When businesses decide to move to the cloud, it means reworking applications, planning and deciding which workloads go where.
About a decade ago, Target started its cloud journey thanks to the holiday season. Hari Govind, SVP of infrastructure at Target, said that during the busiest time of the year, the public cloud provided the scalability the company needed.
“If we ran our seasonal workloads entirely on a private cloud, we would need to overcapitalize our compute and storage capacity and then underutilize this capacity many weeks a year,” Govind said in an email to CIO Dive. “It would be hugely inefficient.”
Today, the company runs on a hybrid multicloud architecture supported by Google Cloud Platform and Microsoft Azure.
Target didn’t just lift and shift to get here, though.
“Over 4,000 engineers at Target collaborated to fully modernize our application stack on an event-driven, microservices architecture rather than lift and shift legacy apps to the cloud,” Govind said.
Govind described a three-pronged approach to successful modernization:
- Rewrite apps to a microservice-based, asynchronous, event-driven architecture.
- Leverage open source.
- Operate a homogenous cluster management platform in all environments.
Despite intensive planning, Target experienced the same sticker shock that has plagued many businesses.
“Public cloud adoption and its associated cost were initially dramatic in 2018 and 2019 as app engineers across Target began adapting workloads,” Govind said.
This is a common problem that many businesses encounter post-cloud migration.
Going from a fixed cost model to a variable cost model requires a different kind of thinking, according to Shawn Ahmed, chief marketing officer at CloudBees.
More than eight in 10 enterprises say spend management is a top challenge in the cloud strategy, according to Flexera’s 2022 State of the Cloud report. The dynamics of cloud can make it easy for spending to skyrocket if there are no guardrails in place.
To mitigate the cost, Target invested in engineering. The retailer developed tools to manage workload placement, data movement, service consumption of raw compute, memory, disk and network bandwidth throughout the cloud and on-premise.
“All of this put together helped us slow down our public cloud growth to single digits in 2020 and 2021 despite Target’s unprecedented digital growth during those years,” Govind said.
In 2021, Target joined the Open Compute Project, the first major U.S. retailer to do so. The project was started by Facebook to design data center products that would be more energy efficient, less expensive and have the benefits of open source with open collaboration to hardware.
Along with open-source projects like SONiC, the company intends to continue driving down the cost of private cloud with disaggregated, commodity hardware.
The company plans to continue its current strategy by remaining hybrid, utilizing public cloud for scale and cost arbitrage and private cloud for data intensive workloads, line of business applications, enterprise systems and edge computing.
“We’ve learned that the long-term cost implications of public cloud make sense with either committed spend or on-demand consumption for transient workloads with spiky growth and extended, deep troughs where on-prem capitalization is inefficient,” Govind said. “We’re also acutely aware that purely from a unit of compute perspective, a private cloud can offer significant cost benefits.”
Advice from Govind
For businesses trying to mirror Target’s digital transformation success, Govind has a few pieces of advice:
- Intentionally investing in architecture to modernize applications, such as microservices, can future-proof the enterprise in terms of workload and data portability.
- Focus on FinOps — previously known as Performance & Efficiency Engineering — a practice dedicated to efficient utilization of cloud capacity and can help manage costs even in the medium-term.
- Large enterprises with a big private cloud footprint can benefit from developing a set of norms for hybrid cloud workload management, as well as building agility into the platforms that enable workload placement.
