AWS for Industries

How Yahoo DSP Supercharged Their Advertising Tech with AWS Graviton processors

In today’s fast-paced digital advertising landscape, every millisecond and dollar counts. Following a rapid migration to the cloud, Yahoo DSP focused on reducing its cost and latency. Selecting the best compute instance for each workload was the target, as it was among the most significant cost items amidst aggressive budgeting goals.

How Yahoo DSP Approached the Challenge

Demand volume drives compute cost, and for Yahoo DSP this drives the cost for the Demand Side Platform (DSP) and the Profile Targeting Service (PTS). These platforms ran on older Intel-based instances (c6i family), incurring significant monthly costs becoming a focus for cost reduction.

The latest microprocessor advancements offered the potential to optimize compute resources and achieve better price-performance ratios. This included Intel’s Xeon Sapphire Rapids (c7i/m7i/r7i) and Amazon Web Services (AWS) Graviton3 (c7g/m7g/r7g) instance families. Applying the strategy to leverage instances delivering the best performance results across both compute families, Yahoo DSP achieved significant cost and performance gains across their DSP and PTS platforms.

AWS Graviton3 Powers Leaner, More Efficient PTS Read API

The PTS team set their sights on optimizing the Read API application tier. They compared the performance of Intel-based c6i instances to the latest AWS Graviton3 c7g instances, and the results from the AWS instances were impressive:

  • Increased throughput
  • Lower CPU usage per request
  • Reduced latency

This dramatic performance improvement allowed Yahoo to significantly reduce their Read API cluster footprint, cutting instances by 25% in us-east-1, 15% in ap-southeast-1, 30% in us-west-2, and a massive 50% in eu-west-1. The cost savings were substantial for Yahoo DSP.

Combined with the better pricing of AWS Graviton3 instances, these optimizations delivered a 40% savings for the PTS platform.

AWS Graviton3 Unlocks Further DSP Cost Reductions

Yahoo’s DSP platform, which facilitates media buying for advertisers, was originally deployed on Intel-based c6i instances. The team migrated the Bid and Optimization servers to the latest fourth-generation Intel Xeon-powered c7i instances to drive cost savings, achieving a 20% better price-performance ratio.

After migrating to the latest Intel family, they performed benchmarks for DSP Bid and Optimization servers using AWS Graviton3 c7g instances. Following positive results, they migrated their Bid servers to AWS Graviton3 c7g instances, which has provided an additional 20% cost reduction. The Optimization servers require floating point calculations, and Yahoo DSP didn’t observe significant improvements on AWS Graviton3. They decided to continue on c7i instances until they could evaluate the latest AWS Graviton4 family (c8g).

Embracing the Future with AWS Graviton

By proactively adopting the latest AWS Graviton processors, Yahoo DSP has driven dramatic cost savings and performance improvements across its critical advertising technology platforms, with minimal changes to the Java runtimes and libraries. The company expects these AWS Graviton-powered optimizations to deliver over $1.8 million in monthly savings, a testament to the power of staying on the cutting edge of cloud computing.

Looking to the Future

As Yahoo DSP continues its cloud journey, the strategic use of AWS Graviton will undoubtedly play a key role in maintaining a competitive edge and driving increased profitability. The next step for Yahoo DSP is the evaluation of the AWS Graviton4 families (i8g and c8g). They should bring improvements and further cost-performance benefits.

Conclusion

After finishing a significant migration to the cloud, Yahoo DSP focused on optimizing and reducing the costs of their Demand Side Platform and the Profile Targeting Service. Utilizing the AWS Graviton family of processors helped Yahoo DSP surpass its current processing speeds (90% reduction in latency at peak) while offering significant cost savings (40% savings for the PTS platform).

Other enterprises looking to optimize their cloud spend and performance can benefit from Yahoo DSP’s example and best practices by exploring the transformative potential of AWS Graviton. Start your AWS Graviton journey today.

Contact an AWS Representative to know how we can help accelerate your business.

Further Reading

Fernando Gamero

Fernando Gamero

Fernando Gamero is a Senior Solutions Architect engineer at AWS, with 25+ years of experience in the technology industry, from telecommunications, banking to startups. He is now helping customers with building Event Driven Architectures, adopting IoT solutions at the Edge, and transforming their data and machine learning pipelines at scale.

Keith McGovern

Keith McGovern

Keith McGovern is a Global Account Manager at AWS, with 25+ years of experience in account management across telecommunications, IT services and cloud computing.