The cloud isn’t just for storing high volumes of data, or for performing resource-intensive processes without requiring costly on-site infrastructures. Exchange operators are considering whether they can move all or part of their exchange operations to run in the cloud.

The idea makes sense: Increasingly, exchanges are not limited to investors from their local market, but attract a global clientele, so they must have connectivity around the globe. They also process huge volumes of orders, requiring massive compute power. Cloud can support worldwide access, and eliminate heavy infrastructure needs.

But there’s a problem: Cloud doesn’t support multicast data delivery. And multicast is the accepted method for distributing data from one source to multiple recipients at the same time.

Exchanges have begun moving storage—and in some cases, delivery—of data and services to the cloud. But they still run their core infrastructure—their matching engines and broadcast feeds—on their own server farms, on their own premises. Matching engines are complex, and, if shifted to the cloud, need to be able to provide the same degree of low latency as they can achieve on proprietary infrastructures. Many exchanges operate hosting and co-location businesses, and have to ensure that the distance via cable from the exchange’s matching engine to each client’s servers is the same, so that no client has an unfair advantage in terms of proximity.

Nobody is there yet, but I think that is the Holy Grail. And when somebody gets there, that is the capability that will unlock a huge amount of new cloud business in the capital markets.

Lee Bressler

To ensure each data message is sent to each client at the same time, exchanges broadcast data using an IP multicast, which sends the same message to all recipients at the same time, rather than broadcasting messages in a serial unicast—one after the other to each recipient separately—which, while more reliable, means that the final recipient of the data would be at a massive disadvantage.

Thus, cloud operators are scrambling to take advantage of a new form of potential business by offering true multicast capabilities in the cloud, says Lee Bressler, former capital markets lead at Microsoft, who now runs his own consultancy.

“Nobody is there yet, but I think that is the Holy Grail. And when somebody gets there, that is the capability that will unlock a huge amount of new cloud business in the capital markets,” Bressler says. “They are all racing to build this capability. The big cloud firms have top engineers working on it. Amazon was the first of the big three to make a product roadmap announcement, but none of them are quite there yet. This is still very much an arms race.”

Nasdaq is one exchange that has begun making some market data available in the cloud, via a partnership with Amazon Web Services (AWS), though this does not yet include low-latency datafeeds that would be a requirement for trading.

“Multicast is typically not available in cloud environments today,” says Nikolai Larbalestier (pictured, right), senior vice president of enterprise architecture and performance engineering at Nasdaq. “That is one key piece of the puzzle. It is not the only piece, but it is probably the most relevant piece. Getting that ability to disseminate from a single sender to multiple receivers—using the same stream of data—is a prerequisite.”

In December 2019, AWS launched a multicast service in the cloud for its AWS Transit Gateway, a cloud router that allows the ushering of data between different AWS environments.

John Kain, who leads worldwide business and markets development for capital markets at AWS, says the new service gives exchanges the ability to expand their reach without making long-term commitments to infrastructure, leveraging existing technology stacks. He says a number of exchanges—he declines to name any—are piloting and evaluating the new service.

AWS’s functionality allows customers to begin to use multicast, though it does not offer the same level of service as on-premise multicast. “It’s not that nanosecond-responsive multicast that you would find in hardware in an exchange datacenter, but it is performant, and certainly if you are not looking at ultra-low latency trading, it is a viable option,” Kain says. 

The key challenge here is that while it works, it can’t be used by those who need to receive data and send trade orders to an exchange with minimal latency, which accounts for most institutional traders. And, as one industry source puts it, the real money is in ultra-low latency applications.

“I think we are still quite a bit away from [supporting] something like a regulated equity exchange that has performance characteristics that are measured in nanoseconds. I do not actually think, for the foreseeable future, that is a good fit for the multicast service that we offer today,” Kain adds.

‘Not in My Lifetime’

Marketplaces that are not particularly latency sensitive, or have a trading model that provides a “last look” on trades—such as broker-dealer currency trading platforms and markets trading digital assets—are natural candidates to move to the cloud, Kain says.

For example, it can be easier for newer exchanges—or those with a different operating model from traditional marketplaces—to adopt cloud as they are not bogged down by legacy systems.

To this point, many crypto exchanges run on the cloud. Hirander Misra, CEO of GMEX Group, says considerations of a crypto exchange are very different because they typically cater to retail investors, even though some wholesale investors have started to take an interest in those markets. Another factor is that these crypto markets have been largely unregulated so far, and thus, some of the same considerations that exist for regulated markets do not apply, he adds.

“I am not saying that the regulators would not allow a completely cloud-based exchange—as long as you can demonstrate security, accurate archiving, [and] timestamping, then from a regulatory perspective, it is totally permissible,” Misra says.

Instead, he says the barriers to exchanges adopting cloud relate more to commercial considerations, and in some cases technical factors, since some of the exchanges are running trading platforms that were never meant to connect to the cloud.

For equities, there is no way that is going to happen in my professional lifetime, just because of the nature of the market.

Exchange executive

As such, traditional exchange environments remain a challenge for the cloud. One industry executive, speaking on condition of anonymity, says he does not feel the cloud model works for highly liquid equity markets.

“For equities, there is no way that is going to happen in my professional lifetime, just because of the nature of the market,” the executive says, referring to the potential for exchanges to migrate entirely to cloud environments.

However, others say cloud is possible for exchanges, with a little work. “To put equity trading markets like LSEG’s in the cloud, you would have to work with a cloud service provider (CSP) to tune the platform to the same latency profile that that we are seeing on-premise,” says Ann Neidenbach, CIO for capital markets and global head of LSEG Technology at the London Stock Exchange Group.

LSEG Technology’s matching engine, which supports the London Stock Exchange, Borsa Italiana, and Turquoise trading platforms, has also been licensed by Hong Kong-based digital asset exchange operator Atom to support its cloud-based AAX (Atom Asset Exchange) cryptocurrency trading platform.

“From a software perspective, we are already running LSEG Technology’s matching engine 24/7 in the cloud for Atom. In order to run equity markets such as LSEG’s in the cloud, you would have to do a lot of tuning for latency. Then you would also have to figure out how to distribute data, and how this would impact low latency hosting clients,” says Neidenbach (pictured, left).

As more of the asset classes that don’t require extreme low latency move to the cloud, Neidenbach says she expects to see exchanges innovating and creating different ways to distribute data.

“I think as you start to see businesses really understand the power of the cloud, they are going to start innovating more and they are going to start developing new markets and new exchanges on the cloud. I do think there’s an opportunity to introduce putting some markets in the cloud much sooner than five to 10 years out,” she says.

AWS’s Kain also believes exchanges will start moving some of their core operations—regulated, large-volume trading—to the cloud in “the next couple of years,” and over the longer term, could fully migrate to the cloud, depending on their trading model and the instruments traded.

“We already have a number of regulated exchanges for different asset classes running within AWS. So that’s already there. And then, as the performance characteristics improve … you will continue to see that keep progressing, even in the more latency-sensitive markets,” he says.

A New ‘Firehose’

Web services data provider Xignite has been distributing data via the cloud for 12 years, but not in the same sense as the “firehose” feeds that deliver every quote and trade, says CEO Stephane Dubois (pictured, below).

On the internet, for example, data is exchanged when two parties connect online. One requests data, and the other sends it—much like a person-to-person phone call. But this is not an efficient approach for communicating data to thousands of exchange members. Multicast takes more of a fan-out approach to distribute data—more like a conference call, or, as Dubois says, a radio broadcast.

“Unless the cloud providers really solve that problem … it really is going to put a damper on migrating the big matching engines to the cloud,” Dubois says.

Multicast is difficult to move to the cloud because it requires sophisticated networking equipment that can achieve incredibly fast data transmission speeds. It uses so much bandwidth, says Microsoft’s Bressler, that it requires massive pipes to support it—or, advanced ways of compressing the data and then rapidly decompressing it at its destination.

Multicast solutions currently available over the cloud have overlays that allow users to emulate or synthesize multicast over a point-to-point network. It can work for some multicast use cases, but that doesn’t solve the fairness problem—ensuring that no recipient gets a speed advantage from being the first one to receive the data, adds Nasdaq’s Larbalestier.

“It looks like multicast, but it does not actually solve that part of the problem. For that, you really need a native multicast implementation, paired with physical location controls to ensure a predictable and deterministic path from the matching engine to market participants. Today that is not generally available in any public cloud,” he says.

Typically, exchanges have addressed this equality issue using equidistant cabling, so that everyone within a co-location facility receives data at the same time. However, where a cloud workload is placed is not exactly controllable. It may be next to the matching engine, or across the datacenter, which will have an impact on the time at which a particular client server will receive information.

Larbalestier says Nasdaq is actively working with multiple cloud vendors on the problem of moving core matching engines to the cloud, but says multicast is both the sticking point and the way forward, for now.

“The only thing that provides those attributes today, in terms of the simultaneous delivery potential, is multicast,” Larbalestier says. “That is not to say there could not be a new creative solution to provide that, but I think that right now, it seems to me that multicast is the answer.”

Such “new creative solutions” that could provide alternative distribution options that don’t necessarily require multicast include Google Pub/Sub, a real-time messaging service that allows the sending and receiving of messages between applications, and similar services, such as  AWS Kinesis, Azure Service Bus, and Apache Kafka.

For example, CME Group leverages Pub/Sub to help provide access to its real-time market data via Google’s cloud, while Nasdaq’s Cloud Data Service uses Kafka to distribute market data for customers in a “fan-out” topology.

‘Ripe for Innovation’

Still others believe that a whole new mouse trap will be required in getting equity exchanges to move fully into the cloud.

“Broadcast and multicast are both legacy networking technologies used for several decades. Communication protocols are ripe for innovation, and cloud offers an opportunity for the industry to reimagine them,” says Adrian Pool, head of financial services for the UK and Ireland at Google Cloud.

Poole cites Refinitiv, which recently made its extensive Tick History archive of pricing and trading data available to clients via Google Cloud, and says that beyond data vendors, many exchanges have already started moving their core activities to the cloud—though he says most “don’t typically publicize these efforts.”

But while multicast is the traditional standard for market data distribution, cloud offers an opportunity to update that “traditional” operating model, and introduce alternative data protocols.

There are other ways that you can distribute data; multicast is just one methodology.

Ann Neidenbach

LSEG’s Neidenbach says that while exchanges have traditionally distributed their market data via multicast, technology has evolved and exchanges should arguably be innovating new ways of distributing data.

“There are other ways that you can distribute data; multicast is just one methodology,” she says. “There is unicast, multicast, and point-to-point TCP, for example. Multicast is the traditional network way of distributing data. But if you are a consumer, and you are in the cloud as well, you can listen to the data through other protocols, such as APIs like WebSockets.”

There also exists a much broader set of market participants, many of whom would not require necessarily the same level of latency when accessing the market. In the cloud, there is a good argument for bifurcating users to meet these different requirements.

“I think we could certainly see things change there in terms of the offerings and meeting customers more based on their needs than on a fixed model,” says Nasdaq’s Larbalestier.

But until that happens, or until a new, cloud-friendly protocol emerges for low-latency streaming data, or until exchanges revamp their legacy architectures to resemble newer marketplaces, multicast may remain a barrier to full cloud adoption by exchanges.