DASH Community Weekly, 4 Jan 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SONiC DASH Workgroup Community Meeting Jan 4 2023

Description

Back to business after (U.S) holiday break!
DASH ENI create/remove tests PR293 - Intel
Jan 4 Update:
Anton addressed comments, Prince signed off, Kristina merged
Anton is also making ECMP changes in the main SAI repo PR176 no longer need to use private repo ptf: testutils: add parameter vxlan_reserved0 to simple_vxlan_packet #176

Egress port selection Section 3.3.5 of SONiC-HLD Underlay Routing
Jan 4 Update:
* Number of VNET is a software limit as VNET by itself does not take hardware resources. This shall be limited to number of VNI hardware can support
** Support 10K peering in-region/cross-region

We need to update DST MAC in BMV2 - see 3.3.5 the 'neighbor' includes IP & MAC (implies that we need to implement some sort of underlay routing)

Change ToR to 'SmartSwitch' in 3.3.5 - Kristina made suggestion in Prince's PR

Discussions of default route and whether we will need another LPM lookup (for 2 default routes), whether we will be leveraging SAI for the routing.
Would be more optimal if we didn't have to do another LPM for 2 default routes (or to have an option for this). Complex tables, higher xn rates. If we leave this open and they make tables w/10k routes, they will need to reserve space and do the LPM.

https://github.com/Azure/DASH/pull/308 - Prince providing an overview of the new content
Located in the SONiC HLD at a high level Section 2.3
3.2.3 ENI Table pl_sip_encoding and pl_underlay_sip (to use in the underlay)
3.2.7 Route LPM Table Outbound we have 4 extra attributes overlay sip overlay dip underlay sip and underlay dip
3.6.2 Service Tunnel - 2 actions associated - 4to6 and action2 such as NVGRE encap, action type is servicetunnel, overlay dip and overlay sip
Check brief examples in the HLD

3.6.3 Private Link
ENI Table has 2 actions and pl_sip_encoding and pl_underlay_sip (common sip for all pl underlay)
Routing Table will have DASH_Route_Table entry, and a DASH_VNET_Mapping_Table entry
Check brief examples in the HLD

Key difference is that for the overlay-sip there is encoding that needs to be applied

A

New Year and thank you for coming to the first Dash meeting of the New Year 2023 um I'm, displaying right now what we talked about: um December 14th, just before sorry, we all went on the break and these were the um these were the different PRS. We talked about that were underlined here, just the different things we talked about, and um you know the conversation around them.

A

So there was 297 302, 301, 303, 304 and I did put these notes in when I post the YouTube video I usually put these notes inside the uh the description. So I did do that, um although I didn't send out notes, particularly that time. So that's what we talked about last time and because it's the first year I thought I would open it up and I'm going to stop sharing this screen. I'll share a different screen.

A

And I'll just share my regular screen here to see if there are any PRS or items you guys want to talk about today. I'll turn my camera off.

B

Yep, so we have one pretty old beer. Is this Anton yeah uh yeah 293 is always also almost on the bottom. Does she and I create remove those cases here? So that's uh the one that we presented a pretty wild ago, so I fixed the comment that there was too many fail files there. So right now, that's only two files test and the test plan. That's it so like. Please, take a look if there are any comments.

B

A

This is what we said last time: separate period: yeah: okay, let's take a peek, then.

A

Can you guys see it? Let me open this up a bit so Chris said it looks good to him. um I believe those comments were from those were from Prince and Gohan. Were they not Anton.

A

Okay, yeah, let you guys chime in then uh Prince. What do you think.

A

um So the last time we looked at this PR before we left for the holiday. um There were some comments from the internal Sonic Team about um creating a a separate file for common functions, making it a little bit easier to review, and so Anton feels that he did that.

C

Oh okay, okay, I didn't get the chance to look.

A

I think and then Chris uh said it looked good to him. So.

D

Yeah basically, Anton addressed the comments from yeah.

A

Okay, thank you. Prince.

D

Anton also is uh making some fixes in the uh ecmp area in the PTF in the main side, repo, so he'll be creating a PR for that that will directly go to the side- repo, oh okay, yeah, which will help us as well over here. um There's another PR. uh There were some uh issues in the PTF that we had fixed and uh uh made uh created the pr in the PSI repo and that's now gotten merged, it's the pr uh 176..

D

So with that now we are going to be able to use sire sub module within the dash Ripple, and we don't uh long no longer need to use the private repo. Because of this fix is snatched.

A

I'm just taking notes for reshma while you're talking so did everyone um get that and get the ramifications of that.

D

Yeah, so I will paste the pr number in the chat so with that fixed now that is being merged. We are able to use the side repo directly into the dash Ripple without having to use the private repo that we were using earlier for.

D

A

I'm just going to pull this up just so, people can put their eyes on it, real quick, so here we go so before laying PTF. This is the one guys.

A

A

um Anyone else have a PR, they want to cover.

C

uh So I have one PR Krishna like a before the vacation. We discussed about clarifying on the underlay routing.

A

Okay, yeah and that one was um I'm. Sorry.

C

The second one, the.

A

Second, one: okay,.

C

Basically, verifying what attributes we're using and how the underlying can really work.

A

Okay, so we need Marion to take a look and me: okay,.

D

Since this is updated in the hld right, okay.

A

C

One minor changes like the v-net number um Theory B net is not translated. Hardware is also the number was some confusions. The people were asking why 10 25? If you go back to the scaling session section.

A

Yeah yeah make it back there.

C

So I just clarified that it's a software limit.

E

A

Sorry for watching me scroll guys it's my first day back.

A

Number of v-net is a software limit, as Vena by itself does not take Hardware resources limited to the number of vni the hard work and support.

C

C

Clarified between the outbound route scale versus the inbound, but previously it was just routes perienna and it was happening and people were asking about.

C

It the maximum that we anticipate is like 10K I know that previously clarify I mentioned I said it has 1K Max, but then um there has been some other discussions where we could also have peering scenarios, in which case like if 10 minutes appeared then and somatic 10K for 20K at range.

A

Does that make sense to everyone on the call is.

D

The total number of routes now 64 times 100k yeah.

C

It's just I right.

D

C

Was always routes 100K, but now that scale is for the outbound, so it is 64 times 100k.

D

D

Versus IPv6, okay, that's possible yeah.

A

Yep I believe that you're still working on that yeah.

C

I think that's right! So how.

C

At this point, we know about the total limit included.

A

And Prince I have that filed in our internal Azure devops um for the sdn team and I can I can app mention you. If you want to see it or I can give you the number.

A

So we can look, keep looking have a marker that we're looking into it.

A

We also had I, don't know if this is related guys, but in that same conversation when we were taking notes, we, you know, we said prints, May update this section and and he did and then you know we need to update the destination Mac and bmv2 since we have Mario- and we want to talk to him about this or ask him.

C

Yeah, so the other section is the underlay routing part right. So that's what's updated um the the second line. Okay,.

D

Right yeah in the hld Christina, the underlay routing.

C

If you go to three three five yeah.

A

D

D

Happens if you can remind, or what is the new change.

C

To the pr, so we.

F

Can barely hear you Prince.

C

Oh, can you hear me now: yeah yeah go back to the pr. We have the the diff so previously I didn't capture the actual PSI attributes.

C

um So now the attributes are specified. So it should be clear on.

A

So please say: attributes here: yeah.

C

That one, and also in the description mentioned like how different attributes are, should be created and used for under layer, outing.

A

So go back to the files changed and.

C

Yeah yeah so yeah, the under layer, routing section yeah.

A

Yeah now we're back here.

A

Was there a question reshma about this section here.

D

um I think we wanted to um clarify in detail how the destination Mac will be updated, but we forward the packets um after the LPM lookup.

C

Oh wait: that is the regular neighbor table right. Okay, so that's that's why it is like neighbor land neighbor entry has both IP and the Mac.

A

Oops, so neighbor is IP and Mac.

G

So again, the Intel test cases use that writing and the program writing interface and all that. So it sounds to be the model that we would use here. Yes,.

D

Okay, so this uh implies that we have to implement some amount of underlay routing um the basic part that is there and that will come directly from what's there inside.

A

Thanks Prince any other questions about this PR guys.

A

H

But here just a question on that.

H

I didn't have time to look at this before so this might be a dumb question or something we covered before. It says in the text there that the tour advertises a default route to the Appliance. Is it? Is there ever anything except a default route advertised.

H

Sorry is it: what is there ever any route other than a default route advertised I.

C

Mean there is no uh restrictions, because it's a general, a generic bgp session between the plants and the tour right, so whatever tour can advertise, like maybe the Gateway or um or some specific routes, there's no uh no real restrictions there, but uh in the current deployment I. What I see is uh the tour just advertised the default.

F

Make this more generic, why are we calling it a Taurus? It's a smart switch. It could be a T1, it could be somewhere else in a network it. We always mention it like. It sits below a tour and that's it's.

C

F

C

I can make that change to.

I

C

That it's like the connected pgp peer or something like that yeah.

F

Or the network device yeah, it's more likely we're going to connect to the tier one. So this it yeah just to make it more generic sure, yeah.

H

So so today there are only default routes advertised, but in the future they could be specific routes and any specific number of.

A

Just thinking out loud I, don't know why we would manually program that would be difficult.

C

I think for the underly. We just need the default route. Right, don't need.

A

Default route points to one. If that goes down, then.

C

Yeah one or two like, or whichever advertisement.

F

F

Yeah you're you're breaking up.

A

One more time, Russia.

J

Oh look: I was sitting for bgp I. Think there was a document. I saw a few months back, uh requesting a thousand routes for underlay and I think with AJ 2000 Max, so probably in that range of scalability. From what I remember, I'll find the document and I'll point to it.

G

I got another question too: it's like sometime, we just Decap the packet, there's nothing in the flow, the airplane flow. That would say what to do once we process the packet. You know Decap and cap whatever it sounds to be that we go directly to writing all the time right, but sometimes we also Decap packets, right, that's not underlay per se, but we're still going to write them right.

G

In other words, whatever we do to the packet at the end, we is there any condition where we wouldn't route the packet or we just do writing period.

A

Well, my understanding is: we look at the mapping table and we look at the routing.

A

C

Oh sorry, what Christian.

G

Look we're talking about underly writing here, but sometimes the packet gets decapped as well right. That's part of the processing, I, don't recall the case, but there is a case where.

C

That's for the inbound routing right.

G

Outbound on the outbound, we never Decap on the outbound.

K

There is a road Direction right.

G

So, okay, let me rephrase anyway, he's riding on outbound the only ever action we're gonna have.

C

On the outbound, it will be only routing okay,.

A

And that was um who was talking there.

F

Why do we I'm trying to figure that out I thought we did everything on the outbound that we do on the inbound, including apples Etc? We don't.

D

Michael should be there yeah.

L

I think the question was about the encap Decap right: okay, yeah.

F

It's true: okay, yeah.

L

Yeah but I I believe you know we we do have to do end cap depending upon which way we are talking about outbound right, whether it's a host or the network right, so we have to really specify which outbound we are talking about.

F

Trying to figure out which action you're talking about that you're confused on, because it kind of goes in a Blow I'm trying to figure out what? What is your question? Is it you want this flow diagram changed um I'm, not sure what it is that you're trying to get.

G

uh Me, oh, that was.

K

G

You process the packet and then you're gonna sign it somewhere is the ascending somewhere? Is writing only or is it going to be something else it sounds to me. We just do writing. That's all yeah yeah right now, there's no option anywhere to say to give options anyway of what you can do so.

C

So I think Vincent. There are two parts right: one is on the uh actual packet where we will do some sort of uh encapsulations and then those actions are part of the the overlay routing tables right and then you have the underlay for the final packet forward.

C

So in that underlay it's a it's only routing right, right, yeah, Mr,.

F

Thanks Prince, do we need to clarify the flow diagrams at all to uncompute that.

A

Yeah, if we do please let me know, I I drew this one, so I'd be happy to change it or improve it.

A

Vincent. Just let me know.

F

I think this is the overlay, low table and I think what we're talking about is once this is all done. You still need to forward it, so you wrote it to the next Hall. That's not in here. Probably, let's.

A

F

A

You know it's here, create the flow.

G

It's the forwarding side of things when you're all done with this processing. Yes, it's correct.

H

So on the underlay routing in in terms of implementation, there's a big difference between having a default route or maybe two default routes and um like a thousand routes um in both like computation and overhead um I, just want to make sure that there's actually a requirement to do something other than a default route or two default routes um and the the statement that we need um more than that, like maybe a thousand routes or anything other than a single round, um is an actual need and not just being done for completeness.

F

So I think he already answered that we don't use it. We just have two default routes, um but the the actual uh open source document here it shouldn't prevent somebody who might have a need for that, but we've already answered it with. We use two default routes, but that's just Microsoft, though, because this can be used. This can be used by by other companies um and if, if it is, they might have different requirements, but ours is two default rows.

G

Are those before what are they provided to the control plane through the next up and router ID, and all those things that we just talked about.

G

I guess I'm gonna uh I'm.

F

Looking at is there a specific site API for those default.

G

F

Because I think this is part of a bgp session, I mean yeah.

G

Yeah, but it needs to go through side to get down to our control plane. That's what I'm getting at and and are we basically there using the router interface and whatever, like those Intel test cases, are using.

G

To program the dmac and S Mac for the rod.

F

So you're saying that it's not defined in this so I have a path. That down is that, where.

G

So, in a generic way, when you're going to forward your packet yeah.

M

G

Gonna look you're, gonna, look up those two three tables for Rider interface and whatever I. Don't remember those things right. Next,.

I

G

Router interface right, but that's fine, that's generic, and it will work in this program through the side with the proper apis. We just talked about this like 20 minutes ago now. If we only have two raw two default rods, we may want to optimize the data plane for that without doing like three lookups to just get to you know we can optimize for that.

G

What I'm questioning is the two default rights you're talking about how they're going to be pushed to the side, how they're going to use the same network, um neighbor, router interface and next hop definition in the site level.

C

So so, when we push the default route, it will have, let's say say: if you have ecmp right it, then it will be like two different next top for that default route right, so it it will be pushed by by the PSI with two next stops for the same default.

G

So so it will use the sign next up definition, rather if it's definition, another okay, yes,.

C

D

So, just for two uh routes: um we should see how uh you know whether we need another LPN table, because here we have a complex LPM table that we just saw in the other diagram which includes eni, along with the destination IP right to look up so yeah in the pipeline for the underlay. Do we need another LPM lookup? Those are the sort of things we should figure out just for two rows: foreign.

F

Maybe let's uh take this as a topic item Christina, because I'd also want Gohan here: I don't want somebody reversing decisions like after the fact wants this whole conversation over again.

F

So when we get um the people here, I think that we do need to resolve this I think we hear that it would be much more optimal if we didn't have to do another longest prefix match just for two two default routes, but let's make sure that we have the right people in the room to make that decision or to have option, for that uh is another way to do it. But let's uh take it as a topic item, so it seems important enough to we can take it up in the next meeting yeah this.

F

uh You just need to take the note to take it up and make sure that the people are here. So we can close it yeah.

A

I got it: okay, I I am typing, as we speak so to be super crisp. Then you know we we just need to know. We need the discussion and the answer the question about when we're optimizing. Do we need to have another longest prefix match table for two default routes and the ramifications around that is that right, guys, yeah? Okay, thanks? Okay, thanks got it okay and then, and then.

F

um I can't eliminate a complex table, so it can, it can create higher Connection rates is what they're getting at and it doesn't take as many resources. So.

F

And um and they have to make tables that are um like 10K routes and anything like that, they have to reserve this space in the first, yes and then, and then you have to do the LPM in the second place. So it doesn't. It doesn't seem up to me.

A

Makes sense- and we only have so much memory and resources yeah, and then we did answer the question that we're using the sinextop API, so, okay, okay, so uh any other PR's or something you guys want to want to go over today. Sorry.

C

I think the next one we want to discuss is the service tunnel and okay.

I

C

B

C

Any other priors, because.

A

Sorry, if I sound, distracted, I'm managing like four windows at the same time, um but is it this one right here the service tunnel route tag.

C

Yeah, but before we go there, I.

H

C

Would like to give an introduction on the service tunnel and.

H

F

That'd be awesome Prince. Let's do that.

C

Okay, okay, so I'll share my screen.

A

Okay, thanks Prince.

C

Okay, can you see my screen yep? Okay, so uh this one is part of the Sonic actually as well I think um we also have some other um details section on the service tunnel and private link packet processing, um but in the Sonic hld it's a tab at a very high level. So we have this section for v-net to v-net packet flow, so service, tunnel and um and private endpoints, like private links, are kind of uh the next phase. It's not the v-net to v-net like it's, not that regular winner to unit scenario.

C

So it's a it's uh solving other customer um uh customer use cases and and scenario so I like I'll go over the the Northbound schema changes and how the packet flow happens, but at a high level the service tunnel is for, like you know, um uh like different customers want to access a common shared results. So, let's say like the storage right in this case uh there will be multiple, um uh multiple beeps that are assigned to this first endpoint and customers can access it through either the public, reachable or the internet.

C

Routable endpoints, that's the service tunnel scenario or like within a v-net like using the the private endpoint. So again like uh what there are like documents in the Azure about what is private link and what is service tunnels in in this section. So there are like more details on how the uh or what are the uh the different cases for that, uh but, like I said like at a high level, it's like accessing a shared resource, either through public endpoints or through private endpoint right so I think.

C

Maybe when we go through the uh the uh packet forwarding section, it will become more clear. uh So the one thing I want to highlight is um when, because it is not within a v-net where the uh the end point uh may want to install apples or those things based on- let's say the encoded, v-net ID and region ID. So that's why. In this scenario uh there are actions that kind of encode the region being at ID, subnet ID in the in the packet itself.

C

So this is kind of a transposition from ipv4 to IPv6, with uh different encodings Okay. So yeah I think it's a it's a bit of confusing. But uh let me let me go over the the details, but um do you want to extend on on what is service tunnel or private link scenario or.

C

Okay, so let's move to the the changes that that is being done to to achieve the service tunnel. Private link, forwarding right so for uh one is in the eni table. We will have the encoding for the private link, so this is only specific to private link. So there are two cases right: one is service tunnel. Another is private link. Service channel is the routing to the public, IP and private link is routing within a v-net by uh by uh associating a private IP to the public endpoint okay. So let me explain that in detail.

C

So this is one is the encoding. Another is the uh the sit to use in the underlay. uh Then let me come to the routing table so in the routing table. This is the main change right, so we know how the v-net to V Network right, um based on how it is explained in the example section, but for for the service tunnel. We have these four extra attributes that are being added, like basically you'll specify the overlay, sip overlay, dip, underlay, zip and underlative.

C

So so uh so overlay ship will be like an IPv6 transposition from the uh from the ipv4 packet and similarly overlay dip is also an IPv6 transposition uh from the the ipv4 packet, um maybe like what the example will be will be more clear and we can come back to this uh so the same routing table depending on the action type. If it's let's say, if it is service tunnel, then it will pick up all uh all these attributes. So so that's why it's optional so for regular v-net.

C

These attributes are not mentioned, uh not uh not needed, but for a service tunnel. This will be the one that takes effect. So the two changes are in the tables are one in the routing table, LPM route table and another eni table for the private link. So let's go to the the example section.

C

Yeah, so here is the service tunnel example right. uh The route table will have an action called service tunnel. There are two actions: Associated one is a four to six like ipv4 to IPv6 transposition and the second action is the encapsulation either nvgre or vxlan.

C

um It can be either one but I'm having the exam Plus in which area, but remember in v-net to v-net case. There is just one action which is like map routing or static and cap, something like this: a single action right, but for service tunnel. There are two actions: one is a transposition and then the other is encapsulation.

C

Now the route table will look like this. You have a packet test into some IP and then the action type is service tunnel. Then here we specify the overlay, Sip and overlay dip right and then in in some cases, we'll specify, um but.

C

Depending on whether it is available or not, the packet Transportation will uh will happen accordingly. So now one example is: let's say: the packet is just a packet is coming in this eni um this time to 51 to one right so 51 to 1. It will hit this entry and the action is service tunnel.

C

So uh the first action for the service tunnel is four to six transposition, so you get a packet from 10 111 to 551 to 1, so the it it first gets transformed with overlay sip, uh which is being captured here with colon colon 0.. So the the last 32 bits of the ipv4 get transposed to this overlay sip right. So you have um you, have an overlay sip with, which is an IPv6 and the last 32-bit is here. Ipv4 uh destination like this is uh what 511.

C

um Now this is the the Sip. So so it will, it will be like ten one one, and then you have the overlay dip, which is also mentioned here with colon colon 0. So the overlay dip will get transport the transposition to the first, like 64 bytes like this, and then uh your last words will be um will be the destination IP. So this is the first Transportation like four to six and then you have the second action for the service terminal mentioned as nvgre.

C

So you you encapsulate the packet with nvgre and then for the nvgre. You, you need an overlay. Ip I mean like um the under Lacy pen, tip right and you get that from this table itself, like under list 41 to 1., right uh until I say pass 41 to 1, but since there is no underlayed dip specified, it will use the packets IP that the destination packet um IP itself. So this is how the the packet transposition happens.

C

Okay and and the other example is 6121, where uh you have both underlay Sip and dips mentioned so the same kind of IP for ipv4 to V6 transmission, but for the Sip pan dip, you have the um uh you have the the attribute, so it shall use from the route table.

C

So so this is uh about the service tunnel chain uh packet, forwarding and uh before we go to the private link. uh Any question here.

A

No question from me: it looks great.

I

D

Far, we have discussed mostly vxlan. Now we are discussing nvidiary as well. Yeah.

C

So so in in in some end points right where we send the packet, it really supports uh nvjre. So this is like um we might have both support like nvgre or vxlan.

C

H

I have a question on the um the case where the underlay dip is not configured.

H

um It says uh line F, it'll use the destination IP from the packet yeah yeah is that is that the the uh overlay? Yes.

C

That's the original packets destination will be like this one right. Six, uh sorry, this 51 to 1 right.

H

Okay and the 50.121 is the.

H

Okay, so you're saying if the um if the desktop isn't configured then use the overlays guest IP is the underlay? Definitely okay. Okay, thanks.

F

We have a way to make that clear, because since we've been over that a couple of times, maybe we should just put it in the document.

C

uh To use from the overlay right.

F

Yeah, if not present, yeah yeah, and that way we won't get that question again.

C

L

Friends, I have a quick question on this uh routing table, so will um other v-nets routing table uh look the same? In other words, service tunnel endpoint, the shared resource will be accessible uh using the same IP address.

C

Service tunnel endpoint see, if you see these, are public routable endpoints. So and if what was the question, so it.

L

So my question is that these uh these public IP addresses will be the same for other v-nets as well right.

C

For other Enis.

L

There have been just just let's say: there are multiple customers right right.

L

Yes, yes, we'll be using the same IP address. Yes,.

C

It could be like the shared resource can have multiple uh multiple whips right. Multiple public IP addresses, but it is possible that one um multiple Enis can have the same destination IPS, but probably the overlay ship dip. Things will be different depending on the v-net ID.

L

Right and then the.

C

So when you make this transposition, you will create a flow right for both outbound and inbound.

C

Okay, yeah, so for the reverse path: we'll have to do the reverse transmission, I, think um yeah, I I'll, take a note and I'll also update the the reverse path. I I only briefly mentioned at the in the section above like because it's it's uh there's nothing from the configuration side for the reverse path, but based on the outbound flow you'll create an inbound flow for for this.

C

Okay, all right thanks yeah! Thank you.

D

um In the tracking table um this will in each Direction it will be the addresses that are completed. The transposition right.

C

D

And uh all the Enis will have the same destination IP if it is a, um are private within the same v-net. If it is public, then of course any v-net can access it. With that IP address.

C

Yes, yes, but for private yeah, I I'll come to that it's part of the mapping table, so it's like within every the same service can be like you know, having a public private endpoint and um it will be more like a winner to be in a scenario.

C

So the private link is but.

E

What is the UI that going to be used.

E

For the for the service town on it, yeah yeah, so.

C

The key is mentioned as part of the uh that.

E

Oh, the key is mentioned: okay, okay, I see.

E

But for the reverse Direction flow, what would be the Packer look like? Is it going to be encapsulated or not encapsulated.

C

I

C

From my understanding, it is encapsulated with the same uh like same as how we do the outbound. It will be similar encapsulation.

E

You also have those uh you just underlay, sorry, this c-pan D, sorry, the overlay, zip and deep.

C

B

E

So when we, when we try to create the tracking um we're tracking the V6 uh connection,.

E

When you see we're creating the tracking table, so what kind of uh IP address you are using.

C

Retracting the V6.

E

Well, we're tracking the V6 yeah.

E

Okay, that means that the um so when we try to create a connection tracking, we use the use, the overlay V6 side, key plus uh class support number to the to create the connecting tracking table right. So yeah, okay,.

E

Okay is that is that mention a document yeah.

C

I think the inbound is based on I just uh had a like a high level statement about, like the connection tracking to be based on the outbound flow. But um yeah looks like I I'll clarify more on this.

E

Because I don't know like to look at the pipeline, I, don't know when the connection tracking table being created is it. After all, these actions we create the connection tracking table or before that.

C

So if we look here um in the flow, the it's after the routing the connection track, update is done. I see I.

E

See, oh makes sense: okay,.

E

Okay, okay, so maybe this particular part we need to uh you know, make it more specific, so that the people know that this is tracking the modified Transporters that had it right so.

C

Yeah, okay, so.

C

C

um Yeah the next one is about the private link. It's uh um again like.

C

uh It's kind of an extension uh to the to the service Channel and and enables the customers to access public facing shared service via the private IP addresser. So this will be within their v-net right. So so it's basically same as um we need to win it, but there are kind of encodings that are uh that are required here. So it's like slightly complicated than the service tunnel. So here one example is you have the eni table?

C

You have the routing action type for private link again like two actions, four to six and then in vgrs encapsulation. um Then you have the routing table right. Sorry, this is the eni table and- and the encoding is like this- so let's say: uh there's a field um offset size and value the field offset size and value. So if this is the encoding, it is common encoding for all packets.

C

uh That is going out of this eni right and then here you have the underlay ship, which is which is a common um zip for all the private link under AIP. Now the route table, you will have a an entry like this. That says, like you know, 10 say: packet is 10 to 10 to 0. 0. Then you have like you know, being a direct we in a direct and um and and here you can specify the overlay IP. This is like the regular winner direct example that was captured as part of unit 2 unit.

C

Now, the thing here is when it goes to the mapping table. So here you can see the mapping table. uh You have a 10 206, a specific entry for the mapping table entry and then uh the routing type and all the remaining things are specified here. So let's take an example like 10208 from 10 111, so LPM lookup hits for this table right and then uh the action is being a direct with the mapping, Loop lookup key of 10206 right and then uh the action in the mapping table is private link.

C

So in for the private link, there are two other actions like four to six, so four to six times position. uh This is what will happen for all the private link cases so uh for overlay. Sip uh using PL sip encoding like here is the the encoding is captured as part of eni.

C

Like it means that, from this overlay, zip modify these fields right, like field 11 um length of 1 with value one, so it will change like if you notice here, um where is that yeah, so it it changed to fd30 instead of like fd40 PK 530, because uh the encoding is mentioned that and over here like what to use and similarly the 48th uh uh offset and and size 48. You have this value, so you change that value here and then. Finally, the last 32 bits is again like the same mechanism.

C

You just replace the ipv4.

G

C

With for that last 32 bits, so this is how the the transposition happens for a private link and and similarly for the destination IP. um So here, like you, know, overlay dip, uh it's again like sometimes you will. You will be provided with the last 32 bits uh as part of the overlay dip. In that case there is no replacement of ipv4 to IP, uh like IP referred to the last 32 bits, but um uh if it is provided you just directly use that.

C

So these are the two four to six transposition, one for the Sip and another for the dip.

A

C

And the second action is same similar to the service terminal. It's just an ngre with the key here and, um and you have the underly dependently ship from the mapping table. So you have like for this example and underlay IP is mentioned, so it uses the underlays. If from this mapping table and the Sip is part of the enl, like all private link, will have this particular see.

C

So yeah the the key difference is for the Sip overlay. Sip there is a. There is an encoding that needs to be applied.

F

Is this clear to everybody, does yeah or are the people on the call familiar with these services.

E

So can you go back to the winning direct mapping table so the IP address is 2 8 and then we map to dot six right so either I just want to make sure. That's.

C

Not yeah, that is part of the v-net direct example. Here, I'll go, and so let me show you that so in the winner to win it, we have different actions. Right, one is, uh let's say: v-net right in the v-net, you don't specify the overlay IP to look up, but the same thing for the winner to win it. If we specify the action as winner direct, then it has to use the overlay IP.

C

That is mentioned as part of the routing table for the map lookup. Even though the packet is designed to this subnet, the lookup happens based on the overlay P from the route table, and that example is also here. um Yeah yeah.

C

So Vena 10 winner direct, is the main difference is one the lookup happens in the mapping table based on a specific IP. This is like kind of.

E

No I get this problem. Is there a case that uh you know um somehow, you know they? They they don't go through the vinegar and they directed winner winner, but the destiny. Actually. Is that the thought six? So then it goes to the mapping.

C

I

C

Note that I'm aware, because it has to match this subnets right. So if, if that subnet matches the base and and then that's where the action takes place, so.

E

Oh, but what if the you know the digestion, the sender test and teaching IP address directly to that 10.2.0.6.

C

For this example,.

C

E

Yeah, could that be a case where you know the ascent direct okay? If you send directly, you will match that route.

C

Doesn't match this yeah.

E

Yeah, okay, so you are saying this overlap. You will always be wishing that within the range. Oh.

C

C

um That's my understanding, but if let's say if it is not uh the same, then it will not match this right. So um I can clarify that, but I think it will have then another entry for that for this. But this one to match for some routing table.

E

Okay, how many, how many uh those encodies uh could we have? Currently you have like two encodies right so could? Could we have like three encoding four encodings? Do they have a do? We have a limitation on that part.

C

uh Yeah, so so this encoding is what you're saying right: yeah.

E

C

It's very ni, and here you can have any number of encoding.

E

C

E

Should be maximum 128? Yes,.

C

Maximum yeah I think maximum would be like 96, because the last 32-bit is always the IP for IP.

B

M

And friends, will you um share the exact encoding of the service tunnel? Ipv6 addresses like what the fields mean and where they are located. uh Dash repo doesn't have this information.

C

Yeah, so that will be.

E

um No, no, that part I'm not sure that we want to make it generic. What do you mean by exactly like.

M

The IPv6 prefix has v-net ID some other fields then called it in it. But do you want to make that standard, or do you want to just leave a generic V6 prefix for uh people to use it as they wish.

C

Yeah that that site married like we do, we want to make it as a generic encoding. So don't want to specify this level is, let's say, vnet or those things.

M

So we have uh like 96 bits of uh apparent data that anyone can put any information. Okay,.

K

M

Tunnel there is no uh encoding uh per unit right.

C

This is only only for PL yeah only for PL for service tunnel. The encoding is uh automatically detected by the North Port and.

E

C

E

Okay, so so I'm wondering for the encoding right, so whether there could be a separate format say you know, you just give a mask under the beat, you know, 128 mask and the beat, and then you know, then you don't have these separate fields that that says right. So.

E

You always give the full mask and the full modification stream and they apply the mask. You can apply to that right. So.

C

E

C

I think there was a challenge if I, if we put the full mask.

E

um You mean challenging the vendor implementation or Challenge on us.

C

On us so I think.

E

I saw you the same right, so we we just specify whatever the Beats need to be masked out and apply the values.

C

Yeah so, for example, this one means like 11th, offset one one bit and change it to this right yeah. So you can.

E

You can have the full mask right, so you know you put the 11 for the bt11, you you say one and then there's another one. You know fail the value right, a mask and value. Do you have one mask? That is a full 128 and you have a you: have a value for 128, for whatever you don't use you set to zero right. So then they don't mask out for the one that you you want to mask out and apply the value you apply it.

E

It's the same thing. It's just.

C

Yeah I think it's the same thing right so.

E

But maybe it's easier for them to to to to pass it right. I, don't know what what the hardware will Implement, because they won't apply those fields one by one right, so they probably just uh you.

M

Know yeah mask is definitely easier to understand and as to and program the hardware.

M

And same goes goes by the way for service tunnel, because they saw that uh the last four bytes could be zero or could be failed and now I need to look into them and figure out what I should do. So if there was a value and mask again, it would be completely transparent.

M

The operation often coding.

C

Oh so here, if I put like slash, 96 versus Slash,.

L

C

But see these are all Source addresses as per the schema. So that's why I didn't want to put prefix and.

C

And maybe like to the okay I.

M

Think but you you can do another field uh overlay sit mask.

C

Yeah I mean otherwise to the side. We can provide the mask right, so you can Define.

K

It that way, yeah.

C

But for encoding yeah, let me we will and revisit this yeah.

E

And when you apply this in Kodi, that or is that just apply what, if the fields on the original uh one or non-zero, are you just replacing it or are you doing the xor or doing no.

C

This is just replacing uh if it is already one, then it is one.

E

So maybe we need to clarify that.

C

K

A

I wonder if if this oh go ahead, Vladimir yeah.

K

Yeah I just have one question, so are you planning to update the dash pipeline in relation to the service tunnel currently.

C

Yeah, so that is one PR by Marianne here, I, don't know if you have time to discuss that today,.

M

It only has outbound, uh so inbound will be done following more details and Prints. But here you can find the Transformations uh for the album, like what you mentioned, four to six encoding and then uh encapsulation and two options for vxlan as well as nvgr.

K

Yeah, okay, okay, I will take a look thanks.

A

Yeah I'm wondering if this is a good breaking point, we're five minutes over um hate to be a PM but I'm sure you guys all have other meetings. uh What do you all think Gohan Prince Gerald.

E

No I think it's good I mean it's uh I, I believe there's a lot of information. People can can review off like.

A

There's a lot and they're wondering we should save some time next week, maybe um for you know, 20 minutes or so to let people take the sand and go over it, and um how about that guys?

A

Does that work for everybody, okay, great um I'm, gonna! Stop the recording, I appreciate all your time. Thank you for coming, um especially the first lighter week after uh after the new year and everything so happy, New, Year everyone and I'll see the behavioral model people tomorrow in that meeting. Thank you thank.

B

I

M

Bye-Bye thanks: everybody.