RSSAC Meeting, San Francisco
RSSAC meeting 15
- Report on anycast
- Delegation response size
Joint WIDE/CAIDA work in progress
Changing architectures will change change measurements
three sites - RTT measuresments from UCSD/Aukland/UC-boulder - lack of coorelation btwn sites - sometimes events on two sites, never three.
timing is all UTC
need more than three sites, may want to do three more. no active measurements, how will this answer the Question, serving poorly served areas of the net? - need more diverse number of nodes.
AK - did you measure RTT values? Plots are of 5min. medians. Should not be much different than means.
AK - its all passive so is there anyway to be other than reactive?
FN - can synthetic queries simulate loading for passive servers? -- CAIDA/WIDE are working on active/passive measurement coordinating.
PV - use seti@home techniques? -- talk to Chris
LT - any analysis of query types? -- No.
IDN -- BM
open interop testing 28-30 May. working with the community on testing. This will impact response size, even more than existing or contemplated techniques. there is a potential impact of ambigious interpretation of the label seperator. this behaviour could generate extranious traffic.look at existing traffic for things that have the 8th bit set.
RA - . same trash w/ IDNencoding or more traffic? –
JM : more traffic
.JP - using IDN next level lables. Active push
Delegation response size -- PV
Kato Internet Draft - stuck w/ 512 bytes per RFC 1035
reponses include a copy of the query. the larger the query,
the smaller the NS record set.
all , none , some
(only for delegations)
minimal useful glue : two
best case, no truncation.
TLD operators can use the common name hack to increase
the number of servers.
adding AAAA will impact the response.
presented to DNSOPS - there are some layer8 issues.
what will resolvers do with mixed A/AAAA replies - Katos draft.
generally push A first, then AAAA.
GM – are there boundary errors? - PV - No.
Anycast -- PV
anycast in the v4 context. no BGP tutorial needed.
7 of 13 are doing anycast. M has been doing this longer than
Not particularly dangerous - lots of experience.
C - cogent - in their datacenters @ edges
F - ISC - international exchanges –
21 sites in pipeline
G - DISA - survivable service for parent org
I – A’nomica - plans for about 10
J - VSIGN - colo w/ gtld nodes
K - RIPE - AMSIX/LINX - add more later in Europe
M - WIDE - More in East Asia
LoadBalancing - nearly all services do this now.
What are the differences in common administrative control? - LT
PV - owned vs unowned anycast. F's AS vs AS112
LT - what failure modes have been seen and may be likely?
PV – scaling issues. who pulls what from where?
JB - self diagnostics? pull myself when I'm out of sync.
PV - cronjob now - will instrument zebra...
BV. - now four SNS points.
PV - one transit instance, everyone else is peering, localized failure
IPv6 - AK
V6 proposal draft - we are stuck w/ 512 so only two AAAAs visible for most of the net. Use anycast services for these two AAAAs.
AAAA records will use IPv6 transport and the anycast use is of the
Will need to add to the root-servers.net zone. Move to all servers with AAAA service. Needs eDNS to work properly.
LT - Contrary to Ted Hardys draft? - Paul, no broad acceptance of this idea. Its not that critical.
RIO mtg - JM
JM will not attend - four operators will attend. Manning, Crain, Woolf, Liman. One of those will need to report to ICANN. Liman is asked to present. GAC meeting invitation to meet w/ root ops (not RSSAC). - LT - GAC does not distinguish between root ops and RSSAC. Cover training, meeting, security, idn, etc... for Monday
.local - AK
.Local is used widely although it is not a formal TLD. Query rates are going up. How can we get .local delegated to things like the AS112 servers. Perhaps this needs IETF action to give ICANN leverage in supporting this type of request.
Next mtg - sunday before IETF in Wien