Two Sigma had a big problem in 2014: the compute power needed for the quantitative fund's research workflows was 10 times greater than what its data centers could provide.
"We said, 'You know what, we're not going to build out a 10x physical presence. That's just enormous, that feels wrong to us," Camille Fournier, Two Sigma's head of platform engineering, told Insider.
As a quantitative fund, the problem was particularly salient for Two Sigma. Quant funds rely on mathematical and computer-based modeling to make their bets in the market, meaning their demand for computer firepower can often be enormous. Two Sigma, founded by billionaires John Overdeck and David Siegel, is known for pushing the limit on computing and data usage, even amongst its fellow quant peers.
Instead of building more physical data centers, the decision was made to push the fund into the public cloud.
Today, the New York-based fund now has a cloud strategy where the majority of its workloads are spread across multiple public clouds.
The $58 billion fund uses a combination of public cloud providers — mostly Google Cloud Platform (GCP), in addition to Amazon Web Services and Microsoft Azure — to host a wide set of programs and applications. About 70% of the firm's capacity is on the public cloud, with the remainder hosted in on-premise data centers, Fournier said.
As Two Sigma's cloud strategy has matured, so too has its hiring of engineers focused on the tech.
When Fournier joined the firm in 2017, around 10% of her platform-engineering group focused on public-cloud work. Now, half her team of more than 100 engineers works on the tech.
And that group is set to grow "by several hundred engineers," in the coming years, as Two Sigma hires more people with a focus on data and cloud-focused distributed systems, Fournier added.
'Multicloud doesn't necessarily mean cloud agnostic'
In the beginning of its cloud journey, Two Sigma was mostly using AWS to host its research workloads.
But the firm has shifted much of that work to GCP in a move that began around 2017 as Two Sigma expanded its cloud capabilities, Fournier said.
The use of BigQuery, GCP's data warehousing and analytics service, for Two Sigma's research work was a driver in its adoption of Google Cloud.
"There really isn't a comparable product that we could be running on prem," Fournier said. There was also a lower cost of ownership when using services built in the cloud versus building something internally, she added.
Two Sigma still leverages AWS for certain storage capabilities, and is currently evaluating scaling up certain research workloads there, Fournier added. It's also using Microsoft Azure, and the fund still maintains its physical data centers despite its cloud push.
But while Two Sigma takes a multi-cloud approach, the goal isn't to have all tools operate on multiple clouds.
The fund adopted a fit-for-purpose approach that sees it use different tools in different circumstances.
"Multi-cloud doesn't necessarily mean cloud agnostic," Carter Page, head of data engineering at Two Sigma, told Insider.
Two Sigma's hires have also reflected an increasing embrace of GCP. The firm poached Page, a former director of engineering at Google where he built cloud databases and led a team of 100, this past June. Page now oversees how data is ingested, standardized, and flows through Two Sigma's tech stacks and to its investment teams.
The fund's former chief technology officer, Alfred Spector, was also a Googler before joining Two Sigma, which has poached the most talent from big tech firms of any fund over the last decade.
A new approach to compute
Two Sigma has also recognized cost savings by using a cloud tool that allows it to leverage GCP's excess compute power at a reduced cost, a feature called preemptible virtual machines.
The tradeoff is that the compute power isn't guaranteed, since GCP might not always have leftover compute capacity. As a result, preemptible VMs are ideal for tools and apps that don't need continuous compute power.
Still, the feature can lead to drastic savings. According to Google Cloud, preemptible VMs have lowered compute engine costs by 80% in some cases. AWS and Microsoft Azure also have similar tools for their excess compute.
The tool has led to "tremendous savings," Page said, and prompted Two Sigma to consider doing more batch workloads, where computing is scheduled to be done at intervals, to leverage preemptible VMs.
The effort is best exemplified in its research, where the firm runs both daily simulations and long-run, historical regression analyses.
"We actually have done a lot of work to make it so that we can checkpoint and restart workloads when they get interrupted so that we don't lose progress, so we can continue to make forward progress on these very large scale computations," Fournier said.
As Two Sigma realizes advantages related to cost and faster access to cloud-based tools and machine types, there's an "inevitability of just more and more of our internal operations mov[ing] to cloud," Page said.
