Salesforce achieves optimal price and performance using AWS Graviton

Marketing agencies and enterprise marketing divisions rely on Salesforce’s Marketing Cloud Intelligence—originally developed by Datorama, an Israeli company acquired by Salesforce—for fast, reliable access to aggregated data.

The service’s dashboards facilitate quick decisions on time-sensitive issues, when query performance directly affects business decisions. These servers hold critical customer data, making trust and reliability essential. The infrastructure supporting Marketing Cloud Intelligence had been running on hundreds of Amazon Elastic Compute Cloud (Amazon EC2) R5 instances, which are optimized for workloads that process large data sets in memory.

In its quest to optimize both cost and performance for intensive analytic workloads, Salesforce’s Data Cloud Infrastructure team made a significant discovery. Vertica began officially supporting ARM-based AWS Graviton processors, a family of processors designed to deliver the best price performance for cloud workloads running in Amazon EC2. This development opened new possibilities for infrastructure optimization.

Following that finding, the team undertook a comprehensive analysis of Amazon EC2 instance offerings, examining both technical specifications and cost structures. AWS offers more than 1,000 different instance types across diverse compute families, each engineered for different workload characteristics. The team’s evaluation involved performance testing across several instance types: existing instances plus alternatives. The comparison included instances powered by the latest Intel processor generations alongside several iterations of AWS Graviton processors.

Salesforce's Data Cloud infrastructure team implemented a comprehensive proof-of-concept to evaluate seven AWS instance types through a systematic approach, which involved the following stages: (1) provisioning isolated environments that mirrored production specifications with actual data; (2) identifying approximately 10,000 representative queries covering diverse use cases to facilitate direct performance comparisons across environments; (3) establishing consistent processing procedures across all instance types; (4) developing a methodology for capturing key performance indicators per query, including CPU cycles, memory allocation, and processing time; (5) consolidating all query metrics for comprehensive benchmark analysis; (6) excluding network fetch time to isolate a pure processing performance comparison; and (7) generating comparative visualizations and charts displaying performance characteristics for each instance type.

The proof-of-concept delivered compelling results: Two instance types demonstrated substantially superior performance compared with all other tested configurations. These were Amazon EC2 R7g instances—memory-optimized instances powered by AWS Graviton3 processors—and Amazon EC2 R8g instances, which are ideal for memory-intensive workloads, such as databases, in-memory caches, and real-time big data analytics.

When viewed through a price-performance lens, Amazon EC2 R7g instances offered the most compelling value proposition and were selected for production migration. The whole initiative, from conceptualization to full production deployment, was done in just 1 month. Most impressively, this rapid migration was achieved with virtually no impact on service availability.

“We can always easily spin up a new cluster of servers without any problems,” says Dmitry Zbarski, manager at Salesforce. “If we need to test something or if we need to scale out, AWS provides us with all the agility we need. For temporary workloads, such as for proof-of-concept testing, we only pay for what we use, which makes experimentation cost-effective.”

Post-migration metrics on Amazon EC2 R7g instances exceeded proof-of-concept benchmarks. The deployment yielded 15 percent infrastructure cost savings. At the same time, query processing became 40–80 percent faster, with gains proportional to workload complexity patterns.

Dashboard performance improvements in Marketing Cloud Intelligence have transformed how marketing agencies access their data. “Our 40–80 percent query optimization dramatically accelerated dashboard rendering,” says Amit Berkun, lead engineer at Salesforce. “Even our most sophisticated dashboards, which used to take up to 15 seconds, now load within 5–7 seconds—so the migration to Amazon EC2 R7g instances has cut wait times by more than half.”

The data cloud infrastructure team also saw a 30 percent improvement in its load success rate. That was a significant achievement, given that the team was already operating at 99.95 percent reliability. Faster query processing helped reduce timeout failures, further enhancing service stability.

Building on its success, the team aims to enhance efficiency through strategic customer redistribution and load-balancing across databases. This optimization could reduce total instance requirements without sacrificing performance improvements. The proven results have created opportunities to replicate the approach across Data Cloud 360 teams, promising significant organizational impact.