Generic Cache Management Support

Last Modified: August 12th, 2004

Motivation

The memory footprint of Eclipse based applications can be large.

Tools build using Eclipse often cache data in memory to improve performance. Most tools will lazily load information from disk to improve startup time and ensure that memory is not used until required. However, there is often no stratgey in place to free up this memory if it is used infrequently by the user.

Another aspect of this problem involves tools that access remote data. Caching data from a remote source may involve two stages where the data is cached on the local disk and in-memory. However, the local disk caching may only be to persist the remote state accross restart and the in-memory data may not be freed once loaded.

Requirements

Caching is used for several purposes. Here is a list of the types of situations that require caches

• Resource synchronization state
• Remote data such as repository content
• Domain models

There are several characteristics to consider for the data being cached:

• Mutable: A cached object may be static or mutable. For example a cache used to show the state of a remote resource would not be mutable. However, a cached domain model or resource synchronization state would most likely be modifiable and changes made to the model objects would need to be written back to the source (most likely a file on disk).
• Computed: Computed data may be cached to disk for performance. The disk cache may also be considered a cache that needs to be monitored depending on the cost of recomputation and how often it is used.
• Replicated: (Similar to computed) Remote data may be replicated on the local disk for performance. This replica is itself a type of cache which could also be cleared at some point, if resources were tight or if it were not used a great deal.

Proposed Solution

CacheManager

• consists of a set of caches
• runs a background job that tests caches for the need to unload some or all of the cache
• after a run, determines when to reschedule based on state/retention policy of caches
• will reschedule based on changes to a particular cache (i.e. if a cache size grows, the job may be run immediatly to reclaim)

Cache

• a set of cache entries
• a retention policy
• a persistence policy (if applicable)

Retention Policy

• size limit: reclaim when size is exceeded based on LRU ordering.
• low water mark: when size is exceed, free down to low water mark.
• expiry time: reclaim expired entries regardless of size
• conditional: reclaim entries whenever a certain condition is met

Persistence Policy

• synchronous: batched or unbatched
• asynchronous: immediate or scheduled