The 123rd member meeting of LINX (formerly the London Internet Exchange) took place on the 20th and 21st November, 2024. This also marked the 30th anniversary of the creation of LINX.
For those that don’t know, Internet Exchange Points (IXPs) are locations where a number of Internet Service Providers (ISPs) and Content Delivery Networks (CDNs) come together to connect to each other, initially at least within a common datacenter, and often via a layer 2 network that places very few restrictions on those interconnections. This happens either via bilateral sessions (direct BGP between the participating networks across the layer 2 fabric), or via route servers (BGP route collectors/route reflectors). Many IXPs also offer dedicated layer 2 VLANs between members.
LINX started in Telehouse in London’s Docklands, and that is still where the Janet network’s largest connection to it is. There were five founder members — BT, Pipex, Demon, EUnet GB and of course Janet. It started off as a 10Mbps half-duplex Ethernet hub, not even a switch, which started to drop packets when traffic exceeded 2Mbps. The first BGP session was brought up between Pipex and Janet.
These days, LINX has two networks in London, each one covering sixteen different sites, further networks in Manchester, Edinburgh and Cardiff, and several abroad such as North Virginia, Jeddah and Nairobi. Peak traffic exceeds 10Tbit/s. Whilst sites within a city are interconnected, the different cities are not connected as LINX is not a transit provider. As you might imagine, these are not simple layer 2 switches any more, but virtual layer 2 networks implemented over EVPN or VPLS. LINX has about 900 members.
The Janet network is connected to the two networks in London and the network in Manchester, and usually we will initially connect to a peer over an IXP such as LINX, and if traffic between the peer and ourselves increases to the order of tens of gigabits per second, move it off to a dedicated connection (PNI or Private Network Interconnect). These days PNIs tends to be at 100Gbit/s or multiples of that.
When we peer with a network, the idea is that we only use that for traffic to/from that particular network and its immediate customers. For any network that we don’t have a direct relationship with we use “transit” from a couple of large well-connected providers. As its name suggests, this allows us to transit across them to other providers who may not be direct customers.
What happened at LINX 123?
Other than marking the anniversary, there was quite a bit of content at the meeting. For a start there is the ‘business’ of LINX — updates from the CEO (both outgoing and incoming), the CFO, the technical team, and governance, more on which later.
The meeting is sponsored, so the sponsors gave presentations: Digital Realty on global trends in AI and datacentres. Precision Optical Technologies on 100G ZR transceivers.
There were a couple of keynotes, one from Vint Cerf, and another from Philip Smith, who used to work for Pipex and racked up the first LINX hubs and switches. If you read thought Philip’s slides, linked above, you’ll note a number of mentions of research and education networks around the world.
There was an interesting talk from Akamai, one of the largest Content Delivery Networks, on their infrastructure in the UK, which handles a daily peak of 17-20Tbit/s.
Infrastructure sharing was the topic of two talks, from CNI and Asset Hub.
Given the cable breaks in the Baltic Sea this month, there was a timely presentation from Meta about how they investigated concurrent cable failures in the Red Sea in February of this year. In the theme of “never attribute to malice that which can be adequately explained by incompetence,” it should be noted that about 60% of submarine cable failures are caused by vessels (fishing, dredging, dragging anchors), 10% by failures of the systems (amplifiers, etc), 5% by ‘geohazards’ and natural occurrences. That of course still leaves 25% due to other causes, including hostile acts. After much study of ship tracking data and satellite images, the failures in the Red Sea were eventually discovered to be caused by an abandoned ship drifting across the cables and dragging its anchor through them.
I also don’t want to forget any of the other talks on Datacentres in Africa, post-quantum VPNs (post-quantum cryptography being a hot topic at the moment), inter-domain routing security (particularly legislation impacting on it) and an update from INEX, the IXP for the island of Ireland (for which I also spent some time as a non-executive director).
LINX Governance
In common with many Internet-related bodies, LINX is member-governed. This means that members get to vote in non-executive directors, and have a key role in the governance of the organisation. LINX has a Memorandum of Understanding (MoU) with its members. That document was created many years ago and has since been extensively updated, usually small sentences being changed here and there to, for example, allow online voting instead of paper ballots, or amend the number of directors. As a result of this piecemeal evolution, the current MoU has become a bit unwieldy and no longer covers some of the services that LINX offers.
To try and solve this, LINX have announced a consultation on a new Membership Agreement and service-specific terms. This is a ground-up rewrite and the new documents are intended to be much more readable and cover not only the services that LINX currently offers, but be able to cover new services more easily.
There was also a General Meeting (GM) at this LINX meeting, voting on remuneration for the directors and approving the proposed fees for next year.
These may not be the most enthralling topics, but earlier on I mentioned that LINX had ~900 members. Around 350 people were present in the room on the first day of the member meeting. Electronic voting for the GM was open for two full weeks ahead of the member meeting, simply by clicking a link in an email sent from an independent polling company (Civica Election Services).
The number of votes cast at the GM? 58.
Not 58%, fifty-eight votes.
Addendum: Over on LinkedIn, it was pointed out to me that 7% is not an abnormally low voter turnout for membership and mutual organisations, and that’s absolutely right. At the RIPE 89 GM mentioned in a previous post turnout was between 5 and 6%, especially without the excitement of a board election. However, there are risks with such a low level of engagement. The legitimacy of those votes could, in extreme circumstances, be called into question. There’s also the risk of bloc voting being easier to organise if fewer members are turning out, and potential capture of the organisation.