AWS Security ChangesHomeSearch

AWS wellarchitected: Rewrite ADVPERF04-BP05 profile data guidance: retention, caching, region placement

Service: wellarchitected · 2026-11-11 · Documentation low

File: wellarchitected/latest/video-streaming-advertising-lens/advperf04-bp05.md

Summary

Restructures the performance best practice for managing high-volume user profile data: adds desired outcome, anti-patterns, benefits and risk level, and replaces the terse guidance with implementation steps covering DynamoDB TTL-based retention (30–90 days for bid path, up to 1 year for full audience profiles), Amazon MemoryDB cache-aside caching, storing profiles in the Region closest to users, and scheduled (once/twice daily) cross-region synchronization via Step Functions/Lambda and DynamoDB Streams. Also fixes headings, links, and related-document titles.

Security assessment

The change is framed entirely around performance, storage cost, and query latency SLOs (sub-millisecond bid path reads), not a vulnerability or incident. The retention/TTL and regional-placement guidance has incidental privacy/data-minimization and data-residency flavor, but it is presented as cost and latency optimization with no threat model, control objective, or security feature documentation, so it does not qualify as security documentation or a security fix.

Evidence

  1. **Configure DynamoDB TTL for profile data retention:** Set a TTL attribute on DynamoDB user profile items and configure DynamoDB's TTL feature to automatically expire items after your defined retention window (30–90 days from last update for bid path profiles). TTL deletions don't consume write capacity and incur no charge. Monitor the TimeToLiveDeletedItemCount CloudWatch metric to confirm TTL is deleting expired items at the expected rate.

Diff

diff --git a/wellarchitected/latest/video-streaming-advertising-lens/advperf04-bp05.md b/wellarchitected/latest/video-streaming-advertising-lens/advperf04-bp05.md
index fdfd514..a4704c6 100644
--- a//wellarchitected/latest/video-streaming-advertising-lens/advperf04-bp05.md
+++ b//wellarchitected/latest/video-streaming-advertising-lens/advperf04-bp05.md
@@ -7 +7 @@
-Implementation guidanceKey AWS servicesResources
+Implementation guidanceResources
@@ -9 +9 @@ Implementation guidanceKey AWS servicesResources
-# ADVPERF04-BP05 Manage high volume user profile data
+# ADVPERF04-BP05 Manage high-volume user profile data
@@ -11 +11,30 @@ Implementation guidanceKey AWS servicesResources
-The user profile database is typically large, ranging from 100-200 million to 5 billion user profiles and contains a wide range of data about users' online activities and interactions. Hence this should be retained for a short time in the range of 30 days -1-year max, to manage data storage costs and data query latency SLO’s. 
+User profile databases are typically large (100 million to 5 billion profiles) and require managing retention, geographic placement, and caching to control storage costs and maintain sub-millisecond bid path performance. 
+
+**Desired outcome:**
+
+  * Your user profile store delivers consistent sub-millisecond to single-digit-millisecond read performance at scale. 
+
+  * Retention policies bound storage costs and keep query latency within SLOs as the system scales. 
+
+  * Profile data is stored close to users geographically, and cross-region synchronization is scheduled rather than real-time to minimize transfer costs. 
+
+
+
+
+**Common anti-patterns:**
+
+  * Retaining user profile data indefinitely without a defined retention policy, which allows the dataset grow unconstrained until storage costs and query latency degrade. 
+
+  * Replicating user profile data across multiple regions in real time, when a given user is almost never accessed from more than one Region simultaneously. 
+
+
+
+
+**Benefits of establishing this best practice:**
+
+  * Bounded retention and local storage of user profiles keeps your database at a manageable size (preserving query latency SLOs as the system scales) while avoiding unnecessary replication costs. 
+
+
+
+
+**Level of risk exposed if this best practice is not established:** Medium 
@@ -15 +44,9 @@ The user profile database is typically large, ranging from 100-200 million to 5
-Use an in-memory database with a data cache strategy using Amazon MemoryDB. 
+User profile data has a staleness curve, where behavioral signals older than 30–90 days add less targeting value but still consume storage and inflate query scan times. Use a retention window of 30 days for high-velocity bid path profiles and up to one year for full audience profiles. 
+
+[Amazon MemoryDB](https://aws.amazon.com/memorydb/) provides durable, in-memory storage with Redis-compatible APIs. Use a cache-aside pattern (check MemoryDB before querying DynamoDB, and populate MemoryDB from DynamoDB on cache misses) to keep the most frequently accessed profiles in memory while DynamoDB stores the full dataset. 
+
+Advertisers use geotargeting, so where you store profile data affects bid path latency. A user's profile is accessed from the region they are located in for a given ad campaign. Store profiles in the AWS Region closest to where users are located to avoid cross-Region network latency that would add tens of milliseconds to every profile lookup on the bid path. 
+
+For architectures that require profile data in more than one Region (global campaigns), synchronize profiles on a schedule (once or twice per day). Real-time cross-region replication adds significant data transfer costs with minimal correctness benefit because users are rarely in two geographic locations simultaneously. 
+
+### Implementation steps
@@ -17 +54 @@ Use an in-memory database with a data cache strategy using Amazon MemoryDB.
-Avoid replicating user profile data across multiple Regions due to high latency and data transfer costs. We recommend storing user profiles local to the user. 
+  1. **Configure DynamoDB TTL for profile data retention:** Set a TTL attribute on DynamoDB user profile items and configure DynamoDB's TTL feature to automatically expire items after your defined retention window (30–90 days from last update for bid path profiles). TTL deletions don't consume write capacity and incur no charge. Monitor the TimeToLiveDeletedItemCount CloudWatch metric to confirm TTL is deleting expired items at the expected rate. 
@@ -19 +56 @@ Avoid replicating user profile data across multiple Regions due to high latency
-In the event of multi-Region architecture, implement synchronization between periodically (for example, once or twice a day) rather than in real-time, as users are unlikely to be in two locations at once. Advertisers also often use geotargeting, so a user's profile may only be accessed from the Region the user is located in for a particular ad campaign. 
+  2. **Deploy Amazon MemoryDB as your in-memory user profile cache:** Deploy MemoryDB in the same Availability Zone as your bidder compute. Implement a cache-aside pattern where bid evaluation services check MemoryDB first, populate on miss from DynamoDB, and update MemoryDB on profile writes. Configure MemoryDB eviction policy to evict least-recently-used items as the working set grows, keeping the most frequently accessed profiles in memory. 
@@ -21 +58 @@ In the event of multi-Region architecture, implement synchronization between per
-## Key AWS services
+  3. **Store user profiles in the Region closest to the user:** Partition your profile storage by user geography (EU users in EU regions, US users in US regions, APAC users in APAC regions). Configure your bidding infrastructure to route profile lookups to the regional store. Avoid the anti-pattern of centralizing all user profiles in a single Region and serving lookups globally over cross-region connections. 
@@ -23 +60 @@ In the event of multi-Region architecture, implement synchronization between per
-  * Amazon MemoryDB 
+  4. **Implement periodic cross-region synchronization:** For architectures that require profile data in multiple regions for global campaign targeting, implement a scheduled synchronization job ([AWS Step Functions](https://aws.amazon.com/step-functions/) \+ Lambda) that runs once or twice per day, replicating changed profiles to secondary regions. Use DynamoDB Streams to identify changed profiles since the last sync and transfer only the delta to minimize transfer volume. 
@@ -30 +67,12 @@ In the event of multi-Region architecture, implement synchronization between per
-  * [Observability best practices for Amazon Memory DB for Valke](https://aws.amazon.com/blogs/database/monitor-server-side-latency-for-amazon-memorydb-for-valkey/)y 
+**Related documents:**
+
+  * [Observability best practices for Amazon MemoryDB for Valkey](https://docs.aws.amazon.com/memorydb/latest/devguide/best-practices-observability.html)
+
+
+
+
+**Related services:**
+
+  * [Amazon MemoryDB](https://aws.amazon.com/memorydb/)
+
+  * [Amazon DynamoDB TTL](https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/TTL.html)
@@ -41 +89 @@ To use the Amazon Web Services Documentation, Javascript must be enabled. Please
-ADVPERF04-BP04 Enable detailed performance and observability monitoring to help tune queries and refine compute and storage 
+ADVPERF04-BP04 Enable detailed performance and observability monitoring to tune queries and refine compute and storage
@@ -43 +91 @@ ADVPERF04-BP04 Enable detailed performance and observability monitoring to help
-ADVPERF04-BP06 Consider AWS Clean Rooms collaboration
+ADVPERF04-BP06 Optimize privacy-enhanced collaboration queries for performance and service limits