As discussed earlier in this chapter, WiredTiger uses an internal cache to optimize performance. On top of it, there is always the filesystem cache that the operating system (and MMAPv1) uses to fetch data from disk.

By default, we have 50% of RAM dedicated to the filesystem cache and 50% to the WiredTiger internal cache.

The filesystem cache will keep data compressed as it is stored on disk. The internal cache will decompress it:

• Strategy 1: Allocate 80% or more to the internal cache. This has the goal of fitting our working set in WiredTiger's internal cache.
• Strategy 2: Allocate 80% or more to the filesystem cache. Our goal here is to avoid using the internal cache as much as possible and rely on the filesystem cache for our needs.
• Strategy 3: Use an SSD as the underlying storage for fast seek time and keep defaults at 50-50% allocation.
• Strategy 4: Enable compression in our storage layer through MongoDB's configuration to save on storage and potentially improve our performance by having a smaller working set size.

Our workload will dictate whether we need to deviate from the default Strategy 1 to any of the rest. In general, we should use SSDs wherever possible and with MongoDB's configurable storage, we can even use SSDs for some of the nodes where we need the best performance and keep HDDs for analytics workloads.