Getting the message out about Aeron’s messaging

“In Celtic mythology, Aeron is the god of battle or slaughter. It can be given as a name to a girl or a boy and means ‘battle ending’. It’s the name of this project … and the name of a chair.”

This is the GitHub entry for Aeron, an open-source suite of low-latency messaging technologies developed in 2014 by Martin Thompson and Todd Montgomery of Real Logic, the low-latency trading specialist acquired by Adaptive Financial Consulting last year.

According to Adaptive’s chief executive, Matt Barrett, getting the word out about Aeron, while assuaging users that Thompson would remain the genius behind the product, has been a key focus since the acquisition. Uptake of the technology suite had been slow—Real Logic had never marketed or promoted the tool on its own—but Adaptive is looking to change that, capitalizing on a moment for open-source technologies in financial services in which what was once taboo is now darling tech.

In February 2022, both Real Logic and Aeron were bought by the trading systems designer, which had already been using Aeron to underpin the architecture behind its Hydra Platform, a technical accelerator that powers single-dealer platforms, matching engines, and execution venues for exchanges, banks, and investment managers. Since becoming part of Adaptive, Aeron has seen the launch of subscription-based Aeron Premium for business and enterprise use and has partnered with Google Cloud, all of which have garnered market adoption in both its proprietary and open-source forms, Barrett says, though how much, exactly, is hard to tell.

You cannot exist in that world if you’ve been doing the techniques from 10-plus years ago, where flow and congestion control were not things that you had in your product using UDP.

Martin Thompson

Open-source tools, by nature, are open and free to use and download for anyone with an internet connection, but this makes pinning down exact usage metrics hard. Aeron now has 823 forks on GitHub, 63 more than this time last year. On GitHub, forks function as a tracker noting how many times users have copied a specific code. High numbers of forks on individual products are strong indicators of user popularity.

Among those in the know, though, Aeron and Real Logic have a strong reputation.

“Those were the guys that started [low-latency messaging service] SBE,” says Marcio Moreno, head of engineering at Proof Trading, a cloud-based trading platform. Developed by Real Logic in 2013—one year before Montgomery and Thompson built Aeron—SBE, or simple binary encoding, allows messages to be sent in computer-readable formats, enabling lower-latency messaging.

Origin Story

Co-founders Thompson and Montgomery were well-acquainted with the low-latency messaging space when, in 2014, a large US exchange tapped them to build an open-source messaging system. Aeron was the result. Around this time, “right about 10 or 15 years ago, we were moving to much higher capacity networks,” Thompson tells WatersTechnology.

Speaking at the Strange Loop conference in 2014, Thompson asked the crowd, “Does TCP not meet your required latency consistently? Is UDP not reliable enough? Do you need to multicast? What about flow control, congestion control…?”

By using the user datagram protocol (UDP) rather than the transmission control protocol (TCP), Montgomery and Thompson had engineered a new approach to messaging for capital markets players who required higher throughput (the amount of data successfully communicated) and fault tolerance (the ability of a system to continue working despite an operating malfunction) at lower latencies. Each of these questions has become more pertinent as exchanges migrate to the cloud and ethernet capacity increases.

While Proof Trading opted in 2021 to develop their own messaging protocol for their algorithmic trading tool, they noted in a blog post that the technology was based on Aeron and Real Logic’s innovations in the space. In computer networking, messaging is the backbone of any system, explains Marios Kogias, assistant professor within Imperial College London’s computing department. The first layer is ethernet, where changes in bandwidth have created room for Aeron to distinguish itself from competitors like RabbitMQ and Apache Kafka, which run primarily on TCP.

UDP and TCP are the two primary messaging protocols, though Kogias notes that academics and engineers often develop their own protocols. While TCP offers accuracy and determinism, meaning that messages arrive in the order they were sent, the protocol requires significant overhead bandwidth, explains Kogias. UDP, on the other hand, is faster and looser.

While multicast had historically been the method of choice for transporting large amounts of data in the pre-cloud days, public cloud providers, for the most part, do not effectively support multicast. With the move to the cloud, “once you get over a certain threshold, you just have to do things differently,” says Thompson. “And the fundamental differences are flow control and congestion control.”

As higher ethernet becomes available in the cloud, especially on high-end servers, messaging systems must handle large amounts of data at once. Flow and congestion control, which were typically features of TCP rather than UDP messaging, allow the technology to prioritize and handle those large streams.

“If you have flow and congestion control, you can exist in that world,” Thompson says. “You cannot exist in that world if you’ve been doing the techniques from 10-plus years ago, where flow and congestion control were not things that you had in your product using UDP.”

They’re going to face the same problems on-prem as people are in the cloud, so why not just face up to it?

Martin Thompson

Aeron has capitalized on an innovative use of UDP messaging, which abandons the slower but more precise TCP in favor of lower latency and higher throughput. It does this through Aeron Cluster, an implementation of computer clustering technology built on the Raft consensus algorithm. That allows the same set of messages to be received by each of the computers in the cluster consistently, reliably, and in the order they were sent in—features that are normally associated with TCP messaging and which mimic the benefits of multicast, according to Barrett.

Speaking to WatersTechnology in August 2022, Barrett called Cluster Aeron’s “most innovative solution”, because it allows Aeron to be deterministic, meaning that the input of messages exactly matches their output. For capital markets, where the order of messages—such as those going to an exchange—is crucial, Aeron can be a catch-all, says Barrett, by powering a variety of use-cases, such as an exchange’s matching engine, smart order routers, requests-for-quote, and central limit order books.

Cluster also gives Aeron fault tolerance by replicating messages, which allows Aeron to avoid single-points-of-failure, or the network nodes that would break the whole system if one of them failed.

Identifying Users

As noted before, while the nature of open-source technology makes it difficult to measure how many people are using the technology, Barrett says that anecdotally, he continues to hear of more use-cases, and he estimates the tech suite has about 1,000 users that use it in different capacities.

“We hear someone mentioning Aeron at a conference. Obliquely, someone mentions it at a party. I had someone tell me the other day that a very large investment bank had built a substantial amount of their technology on top of it, and talked to me about that as if I already knew,” Barrett says. “You find out in peculiar ways. It’s not a set of conversations I thought we’d be having as frequently as we are.”

With the launch of Aeron Premium, Adaptive created a number of features for users willing to pay for a subscription, including message encryption, a feature that at least one commentator noted was missing from Aeron in a 2020 blog post from Scott Logic.

Earlier this year, Aeron community meet-ups in New York, London, and Chicago drew crowds of around 120 people from almost 65 firms. In May and June of this year, Aeron also made the Aeron Cookbook—or the “recipe” for the software—available to download. Since then, about 120 people have downloaded it, and a newsletter created around the same time has seen about 100 subscribers opt-in to receive it.

As part of the marketing push, a number of firms have also allowed Adaptive to list them as public users of the technology, including CME Group, Cboe Digital, SIX, and CMC Markets.

As for the acquisition itself, Barrett and Martin agree it has gone better than they initially expected, even as the two companies faced some challenges integrating their cultures. “I’ve been through numerous acquisitions in my past lives and there are always bumps in the road. That’s just the very nature of it,” says Thompson, who notes that Aeron’s core team is spread across San Francisco, Auckland, Zurich, and London—a setup itself resembling a system of distributed computing.

But, as Thompson notes, “we’ve built [Aeron] to be fault tolerant.”

Out Among the Crowd

There are a number of messaging protocols and providers available for different use-cases. For example, the Financial Information eXchange (FIX) protocol, written in 1992, has underpinned the dissemination of price and trade information among banks and broker-dealers since its inception. More modern banking applications, such as analytics and trade processing, might use Kafka, an open-source messaging system designed by the Apache Software Foundation, but it’s important to note it wasn’t created with the financial industry in mind, Thompson says.

A messaging system that’s been created with the financial sector in mind should be crucial, he says, particularly as exchange outages become more common. During an outage, the systems exchanging information become backed up with instructions. When they come back online, it’s like a dam breaking.

“The simple example would be you get a huge burst of traffic released when the market opens; you will get so many orders, and yet, the latency is a bit higher. But when you’ve had thousands—or hundreds of thousands—of those orders all stacked up, how long does it take you to get through that?” Thompson says.

According to Thompson, the same goes for the on-prem world, which Thompson says is at an inflection point.

“They’re going to face the same problems on-prem as people are in the cloud, so why not just face up to it?” Thompson asks. “If people are going to have to move and adjust, you’re going to see people start moving to the cloud.”