November 7th: Exploring an 80% lower cost hosted Redis alternative - register

Question: How does MongoDB handle cache eviction?

Answer

MongoDB utilizes an internal memory management system to handle its cache, which is crucial for its performance. The database leverages WiredTiger as its default storage engine (as of MongoDB 3.2 and later versions), which implements a sophisticated cache mechanism.

Understanding Cache in MongoDB

The WiredTiger storage engine maintains an in-memory cache where it stores recently read and modified data. Having data in-memory significantly speeds up the access times compared to reading from disk. The size of this cache is governed by the wiredTigerCacheSizeGB parameter in MongoDB configurations or can be automatically managed by MongoDB itself. By default, WiredTiger will use approximately half of the available system RAM minus 1 GB for its cache.

Cache Eviction Policy

Cache eviction is the process of removing older or less frequently accessed data from the cache to make room for new data. This is essential because the cache has a limited size. WiredTiger employs several strategies for cache eviction:

  • Least Recently Used (LRU) Algorithm: This is the primary method of cache eviction. Data that hasn't been accessed for the longest period is considered for eviction before more recently used data.
  • Dirty Data Eviction: Data that has been modified (referred to as 'dirty' data) is periodically written back to disk to ensure durability and to free up space in the cache for new or more frequently accessed data.
  • Checkpoints: At regular intervals, WiredTiger writes a consistent snapshot of all data to disk (a process known as checkpointing). This allows for some data that might only reside in memory to be moved to disk, thereby potentially qualifying more data for eviction under the LRU policy.

Tuning Cache Behavior

While the default settings work well for many applications, certain workloads might benefit from tuning cache settings. Here are a few parameters that can be adjusted:

  • Cache Size: Administrators can adjust the cache size based on their application's needs and available system resources. Increasing the cache size can reduce disk I/O at the cost of higher memory usage.
  • storage.wiredTiger.engineConfig.cacheSizeGB: This setting in the MongoDB configuration file directly sets the cache size in gigabytes.
{ "storage": { "wiredTiger": { "engineConfig": { "cacheSizeGB": 4 } } } }
  • Eviction Trigger Levels: Advanced configurations allow tuning when the eviction process starts and how aggressively it proceeds. These settings should be approached with caution, as improper values can lead to performance degradation.

It's important to monitor MongoDB's performance and adjust these settings as necessary based on the workload characteristics and the observed impact on latency and throughput.

Was this content helpful?

White Paper

Free System Design on AWS E-Book

Download this early release of O'Reilly's latest cloud infrastructure e-book: System Design on AWS.

Free System Design on AWS E-Book

Switch & save up to 80% 

Dragonfly is fully compatible with the Redis ecosystem and requires no code changes to implement. Instantly experience up to a 25X boost in performance and 80% reduction in cost