With AI transitioning from training to implementation, real-time data’s high velocity means even brief outages can result in significant gaps that can jeopardize AI inferences. In this blog, we’ll look at how low-latency dependent workloads make disaster recovery strategies business-critical with a focus on financial services providers as an instructive use case.
Algorithmic trading is increasingly powered by AI data, fed into “algo-trading” models that financial services organizations use to identify trading opportunities, and real-time interconnection enabling split-second (literally) execution of trading decisions.
Suffice to say, any interruption in the flow of data would be “expensive.” However, data loss also could jeopardize “algo back testing” which relies on complete and accurate historical data to validate the inferences directing algo-trading. Considering the value of current and past market data – and the data in any AI use case – it’s essential to develop a disaster recovery plan specifically for AI-powered algorithmic trading systems.
The first thing to consider when creating a DR plan for your algorithmic trading systems is where you put your trading infrastructure and AI model transaction data – data centers built for high-density AI workloads, interconnection and resilience offer an ideal environment.
It’s important to consider the reliability and uptime record of the site(s). Redundancy is key. If power, cooling, connectivity or security are not robust and redundant, your systems are vulnerable.
Latency, both to the trading platform and clouds, is especially important for this use case because it’s a machine-to-machine, real-time data transaction. Consequently, the recovery systems need to perform on par with the systems where the primary processing occurs. That makes a strong argument for establishing both redundant network interconnections as well as inter-site connectivity, making backup compute resources instantly available in the data center campus.
Checking these boxes in on-premises deployments is doable, but can be expensive in terms of space, equipment and onsite staff to monitor and manage them. That’s why many organizations, including those with on-premises deployments, choose to migrate their primary systems and remote failover and back-up systems to the cloud and/or to colocation data center facilities.
The 3-2-1 backup rule is one frequently recommended best practice for data backup and recovery. As described by ConnectWise, the 3-2-1 backup rule involves “maintaining three copies of data, utilizing two different storage formats, and storing one copy off-site. The primary objective is to enhance data protection and resilience, while safeguarding against threats such as cyberattacks, system failures or physical disasters. It is a strategic framework to ensure data can be swiftly and effectively restored during critical situations. By implementing redundancy in the backup strategy, the 3-2-1 rule significantly reduces the risk of complete data loss and helps organizations continue to operate even during cyberattacks or power outages.”
Key to the effectiveness of this and other data backup and recovery methods are geodiversity and real- or near-real-time connectivity to multiple copies of the data that are essential to AI-powered algorithmic trading operations.
When creating your data backup and recovery plans, factor in two key parameters:
Recovery Point Objective (RPO): The RPO isolates the point in time from which data needs to be restored. To define the RPO, organizations must determine how much data they can afford to lose without significantly impacting the business. This information helps organizations decide how frequently they should backup data. Frequent or continuous backups allow organizations to recover data from just before the disruption for minimal data loss.
Recovery Time Objective (RTO): The RTO defines how much downtime an organization can tolerate without significantly impacting operations. This metric factors in the impact and cost of downtime to help organizations determine how quickly they must restore operations.
We mentioned earlier that where you put your AI model data matters. While many organizations chose to deploy various systems, applications and data stores in the cloud when that option became available, many now see the advantages of using colocation for disaster recovery systems (as well as for a host of other purposes).
Colocation data centers, such as those owned and operated by CoreSite, are purpose-built to provide even the most demanding organizations with the secure, resilient, disaster-resistant architecture and infrastructure components necessary to support mission-critical, redundant data protection, data recovery and low-latency connectivity that are essential to every type of enterprise.
Specifically, CoreSite’s portfolio of colocation facilities:
With all this discussion of speed and systems, it’s easy to lose sight of a truly critical factor – the people running the data center. CoreSite data center operations personnel are trained to know how all the systems support disaster recovery, and regularly work through hypothetical “what if” scenarios as well as hands-on drills designed to hone critical thinking and response. The objective? Minimum downtime, even if there's an issue.
Each technician is also trained in physical security, how to communicate with customers and deliver white glove services ranging from equipment installation, monitoring, management and troubleshooting to operating generators, backup power systems and environmental controls.
Additionally, CoreSite provides ready access to its digital ecosystem, where customers can engage CoreSite partners such as managed service providers to outsource disaster recovery, data recovery, business continuity planning and other services.
That happens through the Open Cloud Exchange® (OCX), a software-define platform developed to simplify and automate enterprise-class network services between business partners and clouds. The OCX, as well as cross connects linking ecosystem members, enables secure, real-time transfer of data between data centers in a campus (and to clouds via direct cloud onramps) and dedicated network connectivity to data centers around the U.S., establishing the geodiversity that has been a pillar of DR strategies since the days of “East Coast/West Coast” business continuity best practices.
Know More
Learn how we can help you with your financial services hybrid IT strategy, including our approach to ensuring business continuity and low-latency access to real-time data.
Download, read and share the white paper, "Financial Services Industry Trends: Data Centers Enable Next-Gen Customer Experiences."
When you are ready, get in touch!