Caching Levels

 

No caching

Local Cache

• Pos.
• No Latency
• Cons.
• Require RAM for cache on same machine which limited
• Leak in cache will kill application
• In-most require more CPU for garbage collecting

Local Cache: In-Process Cache for Immutable objects

• Pos
• No (de)serialization needed.
• NO latency - fastest of all possible caches
•  Cons.
• Require an app was designed to work with Immutable Objects which allows share them by  multiple threads.
• Servers: Possible only in thread-per-request servers (Java EE, IIS Asp.Net) and not possible in process-per-request servers (Apache, CGI)

Local Cache: In-Process Cache for Mutable objects

• Pos
• Minimum latency - deserialization and allocation time
• Cons
• (De)serialization penalty

Local Cache: Out-Process Cache 

• Latency between processes
• (De)serialization penalty

Distributed Cache

• HUGE network latency
• (De)serialization penalty 
• Scalable

http://syntx.io/in-process-caching-vs-distributed-caching/
http://openmymind.net/Do-More-In-Process-Caching/
http://java.dzone.com/articles/process-caching-vs-distributed

Still exploring this topic.
Don't really understand why many enterprise projects use Redis for caching DBs query caching:

• Same HUGE latency over network (x10-x100 than RAM)
• Speed also not significantly higher (x2-x5). This is because only simple DBs queries can be putted into cache (in other case you will should solve a lot of invalidation issues), and simple DBs queries are already in DBs cache.
• Apart, less locks inside DB 
• Apart, local memory have limits, plus GC.
• Apart, Redis have other sweet functional benefits SO: Redis Cache vs Memory , but this is not related to performance.