evyncke / v6ops-james Goto Github PK

On 7/11/22 23:25, Xipengxiao wrote:

Hi Justin, Eric,

I read the draft and I think it’s a good one. A few questions and comments:

1. In the draft you stated twice that “assuming that the 2% of dropped
   packets are within the measurement error”: Is it really measurement
   error, or is it a polite way to say that 2% packet loss is common
   for IPv6? (BTW, this would be somewhat consistent with APNIC’s IPv6
   packet loss rate (PLR) measurement). To find this out, maybe you
   can do some measurement with IPv4 (or IPv6 without EH) to see if

It'd indeed be a polite way to say what you mentioned. We can assume so because (a) we already tried with IPv6-without-EH (same thing was observed); and (b) these measurements are based on cooperative nodes only for now, which is to say that we were able to capture traffic received on destinations and make sure packets were received or lost (without only relying on the sender and ICMP responses received).

packet loss rate can go to 0%).  The reason for my question is, if
2% packet loss rate is common for IPv6, it’s worth digging why and
trying to improve.

+1, it'd probably be worth it.

[...]

4. Is it possible to use the test bed to measure the performance (i.e.
   packet loss rate PLR and latency) difference between IPv6 and IPv4? I think that this may reveal some sub-optimal operations issues for
   IPv6 so that we can improve.

Yes, we could consider doing that.

On 7/26/22 17:26, Fernando Gont wrote:

6. Section 6 says:

[I-D.draft-vyncke-opsec-probe-attribution] to allow external parties
to identify the source of the probes if required.

My approach here is to configure the reverse DNS mapping for the probing (source address), and host a web page on that domain/site to eplain what you are doing. e.g., people might not be aware about your suggested encoding. Also, when e.g. SYN-scanning, you might not even be able to encode the info you want to encode.

Your approach, which of course is the usual one, is also a possibility referenced in [I-D.draft-vyncke-opsec-probe-attribution]. What we proposed in that draft is to provide other possibilities too (in-band alternatives, actually), depending on the situation (encapsulation protocol, L3 or L4, etc).

On 7/11/22 23:25, Xipengxiao wrote:

3. In Section 4.4 for Fragmentation Header, should the success rate for
   “Atomic” be 0% for both UDP & TCP? See the picture below. Does
   this mean that some operators are not filtering out atomic
   fragmentation properly?

Well, yes and no. I think we need to update this section to be clearer. In fact, the reason for having only half of atomic fragments going through is that the other half is filtered by some ASs, i.e., removed from packets, while packets are reaching their destinations no matter what. Note that we did not observe such filtering for non-atomic fragments. So, having 0% going through would not be correct IMO, as fragments headers should be handled by destinations, not by routers. On the other hand, who knows why there is a difference between how atomic and non-atomic fragments are treated by routers. We'll investigate as well.

On 28/07/2022, 03:18, "Laurent Schumacher (UNamur)" [email protected] wrote:

Bonjour  Éric,

draft-vyncke-v6ops-james ne gagnerait-il pas à faire référence au RFC 
9098 (Operational Implications of IPv6 Packets with Extension Headers) ?

A+,
LS

On 7/26/22 17:26, Fernando Gont wrote:

Hi Eric and co-authors!

First of all, thanks for writing this I-D, sharing the code, etc!

**** Some technical comments: ****

1. As noted during the meeting, in line with RFC9098, the drop rate is a function of the total IPv6 header chain. That's why, for a given length of an EH, the drop rate is different for UDP than for TCP (8-bytes vs 20-byte header).

Got it. We'll check that out, it could indeed very well be the reason why. But, talking about it, I think I remember some cases where we did not see such difference, so we'll dig deeper anyway.

On 7/26/22 17:26, Fernando Gont wrote:

2. Regarding the results for Routing Headers, I think they associated measurements should be performed for different RH sizes -- since my expectation is that the same principle applies to Routing Headers: the longer the IPv6 header chain, the higher the drop rate. SO my expectation is that once the RH is 64-bytes long or so, these headers also become unusuable on the public Internet.

Thx for the suggestion, we'll check that too.

[...]

5. Section 5 says:

• only routing headers types 0, 1 and 4 experiment problems over the
  Internet, other types have no problems;

As per the above, this really needs to be measured for different RH sizes. My expectation is that you'll get similar results as for e.g. Destination Options header.

Will do.

On 7/11/22 23:25, Xipengxiao wrote:

Also 2 minor editorial things:

5. Section 3.1, “Beijink ?”

Good catch, thanks.

6. Section 3.2.1, “… using link-local per [RFC7704])”: RFC7704 is “An
   IETF with Much Diversity and Professional Conduct”. The word
   “link-local” did not appear a single time in it. Not sure if you
   really want to reference 7704.

We'll correct this.