14.8.4: Case Study Analysis

Introduction to Cloud Computing · LibreTexts

0%

Contents

    Text

    Size
    100%
    Theme
    Typeface
    Line length
    Spacing

    Read aloud

    How

    Speed: 1×
    Light up
    Highlight

    Symbols: Mulberry Symbols, © 2018–2026 Steve Lee, CC BY-SA 4.0 (licence, the set), the site's own copies.

    From LibreTexts, under CC BY 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.

    14.8.4: Case Study Analysis

    Case Study Analysis

    FINRA: Serverless at Financial Regulatory Scale

    FINRA (Financial Industry Regulatory Authority) oversees U.S. securities markets, monitoring approximately 75 billion market events per day for fraud, manipulation, and regulatory violations. Their cloud transformation — moving from on-premises infrastructure to AWS — represents one of the most ambitious serverless deployments in any industry.

    Scale: FINRA processes more market events daily than Twitter, the New York Stock Exchange, and Visa combined. Their analytics platform must ingest, validate, and analyze this data within hours of market close to detect suspicious trading patterns.

    Serverless Architecture: FINRA uses AWS Lambda extensively for event processing — functions are triggered as market data flows in, performing validation, transformation, and pattern matching. They complement Lambda with Amazon S3 for data storage (500+ PB of historical data), Amazon EMR for large-scale analytics, and Amazon Redshift for interactive queries. The serverless approach means FINRA doesn't maintain idle servers during market off-hours — resources scale automatically during market hours and scale to near-zero overnight and on weekends.

    Cost and Efficiency: Before cloud migration, FINRA estimated it would have needed to purchase 20,000+ additional servers to handle growing data volumes. AWS serverless reduced their infrastructure costs by approximately 75% while handling 10x more data. The pay-per-use model means they only pay for actual compute time during market analysis.

    AI/ML Integration: FINRA increasingly uses machine learning models to detect complex fraud patterns that rule-based systems miss. They train models on historical market data using Amazon SageMaker and deploy them as part of their serverless analysis pipeline. As market manipulation techniques evolve, models are retrained and redeployed without infrastructure changes.

    Challenges: Cold start latency was initially problematic for time-sensitive regulatory alerts. FINRA addressed this using provisioned concurrency for critical Lambda functions. They also had to design careful data partitioning strategies to work within Lambda's execution time limits for large analytical jobs.

    Discussion Questions:

    1. FINRA processes 75 billion events per day using serverless functions. Analyze why serverless is particularly well-suited for FINRA's workload pattern (intense processing during market hours, near-zero outside market hours). Calculate the approximate cost difference between running 20,000 servers 24/7 vs. serverless pay-per-use for a workload that is active ~8 hours on weekdays.
    2. FINRA uses ML models trained on historical data to detect fraud patterns. Design a high-level MLOps pipeline for this use case: how would new market data be incorporated into training, how would models be evaluated before deployment, and how would you monitor model effectiveness in production? What happens if a model starts producing too many false positives?
    3. FINRA addressed Lambda cold start issues with provisioned concurrency for critical functions. Evaluate the trade-off: provisioned concurrency eliminates cold starts but means you're paying for idle capacity — which partially negates the serverless cost advantage. Under what circumstances is this trade-off worthwhile? How might future serverless improvements make this unnecessary?
    4. FINRA stores over 500 PB of historical market data on S3. As AI and new analytical techniques emerge, this historical data becomes increasingly valuable for training models and backtesting strategies. Discuss how the value of data changes over time in a cloud environment. How should organizations think about data retention, storage tiering, and the future analytical value of data they collect today?



    This page titled 14.8.4: Case Study Analysis was last modified on Tue, 22 Sep 2026 21:13:46 GMT and is shared under a CC BY 4.0 license and was authored, remixed, and/or curated by Felix Amoruwa.