Internet Engineering Task Force 93, 20 Jul 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF93-GROW-20150720-1850

Description

GROW meeting session at IETF93
2015/07/20 1850

A

Good afternoon this is the grow meeting. If someone, the back, could kick the doors close and lock everybody else out. That'd be great those people on the very far as your facing the slides right side of the room, there's lots of space on the left side. You could come to the dark side.

A

Ok guess not come to the dark side. We have cookies, ok, so I'm Chris! This is Peter. This is the grow meeting at IETF 93 in Prague, if you're, not in Prague, that you're in the wrong place. If you're not the grow meeting, you're, also in the wrong place, here's the note! Well, it's new fancy, no toil it's much shorter! You can read it quicker see. It's all done.

A

Yeah we're going to play AV games for a second.

A

Anybody has anything that should be on the agenda. That's not on the agenda should speak up now.

A

A

Look that's good. I, like your focus, seems like it's totally on on schedule. I just leave like that. That's fine, all right, so we'll have a great display. um So we have a draft status thing. Then we're going to go over a couple of the current drafts there in work for people to get some updates. Two quick things: John Scudders, not here, but his his draft he sent in instead of comments. It's think he's almost all done for we're in we're in last call at this point in which ends a week after the ITF ends.

A

You all should make sure to read the draft. There's a bunch of invitations already in the field, they'd be good to make sure that what the draft is and the implications are matchup and for the ordering we're going to have job go before shriram, so we'll see thomas, then job, then shriram. So let's go over the draft status oops, alright, so the filtering threats, there's an update coming you. The last call comments. I think this is the one that actually got comment gotta rev in today, so there should be a new version of this today.

A

Eric had some pending comments from a discuss from IETF class called the air-handling document is there's an update coming for that. Those folks have said that they've they forgot and then we'll get back to it short.

B

A

The route leak problem definition is what sure I was going to talk to us about today.

A

The bgp shut up- I talked about this orgy shet life. We talked about this at the last meeting. I, don't honestly remember, what's going on with it, it's expired and I have to pay my offers or, if you're here this is this is the one that lasts is a venom on deck for ever and ever, and it's waiting on something else to get done.

A

Oops jobs are psl via he's going to talk to us today about that. The routing policy considerations Oh sorry. This is the one that just got to read today, Eric put in the comments and today he's had a brand new baby, so he's been busy with other things and the route server operations thing is in the editor: q. That's actually waiting on someone on the eye. Dr version.

A

Okay, so we have other presentations so Thomas. First.

B

C

Do you hear me okay, good good, so my name is Thomas King and I'm, talking about the black hole, xpppp community for plague holding at Ike's peace, internat rafts? That is joint work together with some folks from the aurochs community and the mother folks. Next, please, let me quickly motivate this works. The problem is, if you are connected to an IXP and hit by a massive dealers attack, it might happen that you experience is people congestion, which means that yeah that Majesty does attack contrast.

C

0 plot at I, XP and legitimate traffic will then be disturbed by that next epic can be reserved by using black awning, which some XP is already provide, and the idea is that you trap the black holding traffic already on the ice P platform so that it does not disturb with legitimate traffic and your IX people are congestion is resolved. The black holing itself is triggered by PGP announcements.

C

You, you add to your already working pitch p announcements, and usually you take a more specific for that next and the trust actually is about standardized or a commonly agreed friggin triggering mechanisms for play-calling.

C

The reason, if that, within the authorized community, we looked at different trigger mechanisms that are already available at IXPs and we have trusted actually a 5x piece, and you see that there are five different trigger mechanisms so for our customers, it's quite cumbersome and complex to to use this triggers at a different IXPs, and you also see that sometimes the three mechanisms collide with existing policy controller traps are long mechanisms so, as I sets, the idea of was to to draft a proposal which defines the commonly agreed way to tray up like whole.

C

Next, for that we had some discussion or on the irex irex community for that mainly into Ryan August last year, and we finally came up with with a PTP community that is 65535 trade 06. The reason is that 655 switz 5 is a reserved, a zen and the whole number is in the well-known PP committee space and not used. So we might ask iana to provide this number to us, and the good thing is that it doesn't conflict with any existing trigger mechanisms. So we can have those during a transition phase on xix peace.

C

Next, so um then what happened as yeah as I said, we drafted that internal draft is ability to the working group and then wait quite some good discussions about this concept under ideas. And let me could you summarize what we discussed so far. First trip snarls, Christian diet and some other fans are always committee already requested to extend to be gb community for trading black holes at XPS and ISPs, so that you can also use this black hole community for throwing black holes that you're trans provider upstream tomorrow I'm from the XP communities.

C

So that's completely fine with me. If that helps, we can do what I'm happy to to do the work for that, but I might need some guidance from the transit isp guys so I'm looking at job, especially.

D

Job Snyder's city I'd be happy to help. You thank.

C

You that was fast good earning then. The next comment was from shops, mallets and Treasury house. They requested that p2p speakers receiving a black hole announcement must add and no expert on or advertised if it's not present there to that announcement. I think that the reason they will request from my point of view, we should add such a statement in the traft already in the draft.

C

Is that, because, should add and not must add, so we should clarify that they have to do that and but before we can finally do that we we have to find a solution for the third proposal, because traffic are sequestered said only that the black hole behavior should only be permitted if and only if it also contains no expert or advertised. So my point of view, that's a little bit little bit harsh, but in the end, it's up to you guys it's working good position.

C

If you want to have that happen, we can do it, but, as I said, my personal flavor is on the second proposal. So any comments on that already.

E

Which, if used in a panic I'll, go actually go back to your example and why you chose a is 65535.

E

How soon we forget, because I'm RFC 7300 of show june or july 2014 said: don't use that particular AS number and a couple of others just do not use it. So if you are going to pick one that was a bad choice, Bob's try a different number. Ok sure we can welcome that fat, but.

F

Julie I wasn't going to comment on that, um but yeah I mean this is something that will be coordinated with I Anna when we actually get around to the point of picking a number, so um I think we'll be cognizant. Mindful of yeah existing existing art.

C

Thanks for the comment, and can you please do me a favor and send me a hint, this particular of see, because I am not aware of it so that I can find it that would be nice. My name is Thomas King thanks good and the next step would be rocking group adoption. If that's fine, miss working book.

G

How many people are in favor of the strap being adopted as a working group document.

G

Okay, I think there's a. We should have some discussion on the list and discuss the draft and go from there but I'm guessing it's a general consensus that that should be adopted. Thank you.

G

D

Hi everyone, my name, is job Snyder's, I work for ntt communications and I want to talk with you about the current status of the RPS elvia attributes.

D

This morning, I took a look in the 13 I, our databases that we near Earth entity and I found 293 entries that refer to either export via or import via. So to me, this kind of proofs that people are actually using the methods in the world and I already the software that, for instance, already be an entity used to.

F

D

Like him better soon, all right and there I rd, supports it and it's deployed major sites. The right who is server also will accept attributes and do some syntax checking on what is presented, and it covers the the two major irr server implementations that are out there in the world and there's also an actual organization that does interpretation of these attributes, namely the MSM internet exchange route servers.

D

It is through these attributes that you can control what is either exported towards your ASM or how you want your route to propagate to other participants at that particular internet exchange. Maybe there are other deployments in the world. I don't know, but this is what it is today next slide. Please here are some examples from the world.

D

It appears that most people seem to be using these attributes to announce to have their routers propagate through the route server to all participants, except a few, for instance, maybe except your customers, but I didn't dive deep enough into this to make a clear assessment what it is they actually do, but it proves it's used and.

D

Next, like thank you originally I, took the eb, + f specifications from the RFC 28, something something about our psl and that particular RC. It states that you can copy paste the e BNF into a program like bison bison and then generate itself.

D

So what I did is I just copy pasted some of the stuff added the attributes, as I thought, would be appropriate and submitted that in the draft next slide. Please.

D

Unfortunately, I learned that that particular bison codes from 1999 doesn't really work at all and it is in no way or shape to actually be used in a production environment, because it doesn't really do a lot next.

D

So my plan is to remove the the ed in F specification as a currently stands in the draft. I want to replace it with some Python code that also uses a dnf style syntax. So people can actually use the item program to either validate whether an entry is appropriate or not, and the Python program could also output some data that can be used in the actual routing policy on a route server, so I hope that will help people that want to implement this particular attribute at their route.

D

Server or their network to you know, help them along with something that is actually tangible and not the 90 90 97 x like this.

D

This is what I have today. So if you have any questions, please let me know.

G

So you'll update a provide, an update of the draft. Then, if there's any discussion on the lists go from there and if it's all right then we'll go last call after you have updated the draft and presented.

D

It to the list sounds good to me. Thank you for your time.

H

So good evening, this is it and give update on the route Lakes definition document. It has already gone through two or three revisions, so the difference is compared to the previous version.

H

Since march, since the dallas meeting, there have been two more they've been to more reports about major significant route leaks, one involving hathaway airtel and another one involving telekom, malaysia, a level three. So these two both fall into the tight one category. If you are trying to figure out what type 1 type 2 are, you can quickly go to the last slide and I have listed them just in case. You need to be refreshed on that.

H

But hopefully you know that know that different types type one is basically where the update makes a u-turn like a yeah. It's it should not be propagating from a customer to a provider, but but it does customer leaks it to a provider. So both of these incidents in March and again in june, fall into the type one category in and then we added the again the example of telekom, malaysia, wire level 3 in type 2 as well.

H

I type to basically says that it's not the whole full prefix, but it's a sub prefix, a more specific that is making the you turn or it's being leaked from a customer to a provider, and what happened in this case is that not only did telekom, malaysia leak, 179,000, prefixes to towards level 3 and level 3 propagated them accepted and propagated them did that 179,000 number also happened to include 10,000, which were supreme x's and those were meant to david, not meant when the aggregates of those were already announced.

H

But in this case, while the leak happened, they also got d aggregated and the supply fix kind of made, a u-turn from from Malaysia telecom towards a level three and into the larger internet. So those are additional. Examples of incidents have been added and Brian Dixon is included as an author now, and we have added several new references that are quite relevant makes life. So this is just to give you a little bit more view of robotic what actually happened in the case of hathway Airtel route leaks involving google prefixes.

H

So this is not the complete view, but it's a partial view that makes it clear how this route leak was was a propagating so Google piercer with hathaway in India, and they they share. Some 3336 prefixes with hathway had to lick them to airtel.

H

Normally, a orange, a lot of Google users in France Europe would go through orange. They take I would go through orange to level three to google. That would be the normal data path, but in this case I had to a leak, those prefixes to airtel. It will didn't, detect it and propagated it on and because of the preference of you see all those links labeled as c2 PP, 2, b, etc.

H

The any prefixes that come from a customer are preferred over prefixes learned from appear or a provider, and in this case it was set up to a pod for the hacker, a leak to propagate through Airtel to orange sa and level 3. Because of that policy. Prep little customer policy so as a desire to be I mean the millions of users in your open, also in Singapore and other places in Asia.

H

They they there traffic, ended up at Hathaway and and got dropped on the floor there because of lack of capacity, as so that's a I mean. So this is a maybe be talking about these examples and how to tackle this in the other draft. The solution draft next light, so here I thought talk a little bit about good trying to differentiate between accidental versus intentional, and this discussion came up on the eye. Dr list, so most loudly assume, I think, are accidental, maybe ninety nine percent of them.

H

You can pick a number, but certainly a large vast majority and a one-percent. Maybe a small correction, like one person may be intentional or malicious, and some examples of the malicious, malicious ones would be intentional. Leak of a more specific prefix, as in the capella philosophic mo so more specific prefixes may be intentionally leaked with the asf are intact, so that there is Pat poisoning and exactly like in couple apples of the update, propagates everywhere and wings over thus less specific.

H

But so that would happen accidentally like we saw in the previous slide with Malaysia telecom level three, but it could also be intentional or malicious. The next example is sorry, go back please. So the attacker keeps the legitimate originals, but removes all other proceeding a SS in the a s path, so shortens the earth s part, and by doing this, his announcement will have a shorter route to the destination.

H

Also, here the attacker is trying to deceive the origin. Validation, assuming that rpki and origin validation are being used. This by doing is the attacker is also trying to deceive origin validation, because the legitimate is is still showing as the origin. So these are examples and the third one would be a question is like: could there be? Is there any attack vector suppose we do come up with some solution, which involves putting some new bits to try to detect route leaks and can those in those new bits?

H

If they are unsecured, can they be, and they can they be used as a new attack vector x, hackett by changing those bits, so they are altering those bits. So by looking at these, one might think might think of differentiating between intentional unintentional.

B

H

Vs. intentional or malicious next like this, so just once you start thinking and we are taking a little bit into the solution here. But yet that's not the intent in growth growth.

H

The work here is mainly to define the doubt leaks, which we have done in there. They are documented in the definition draft, but forgive me here for going a little bit into the solution, but just this one slide, so the solution steps. If you are concerned about the X in accidental versus intentional, what would happen, as you start, considering the solution so initially current DGP? It is vulnerable to the accidental, which may be ninety nine percent, or you can pick your number and it's also vulnerable to the malicious, the one percent route leaks.

H

So if we start working towards a solution, for example, put some route lake protection bits as in the other draft the solution draft in IDR. If you start incorporating a solution, then new attack vectors can begin to happen. So in this case, if you have those are out lake protection bits and you have the rpi origin validation, what happens is that be?

H

It is able to detect the the accidental ninety nine percent, but the, but it's not yet I mean it it we don't yet have the capability or the attacker can can do something to bypass the detection, and that would be a malicious attack that would still be in the one percent ballpark us and then, if those bits are so so it does provide you benefit it.

H

It solves the accidental at out leaks a problem, but it still leaves the malicious unresolved or unsolved, and when we go to secure those Bitzer, then then the ninety-nine percent accidental, as well as the one percent malicious they are all detected and mitigated by the solution next five weeks. So I'll have a presentation about the solution on friday in cider. I already did that in an interim meeting with IDR three weeks ago.

H

So since then there has been substantial discussion about the solution stopped on the idea, at least, and it's been adopted by the chairs as a working group draft. Now next like this. So if you are interested in more details about the solution proposal, please come to the sided meeting on friday and so finally, coming back to the definition draft. So all the comments we comment receive them quite a few comments from many people and they have been all incorporated and just have been devised already, two or three times I.

H

So possibly it's a good time to request. Working group last fall on the definition draft and we could possibly include a section to discuss accidental worst versus malicious route leaks, which would be a minor change, a small section, a paragraph. However, that starts to get into a solution space. So we would be a little bit wary about it.

H

I

Questions any questions.

I

G

I understand you think it's ready for last call at this point. I.

H

Yeah, we think so. Ok.

G

Then I think we can start last call and see how many comments that triggers and see what the next step is from there. Sometimes it goes smoothly, and sometimes it generates a lot more back.

I

Thank you. Thank you.

G

But that's it so there's nothing else. We will see you in Yokohama.

G

If anybody hasn't sent a blue sheets, it's a good idea, because the number of people at the beginning was smaller than the number of people at the end.