits.

And it needs to update that information sink that information to with the with receiver.When we when the receiver um consume um a package from this buffer um removed process so credits are freed.So the receiver needs to return that ticket token credits back to the center so that the centerre can keep sending.Which is very important.And that is actually called us a credit free update messaging.So actually in in the c BFC two signaling messages are are being used you probably already can see that the credit free update message is very important without it the sender will halt right.So that is why um the CF update um is actually implemented not by a frame by a layer to frame is at a SPY a control order said.

Making the total number of controller sets on link layer of abouteriesa in it and a two five four from ill a alarm one from CBC.The other one the credit consumed update um it's from the tender to the receiver you might wonder why.Um that is actually for a very interesting and special case that is called a credit leakage recovery.Imagine that DES sender send ten packets but one packet get dropped the receiver never received it.So if um if the centre doesn't synchronize the credit assumed with the receiver you know the total number of credits are lost.Because the receiver doesn't know that.So this um by periodically sending the CC update um you know we can recover from this credit lickage.

But that is less important and so the implementation of this messaging is actually um an ethernite packet.Of this.So one is even a frame and the other is controller it.Well in c b FC there are up to thirty to um VC virtual channel.And you can imagine that is'a buffer or a channel or a tunnel whatever right.So um these v CS um as you can seede here by default from the very start initially there are all best effort.The VC has only two types best effort or losses.Lots of VC we use this credit exchanging you know to to to make sure that there's no um congestion best ever it doesn't care it justsent if a package get lost yeah.

手。So um in syby there is about a lossness initialization addition and removal.Very interesting so at the very beginning.All these v CS are best effort so your switches my support I don't know eight bc or sixteen or thirty two yeah all all of them are best effort.Um then you will have to do the initialization of the lossless bc.Picking which one you wanted to use as of losses a bc and then what is the credit limit um what is the um the credit size you know ah how do you map a frame into this VC so.Which means some best effort will be tackled into lawuus we see here.

C BFC also supports while the traffic is running while your port is operational um you can actually add more losses VC or remove um some losses we see without stopping the traffic.How that done is actually like this.So an imagine that you have traffic on seven and six and you want to have another losses VC you just turn or or or or or convert um a best effort VC five into um a losses VC five.And then you can have traffic on them.There.Removal it it's the opposite direction um you can just you know toggle these loss of VC back to best effort but of course then you have to stop the traffic.So addition removal in in initialization is basically about one particular VC um um turning between transition between best effort and lobster.

So basically that's how on c bc works.Yeah let's take a look at some key um test tested scenarios for analar and CBC.So um bearmin link layer um anal large c bc that means these two protocolles are running at the link layer is not and to end like transport.It is ho.Meaning that the link is actually terminated at between two ports right.Not between two two notes only two ports.So what we're showing here is is trying to help you um test.How the link performs.So the the upper one here is about you have a dut and you probably have a third party device.

When you you guys are you know talking the protocolst and sending traffic in case of something is not working you probably need a traffic in line traffic analyzer to capture to give you like in depth.Visibility into what is happening on the wire.Who is doing doing things wrong right.And that is the first as scenario which is requires an inline traffic analyzer for capturing for debug Gang for analysis.The the second one here it's about um having traffic generator as an endpoint talking to the dut.So with this actually you can control you can mutate the traffic generator in to doing all sort of sort of things to give you many types of of test performance compli negative test.

So they have different youth cases.So testing link layery try.Um we categorize all the tests into three major categories the first one um is compliance.Um the u VC um has the ah very um.Thorrough you know compliance checklist that you have to go through them one by one for compliance test.And I'm just listing some of them here and for an lar.Um yeah.And once complying is done um then you can you can you can move to performance test for example having lah um enabled on your leg and then you start um shooting traffic.

But then the objective um the measurement of objective is not ill anymore is your traffic.Then you can see like how much you know traffic the through put latency jitter frame forwarding I can have.Yeah for compliance your test of objective is illary self.But um in case of um errors you want to have visibility you actually want to have a view or a knowledge of what is actually happening on the wire what is getting wrong right.Then you can either use the inland capture or you you can use the endpoint to do some triggered capture.SOI'm listing some um um some tested generator here for performance test you will probably need a traffic generator to to tois shoot despite directional traffic with your dut and bearram mind elard is by directional.

And for preamble test which is one of the compliance test that requires um.The t sense for example and non compliant preamble to the dut to test how the Du t will react right um so in this case you would need some air injection or for negative testing um yeah for compliance.On.And if you remember I mentioned the replay buffer is a very interesting scenario um yeah.To talk about it here.小。嗯。As you can see here I'm re um showing this diagram.嗯。All these packages they are une acknowledged package they're put into the copy of those packages are put into the buffer right.

And those copies will be removed by the act.So this behavior can actually be simplified into this CU model here.So you can imagine that if.Ancknowledgment is lost.That means for this period of time you don't have onn acknowledgement that means some of these packages will have to stay in the buffer for some time.Which results in some stress in the replay buffer you know the buffer size will grow.嗯。And also if somecknowledgment is getting delayed for some reason.Um that means you will have and or and um the interval between twock acknowledgement um arrived at the ther will be different.That can also cause the buffer size to to vary.

So in this case you can play you can use a lot of these parameters configurations to um to test how your anal law replay of performances what's going to happen if.Um ANA ah even um um an act is lost what's going to happen if um an act is delayed um yeah.So that means.On the traffic generator side of it you're going to use the delayed a or a drop you know.Act response time configuration to mimic all sorts of scenarios just to tested this replay buffer.For c BFC um is the similar pattern here.First we check the compliance test um you can get them from the compliance checklist downloaded from u VC.

And then performance and visibility.And here as I mentioned before we have the credit liakage recovery meaning that um some packages are lost so that the the actual received credits by the receiver is less than with the sender consumed.Um to simulate this scenario if you don't use impairment to actually drop a packet you can actually use you know um.Force a CC update just sent a CC update with a larger value.To simulate this scenario right.And for losses VI initialization addition removal um if you notice this back you definitely know that in there um you can either use manual configuration or use linked negotiation which is an in a more um.

Automatic way for the ports to advertise and negotiating conder themselves and the the whole process is automated automatic and then that would needs to be require ah at tested um authororoughly.So let's take a look at the tested solutions.So at telena lura here um for a from a traffic generator point of view um the solution includes um a z ah sixteen hundred Eden and the eight hundred trium and they do different stuff.Even is is supporting the um two to four the thirty s um.And and one one two gig of thirty as well um meaning that it can support um port speeds um from one hundred all the way up to one point sixty right.

And freya is only a concentrating on the one twelve gig thirty s four.Meaning that is's poor speeds ranges from eight hundred gig to ten gig.They have different um um spectrum here but both of them they support a treeys in that link layer and link layer we mean the the link layer rettroy um the credit base of flow control linked negotiation for both Eleanor in CBC and that is not enough.You will need a lot of um different sorts of air injection or protal configuration out of the you know compliant um range so and and also layer one capture.For you to have.All the tools you need to do the test not only the compliance test but but the performance the negative um and functional you know all of these.

At the ah protocol analyzer the inline and lir point of view we're offering to syreate and twelve eighty eight it actually pretty good um a powerful inline device and that cannot only capture but also doing the analysis decoding of um the archery and a or control order sets and and all these frames struck um um formats for you to see what is um going going wrong who is sending the wrong packet.Who is not doing who is not compliant to the ultra ethhanist specification.And this is very useful especially when you are talking to a third party.Device.You know noed traffic generator as freya or you't a third party then you'put this in between then you have all the visibility there for you to to do the complianceance test very easy.

All right so next I'm going to show you um a quick DEMO and followed by some qaning.So yes.To be sharing.Yeah I hope you can see my screen here.Yes.Yeah so um what you're seeing is is our dewei called za manager three and I'm going to use for simplicity I'm going to use two ports.Um ah one um s simulating center and the other is a receiver and in between I have the um m twelve eighty eight as an in line capture decoding analysis tool that can show you this visibility here which is controlled by disc school here.

So what I'm going to do I'm going to put on first I can.You know for this.Into record.And now I'm going to turn on at my l alarm.Um yeah so this is my send.I'm using fum um to do the demonstration so this port is sending traffic and this port is receiving traffic.I'm going to turn on elar as soon as it turn on the elar it's going to send the send's going to send elar in it periodically and we will definitely see something yes you can see is indexing.

I'm gonna wait a little bit.Let's just stop here stop the capture.I'm gonna make it larger for you yeah.So as you can see as yeah we have a lot of anar in it and um and if you click on one of them you can actually see the gooy is decoding yeah this is um a controller set you actually can see the X g m mi um um format for b.And the the message type of the controller said zero three that is four in it you know one two three four and c bc the CF update is one ten ah one zero in here it is um.嗯。

A decoded information for example you can actually see in Eleanor in it there's an in initial sequence number and and also the initial um data um yeah and what is control order is and so on.Yeah okay I'm going to do the recording again and now I'm going to send some.Traffic.Because my EL law is you know now up and running.The court good I have a stream and I have put the stream into elar eligible or desire stream meaning that the um um.The frames generated by the stream will be protected by anar.Um in in order to see something I should reduce the the package rate a little bit otherwise the um ah the cap will will be done very quickly.

SOI'm going to start this.And I can actually actually inject some f CS error on the screen you know so this would definitely um.Cause a um.A nap you know at the receiver.I can do again.嗯。So what we're seeing here is the statistics measured by the endpoint traffic generator and is group grouped by entity um the t X path in the a path.So from here what we are seeing is um of course um you know a send but what is interesting is hey we received twonas that that's that's actually because I collect twice on this f c s air injection on the stream um yeah.And that is because the receiver as we can see here it transmitted to neck.

Back to the center.And upon receiving that knack you know we have some replate frames.Only two that is because um I'm doing this at a very low haggard rate because I want to see the captured frames in m twelve eighty eight.Let's stop it.Of course if you um you can you know you can configure your captured buffer.But the lottery goes you know the more time you're going to have to wait in front of this um computer and for demonstration purpose I don't want to do that so that's why I'm lowering.The package rate.Great now we'can see something interesting to happen.So um this is a packet.

For example.Not that one.That one is a pack.And we click on that you can actually see.This is a elar eligible.Frame which means it has a special preamble.And where is it we can actually see it here the preamble.It shows you the started frame here preamble it's ah EL elll our eligible prem but it had it it it has um you know um ah a special value here and then foughted by a sequence number sequence number zero and this one must be ah sorry sequence number one.And thereafter that is the regular ethannant um um um Mac frame um the ethann header and so one and then at the end we have the end of frame here.So that is the ethnic packet.And this one is actually a controller set l a.

And we can see that it is acting a sequence number one.So this just keep going and going.But I'm trying to catch that.Neck moment it'gotta be somewhere here.Or maybe I can just do a future.Trigger position and.Let me try to find it.In in yeah when you're using this you can use a filter um to filter out the lr act but you get the point.

So this actually shows you not only um the layer two um frame a structure format it actually down to one um yeah you can actually see the controller said the preamble and so that you can figure out on who is doing wrong and what is going on.So.The back to it so basically that's a that's a very simple demonstration to demonstrate that um um um in a very typical um a case for elar with an air injection on the on on the stream with an fcs um era.So.That's it back to Martin.That's great thank you live for this great presentation and showing more about how you can do testing on the link layer on c BFC we have a number of questions um and please keep them coming you just go to the question mark in the top right corner and and write it in there but um there's a question here to you live and.

If the Ella eligible frame are held in the transmit side wouldn't that cause buffuff overload.If ah anal large eligible package is had the transmission siteight is is held on the transmision side so package are buffered up and while you're waiting on get your your acknowledgement.Oh yeah that is why um there's definitely ah ah requirement on on the l um on the elar replay buffer size um it cannot be too small it cannot be too large um if it if it is too small you won't have the enough space to hold.

Um the data that you transmitted um during this round trip time that I showed you in this diagram right.But it cannot be too large in the specification actually um gives you a very good reason why it cannot be too large.Imagine it is so large that is infinite.That means if um when you're in this replay you would just keep replay.Remember in the replay when the transmitter the the sender is replay packet um pfc um um and pause has no impact on it.That means?Your sender will keep sending um making p FC and and and passse um.'ass seembly not working.So in this specific and actually sets it should not be um too large um.Ah compared to the ah bandwidth delay product meaning that it is the ah you know the ports speed times the round of time so.

Um but if you send a lot of packet without receiving anycknowment.Um then there are parameters to control um are you still allowed to put um you know are you still allowed to accept packages from the oper layer and put them into the replay buffer or not that is called the maximum outstanding a sequence number basically a number of packages and also the um maximum outstanding data number of bits.So very you you you think this two dimensions to control um how many how many data can go into the rigpa platformuff.So yeah so yeah this there's a lot of um configurations tuning um around this replay buffer so that's why test in it is it's important.

Good there's another question here wouldn't the additional state machine import in Ella X and X had latency.A come again wouldn't the additional state machine involved in l r X and X at latency.Oh yes yes um e lard is not solving latency issue um it is.Is preventing.A palas.Um because actually um latency is really not what l are is's is is solving imagine that you have a a local loss at the link right and then you do reconmission if you if you remember the diagram before um when you do replay.

Um you actually increased latency.But hey there's no pack a loss.Which means on the upper layer the transport we'LL see um okay I'm experiencing some some delay.But I thank god I don't have a a pack a loss because if there's a pack a loss the transport layer needs to do the end to end re transmission that is super slow um so yeah.So.Yeah you're you're right um it will introduce latency.But latency is.Um and a large objective here um it is just.To prevent packaglos the latency cost by packaglos not the actual you know that latency um.Good okay thank you then there is a question do you see problems reporting.

By both I ALA and cpfc and does still time support both simtaneously yeah according to this specigate that that is a hard requirement from the UC spec um um I haven't touched on that part because of I only have one hour here Anna LA if we'VE seen needs to Co needs to interact or or a or Co work um together and um there is.Um a very detailed description in this in this specification um they have four different different scenarios basically for a packet you can you can you know.You have to.Um types in elar eligible in eligible right and then if cbfc you have loss list and best effort so two times two that's four so in total have four different scenarios um that you can um you can um you can check.

And what is the most important for CBC um to interact with a cowork with lar is the CC update.Because the CC update is telling the receiver hey receiver when you receive me the c update that means.Um the value carried by the cey update.Is.It's covers of the packets that you have received before me.So that means um when you add l LA on top of that doing re transmission you you need to make sure that c b FC sorry the CC update is not you know.MIS ordered with the the um the frame it it covers so there is a two um scenarios um um did did described in in in detailed um by this by the specification they use AB bc d um for different scenarios so.

If you want to know more we probably can have another weapon next time.So a followup question based on what you explain before so if lr X next are adding latency.Then wouldn't it take away the value propos of ultra ethannic objective being low latency high ban with solution for ARM army.You have to know this latency is um it's a link layer feature meaning that it happens between a port from a Switch.To another port of a of a Switch.Connected by a cable.Um that is so that time is so short compared to end to end latency that happens.

Instantly right.嗯。But also this ack andna.It'S.It's aonne it's it's periodically transmitted by the receiver.It's like you know if they have a rhythm here.So.Yeah so like for for for every um um window there will be an acknowledgement or.Negative acknowgment sent from the receiver to the centre.So probably your concern is only about the very first delay that is you know if you remember the diagram the very first rt t because after that um the receiver just keep sending the the act an act and that is can be configured um by by the management you know by software or by or by um.

Um by user so there is the minimist bascing and maximum space in between two successive and AC andacc um and there's some ah recommended um spacing that is measured in um bittes.I remember the minimum is four hundred bytes.To four hundred bytes.With the link speed at let's say one hundred gig or eight hundred gig so um that in interval is nothing.But but the round for time yes there will be a some run but that's just for the very start when the traffic starts when the you know the ports start operating um that effect will be gone.Good there is another question here how do you determine whether package laws is caused by a congestion issue or a link reliability issue.Oh yeah.There can be a smart way for for example um if you have c b FC if there's a loss you probably know that's probably caused by physical corruption.

But also it is very important then for you to bring up the inline device or a trigger capture to actually capture what is happening on the wire.So um so that you can see what is causing the package drop it is the um the corruption.Or it is the um the the congestion at the receiver because with the a conditional triggered cap in case there is a corruption happen um that would trigger a capture and that will give you the information right.And yeah that actually um.Meaning that um with some visibility um tool for you to see.That is very important for you to to to check who actually causes the pack lots.

Very good one more question and what are the most common implementation issue you see when validating ill are interability.Oh in in in interability um yes.That is when um for example when um different vendors they have different timing um maybe um you know my act is a little bit slow or.This.But first of all we need to make sure that everybody is compliant um they are implementing what what they are supposed to implement and then um some tuning of the parameters.And that all requires um the capture visibility you know the the inline device and um analysis and and so on.

Um and of of course um there are some you know corner cases around this you know state transition area recovery um and and timing and.Yeah and many many of this of these issues will only be exposed if you have the the faultt or error injection capability because you do want to use a traffic generator to emulate different scenarios to trigger that to happen and um with and also with them visibility to to give that information.When you're validating the innerability between different vendors so?Greatly thank you very much we still have a few more questions but we are running out of time but we will follow up with you um that have more questions in the line here um in in writing so but ah thank you very muchly for teaching us more about how to test a linked layer art for Ethan it so I'm out there thank you very very much for joining have a great day and thanks for joining in.

Thank you.Everyone by.