EIX Working Group, 3 November 2011
Co-Chairs: Andy Davidson, Fearghas Mackay
Scribe: Amanda Gowland
Chat: Chris Buckridge
A: Administrative Matters
- Chat monitor
- Microphone etiquette
- Agenda Bashing
EIX WG co-Chair Fearghas Mackay welcomed attendees and thanked the chat monitor and scribe. He asked if anyone would like to add items to the agenda. No one commented. Fearghas announced the IPv6 Privacy BoF at the end of the day.
B. The Local Peering Scene “All about Peering in Austria”
- Christian Panigl, VIX
The presentation is available at: http://ripe63.ripe.net/presentations/155-Peering-in-AT-RIPE63.pdf
Christian asked the audience if anyone was connected to the Grazer IX, one person put their hand up.
Bern Spies said that it’s still alive but doesn’t know a website. Christian said he found something in Salzburg but it was broken so he didn’t put it in the slides.
There were no other questions.
C. EIX Activity Updates – New IPv4 Policy and Switch Wishlist
- Andy Davidson, LONAP
The presentation is available at: http://ripe63.ripe.net/presentations/158-ad-update-eix.pptx
Andy shared the aim of the project and asked for volunteers to help authoring after running through a handful of themes and missing features for the switch wishlist.
Remco Mook said he liked the topic and offered to help.
Martin Pels, AMS-IX also volunteered to help on VPLS.
There were no further questions.
D. Monitoring Platforms for Internet Exchange Points
- Harald Michl, VIX
The presentation is available at: http://ripe63.ripe.net/presentations/148-ripe63-eix-vix_icinga.pdf
Martin Levy, Hurricane Electric, asked him to talk about the RD cache and the disc side.
Harald replied that it was ten to three and ten to four, it’s related to the active and passive checks. It doesn’t do anything until it gets the results. If you have a check that doesn’t work it takes five seconds to get the results, multiplied to 6k.
Martin asked if the RRD, cache part was just “icing on the cake” and where it came from.
Harld said he wasn’t sure.
There were no further questions.
E. Jumbo Frames in AMS-IX
- Maksym Tulyuk, AMS-IX
The presentation is available at: http://ripe63.ripe.net/presentations/129-Jumbo_Frames_RIPE63_Nov2011.pdf
Kurtis Lindquist, netnod, said they have supported jumbo frames on separate VLAN since 2002 and they let customers decide what they want. He said he didn’t understand the need for a second port because they can detect ports and connect to one VLAN or another or both. If they want to send large MTUs, they can and do.
Maksym said his presentation wasn’t about what was good or bad, but who could use jumbo frames.
Kurtis said they left out the most common protocol, BGP.
Maksym said it was a big network and a stable platform, so why increase size to put three months work into it.
Kurtis replied to share experience, it works, and BGP works well.
Patrick Gilmore wanted to know if they asked other exchanges to share experiences.
Maksym said that’s why they did the survey.
Patrick asked if anyone in the room was surprised that half the packets were small. He question Maksym’s conclusions about the small packet sizes, adding that while factually correct, the data can be interpreted very different. He suggested that before asking people to fill out another survey, he should talk to his peers in the IX community.
Maksym asked if he was interested in jumbo frames on AMS-IX
Patrick said Akamai does not send any because the majority of their end users are at 1,500 byte MTUs or lower but there are many other people to ask.
Martin Levy, Hurricane Electric, commented that Makysm’s opinion on what they do on layer-3 is totally uninteresting. He is a customer and wants layer-2 service, so that’s the first issue. He added that the studies he mentioned were very old and advised him to look at the latest speed test results which got much better TCP and jumbo frame usage. He suggested updating his stats. As far as the number issue, he said it was well known that 9,000 had no merit; it was just a nice number. Martin added that as a customer, he came to AMS-IX ages ago because they have customers that want jumbo frames, and to get to Akamai information is not their issue, that’s not what they came to them and asked them for.
Maksym said he knows they came to them and asked if he knew anyone else that wanted to peer with jumbo frames.
Martin said he did.
Maksym said they could discuss it at their meeting in two weeks.
Martin said that netnod’s experience is valuable; there is already information out there on how to do this. He said that as a customer, he expects more of a positive response from AMS-IX and they should go and talk to their peers to find out what’s really going on out there.
Maksym said they prefer platform stability, 37% said they shouldn’t support it.
Martin replied that he has a future as a politician and that that percentage wasn’t convincing when 60% said yes.
Gert Doering, speaking as a network operator, said there are benefits of large packets and that he’d like to see bigger MTUs in the core. But on the other hand, having providers figuring out how to get by MTUs working with two different VLANs means a lot of extra work, so he’s torn as to what the answer is.
Maksym said that as far as he knew, netnod was the only IX in Europe that provides jumbo frames so if people really want it they can follow netnod’s way and start supporting jumbo frames.
Jen Linkova, Google, said that another disadvantage is vendor bugs on the equipment.
David Freedman, Claranet, via Jabber, said that the first issue is that packing more info into large frames means more is lost when they’re dropped. Second issue is that if MTUs are mismatched, your peer does 9k and you don’t, you have to issue a “too big”. He added that, in response to Jen, if you can’t do PM2RD, you are screwed.
Andy Davidson said if peering LAN isn’t announced by BGP the “too big” will never get out and if they can’t get any more V4 for new LANs in this service region then that would never work as well, another reason to support 2011-05.
Nick Hilliard, INEX, mentioned an unsuccessful experiment introducing jumbo VLAN to INEX with only two incompatible participants. Added that they will provide jumbo frames on a private VLAN connection, tagged ports, can support on primary VLAN but strongly recommend against it.
K. APIX Update
- Gaurab Raj Upadhaya
The presentation is available at: http://ripe63.ripe.net/presentations/159-20111018_RIPE63-APIX-update-v00.pdf
There were no questions.
Thursday, 3 November, 14:00 — 15:30
EIX WG co-Chair Andy Davidson began the session and introduced a change to the agenda, Ondrej will present first, then Kurtis. Andy also announced an open mic session and the IPv6 Privacy BoF.
G. Extended Communities for Route-Servers and ASN32
- Ondrej Filip , NIX.CZ
The presentation is available at: http://ripe63.ripe.net/presentations/160-BIRD-20111103-OF-RIPE-EIX-RS.pdf
Maksym Tulyuk, AMS-IX, thanked Ondrej for introducing it to AMS-IX, asked for stats on users in extended communities.
Ondrej said it’s just two members.
Maksym asked when it was introduced.
Ondrej said three months ago.
Maksym asked if there were issues with vendors.
Ondrej said just the one mentioned with Cisco.
Bertrand Duvivier, Cisco Systems, commented that he defined the priority in Cisco and volunteered to help him with that. He added that there was a new draft about extending it to a wider community and asked Ondrej if he was interested in this.
Ondrej said if there was interest for wider communities they would look into it.
Bertrand said that the benefit of wider community is that it’s less restrictive.
Andy Davidson asked if anyone had tried extended communities with open BGP. No hands were raised.
F. How to resolve edge redundancy for peering
- Kurtis Lindqvist, netnod
The presentation is available at: http://ripe63.ripe.net/presentations/168-eix-peering-redundnacy.pdf
James Blessing, Limelight Networks, commented that different networks handle this in different ways, some only have one peering point, others 20 or 30. He added that the other issue is how long is an IX going to be down for, if it’s a couple hours, not the end of world.
Kurtis said that an outage still affects end user experience.
Will Hargrave, LONAP, asked if netnod’s position is affected because they require people to get two ports.
Kurtis said he didn’t know. They get feedback that netnod is more expensive, user feedback say they really want two ports.
Will asked if the problem was that they were asking existing members and not prospective members.
Kurtis said he got similar feedback from prospective users.
Maksym Tulyuk, AMS-IX, thanked Kurtis for promoting AMS-IX and commented that they only provide backup for 10g ports and despite that, they have customers that connect on two sites.
Dave Temkin, Netflix, asked if the question was still peering vs. transit for redundancy and whether it was viable for the peering volumes today.
Kurtis said he was more thinking peering vs. peering and what the planning for redundancy was. He wasn’t concerned so much with peering switching to transit.
David replied that the larger question is more about if people are going to other exchanges for redundancy.
Kurtis agreed that people do go to multiple exchanges but asked if the failure between those two exchanges truly redundancy, if there’s a 30 millisecond redundancy between Stockholm and London, is that really redundancy because it seems like significant latency for high traffic volumes.
Patrick Gilmore, Akamai, said that BGP convergence does hurt user experience. He said the question is if netnod fails and they move to AMS-IX, is that 30 milliseconds worse than if netnod fails and they move over to netnod B. The answer is yet, but at the beginning of this you still have a failure going between two sites in same building you’re still going to have user experience problems. Question is how much is it worth to not pull back to 30 milliseconds later.
He added that in the jumbo frames presentation, Kurtis had said that they let their users decide, yet somehow you guarantee that I pay higher price if I need a 10g port, I’m going to have buy two whether I like it or not.
Kurtis said he thought that question on the survey would have been a no-brainer, yet the feedback surprised him.
Patrick replied that netnod membership is unusual, no other major exchange that posts individual member ports to the world. He added that if he honestly thinks that members should decide, then have two switches and let them.
In response to Patrick, Martin Levy said that the analogy to jumbo frames is different; it’s like relating a budget airline to a high-quality one. You can choose to connect to netnod with two ports or a smaller competitor that only run one port.
John Souter, LINX, said that their members in the UK want diversity in data centres, they want choice and want to split traffic. John offered to share the data on that. He also added that they had data on what happens during major outages and offered to share it.
Mike Hughes, used to run IX, thanked Kurtis, said it was needed. Commented on seeing same MAC address on same switches in Stockholm, interesting how many MAC address appears on multiple exchanges in Europe. He added that it was indicative of how much some providers care about redundancy.
Patrik Falstrom, Cisco, said the discussion should be continued and asked what the problem to be solved was and what they were afraid of.
H. Update on ISC routing project
- Joao Damos, ISC
The website Joao presented is available at: www.opensourcerouting.org
There were no questions for Joao.
I. Proxy-Arp and why it must be disabled when connecting to an Internet-Exchange
- Wolfgang Tremmel DE-CIX
The presentation is available at: http://ripe63.ripe.net/presentations/150-e-wt-20111103-ProxyArpIncident-RIPE63-01.pptx
Benedict Stockeband, freelance IPv6 guy, commented that some people are using proxy neighbour discovery for mobile IPv6, and if it really takes off, it could cause problems with router configuration.
Wolfgang urged to leave it in code but make it default turned off.
J. Danger of proxy ARP in IX environment
- Maksym Tulyuk, AMS-IX
The presentation is available at: http://ripe63.ripe.net/presentations/130-Proxy_ARP_RIPE_Nov2011.pdf
Gert Doering, connected to DE-CIX, remarked that Proxy-ARP is useful in certain circumstances. Cisco made it on by default, stupid choice, so if anything is broken it’s because it’s on by default.
Alexander Isavnin, Media Alliance, asked why such problems in a short period of time happened in at least two exchanges.
Wolfgang said they were completely unrelated. DE-CIX’s was remotely triggered and that AMS-IX’s was remotely triggered by someone moving into the peering LAN.
Bertrand Duvivier, Cisco, said he heard the message and will take it back to Cisco, he said that any change in default will only come in the new release and this will take one or two years to take effect. Not making promises but will convey the message.
Maksym said in AMS-IX’s case it wasn’t the Cisco router but the Linux router, it was the bug was in the Cisco kernel.
Andy Davidson said he hoped it would be solved by default in future.
L. Euro-IX Update
- Bijal Sanghani, Euro-IX
The presentation is available at: http://ripe63.ripe.net/presentations/173-EIX-RIPE63-final.pdf
Christian Panigl, UniVie/ACOnet/VIX, asked if she knew how much the IPv6 traffic would grow over the next four years.
Bijal said they didn’t have those stats at the moment but that they would start collecting them.
M. IX Announcements
Andy asked for volunteers to give a brief update on their IXs.
Thomas Pangin, IX Leeds, announced two newly incorporated IXs.
Ben Hedges, LINX, said they migrated their primary LAN to Jupiter and they were launching a resale programme so people could connect to LINX.
Andreas Sturm, from DE-CIX, announced a vacancy for a Jr. Business Developer.
Izumi Okutani, JPNIC, gave an update on a 100g ethernet experiment with multiple vendors, it was successful and they are preparing for implementation. He also mentioned new route servers in Osaka.
Andy reminded attendees about the IPv6 Privacy BoF that evening, thanked all attendees and encouraged everyone to take part in the EIX mailing list.