Global financial markets today continue to operate mostly from physical data centers sized for peak capacity, resulting in a heavy infrastructure footprint that can be costly to scale. The requirement for low-latency applications, such as trading engines, which have complex and carefully designed physical networks, creates unique challenges for migrating high-performance workloads to cloud-based environments.

Singapore Exchange Limited (SGX), one of Asia’s leading multiasset market infrastructures, set out to explore the possibility of running a low-latency trading engine on the cloud. In collaboration with Aquis Technologies, the software licensing arm of exchange services group Aquis Exchange PLC (Aquis), SGX turned to Amazon Web Services (AWS) and AWS Transit Gateway, a managed service that connects virtual private clouds and on-premises networks through a central hub.

One of Europe’s top 10 stock exchanges, Aquis operates pan-European cash equities trading businesses and regulated primary markets in addition to developing and licensing exchange software to third parties. Together, SGX and Aquis executed a proof of concept (POC) demonstrating that a low-latency central limit order book trading engine can be successfully hosted entirely on the cloud.

The results of the POC exceeded performance expectations in terms of deterministic latency and opened up a host of new possibilities for exchanges. AWS Transit Gateway is one of the only cloud-native tools that supports multicast, or the ability to transmit data to a group of destination computers at the same time, facilitating an efficient and equitable distribution of market data. Since many exchange architectures rely on multicast for transaction messaging, this functionality made the POC possible.

Building a Multiasset Exchange on the Cloud

AWS was a natural choice for the POC. “AWS has collaborated with us over the years,” says Peter Shen, head of digital services for the fixed income, currencies, and commodities unit at SGX. “We have worked alongside the AWS team on several pioneering initiatives, including one of the first cloud-based mobile apps in Singapore’s financial services industry and, more recently, the Amazon Managed Blockchain pilot, as part of our interledger delivery-versus-payment project.”

SGX knew that the cloud could support its goal of creating an easy-to-manage infrastructure that is scalable and retains performance while reducing costs. However, the team was unsure whether the cloud would be able to provide both the latency and jitter ranges required to host a trading engine, where market participants expect deterministic latency for order entry and execution.

As Asia’s most international and connected multiasset exchange operating equity, fixed income, and derivatives markets, SGX had to consider multiple factors when developing a cloud-native architecture. The infrastructure required to support the market must meet high availability and consistency standards, especially in regard to deterministic and repeatable latency during transactions. Exchanges are expected to provide fair access to trading systems and market data for all participants, which means that consistent and deterministic latency is more important than outright performance. “There’s the fair and orderly market requirement, as well as requirements for equality of access to market data and a level playing field for all participants,” says Adrian Ip, director of product management and technology sales for Aquis. “This POC was a way that we could prove that we can operate in the cloud and still meet these requirements."

Exceeding Performance Expectations

To test a cloud-based trading system, SGX collaborated with Aquis to implement a full exchange infrastructure encompassing the control server, matching engine, FIX and API gateways, and market data and logging services as well as an order management system to manage the tests. Aquis’s software required multicast to run the matching engine. “Multicast
is one of the primary mechanisms by which exchanges distribute market data today,” explains Shen. AWS Transit Gateway is one of the only cloud-native tools compatible with multicast, and with this feature, the service can handle the massive data volume that exchanges need to support.

The SGX team wanted to test whether the POC could provide a latency distribution of less than 100 microseconds—an important benchmark to see if the system could meet market
requirements. In this case, the SGX team defined latency distribution as the variation between mean latency and 99th percentile latency. “We wanted to demonstrate that we could achieve a consistent performance on a generic cloud infrastructure without resorting to more specialized services or bare metal servers,” explains Shen.

SGX ran a battery of 10 different benchmark tests—including tests on order entry, trade, and market data messages—to measure the round-trip latency distribution. After optimization, all 10 tests demonstrated a latency distribution of less than 100 microseconds around a mean latency of about 200 microseconds, a better-than-expected result on the cloud. “What this means is that we can give trading firms the reassurance that this is a market that has a consistent latency profile,” says Ip. Based on the results of the POC, SGX determined that hosting a central limit order book matching engine on the cloud is theoretically possible and that the POC architecture is a viable and scalable solution.

Migrating to a cloud-based architecture is one of the next logical steps in the evolution of the financial services industry. “All the financial transactions in the world take place out of maybe 20 to 30 physical data centers,” says Shen. “The cloud can provide not only the ability for financial markets to scale their services based on customer demand but also the ability to consolidate resources that will likely be required to remain in physical data centers, such as cloud-based infrastructure for hardware security modules.”

Moving toward a Distributed Financial Market

SGX is currently exploring how to digitize its businesses across different asset classes. According to the company, moving a matching engine to the cloud alone is not sufficient
to drive real transformation; trading participants, clearing members, brokers, and intermediaries will also need to migrate their execution management systems to the cloud to realize the full benefits of this next-generation architecture.

“We have managed to demonstrate that the latency we experienced on AWS Transit Gateway while using Aquis technology is within a reasonable range for matching engines,” Shen says. “I think with system adoption, there will be more regulated market infrastructures moving to the cloud in the coming years, hopefully sooner. And with that, you will see the next evolution of trading systems.”