Grab's Time-Aware LRU Cache for Android: Optimizing Image Storage

The Challenge with Traditional LRU Caching

Traditional LRU caches, while effective for many use cases, present specific challenges in mobile applications with large media assets. Grab's Android app used Glide with a 100 MB LRU cache for images. This led to two primary issues: the cache frequently filled up for active users, causing performance degradation, and for less active users, images could persist for months, wasting storage space even if they were no longer relevant after a certain period.

Introducing Time-Aware LRU (TLRU) for Better Cache Management

To address the limitations of LRU, Grab implemented a Time-Aware Least Recently Used (TLRU) cache. TLRU extends the LRU eviction policy by incorporating a time-based expiration mechanism, allowing cached items to be evicted not only based on their recency of access but also their age. This hybrid approach helps ensure that stale data doesn't consume valuable storage indefinitely.

TLRU Key Parameters

• Time To Live (TTL): Defines the maximum duration an entry can remain in the cache before being considered expired.
• Minimum Cache Size Threshold: Ensures that critical images remain cached even if they expire, as long as the total cache size is below this threshold, preventing an underpopulated cache.
• Maximum Cache Size: The hard limit for the total storage the cache can consume.

Implementation Strategy: Extending DiskLruCache

Instead of building TLRU from scratch, Grab engineers opted to fork and extend Glide's existing `DiskLruCache` implementation. This decision leveraged a "mature, battle-tested foundation" that already handles complex aspects like crash recovery, thread safety, and performance optimizations. The extensions primarily involved adding last-access time tracking, implementing time-based eviction logic, and devising a migration mechanism for existing LRU caches.