[rssac-caucus] Best Practices for the Distribution of Anycast Instances of the Root Name Service WP Conclusion

[Wessels, Duane]

> On Feb 2, 2018, at 10:06 AM, kranjbar <kranjbar at ripe.net> wrote:
> 
> Dear RSSAC Caucus,
> 
> Please find attached outcome document of  ‘Best Practices for the Distribution of Anycast Instances of the Root Name Service’ work party. As it was mentioned in the caucus updates, the work party did not conclude with finishing all of it’s tasks but had many deep discussions about the questions raised in the work plan. This document summarises all of those discussions and conclusions and, at any point in time, if RSSAC or RSSAC Caucus feels one of the questions should be explored in more detail, we always have the option of booting up a new work party around that single issue and achieve more fine grained results.
> 
> Please let me know if you have any suggestions or questions.

Thanks Kaveh and WP members,

I read through the document and have the following comments:

Table 1: Rows in this table should be either Letters+Orgs or just Orgs.  I suggest removing "(A/J root)" from the Verisign row.

Section 3, third paragraph: I worry that this paragraph, which talks about DOS events, makes it sound like these are more serious or likely to occur than they really are.  Has there ever been a time when a significant proportion of servers were unable to return answers in a timely manner?  I think its important to state these as reasons why we utilize anycast, but we should also not give the false impression that widespread unavailability is likely to happen.

Section 4: I think the values of 30ms low latency and 100ms high latency are good.

Section 4: Somewhere in this section maybe it would be appropriate to state that the nature of traffic from recursive name servers can impact the perceived latency?  If a recursive mostly asks the root about existing TLDs then those are highly cachable, but a recursive (or user population) that sends a lot of queries for non-existing TLDs may have a worse overall user experience?

Section 4 "The RSSAC asks" bold text: I like this line of questioning, however I wonder if "maximum latency" should maybe be something else like median or mean, or 90th percentile instead?  I think maximum might actually be unbounded.  Also I wonder if it should just be left at "with the DNS root service" and drop "as opposed to a single root server" from the question.

Section 4, last para: Glad to see the line about how increased latency may be due to peering relationships between ISPs.  I think that is very important.

Section 6.1: both the first and third paragraphs seem to talk about forged routes for longer prefixes?

DW



-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://mm.icann.org/pipermail/rssac-caucus/attachments/20180205/6520643f/signature.asc>