A strategy for handling cached write accesses.
WriteBack means that data from a write operation which goes through the Cache is not written back to the cached backend immediately. Instead, flusing is deferred until activity is low (or until a threshold on the amount of not yet written back data is reached).
This is more complex to implement than WriteThrough and is more risky (because the data in the Cache is not in sync with that on the cached backend), but can offer huge performance benefits.
F.ex, if the cached backend is a HardDisk, then the Cache may be able to entirely skip most of the updates to the FileSystem MetaData during extensive operation, because they tend to concentrate on the same piece of MetaData. Instead of updating the same directory 20 times in a row, once per file that gets moved somewhere else, the cache can defer updating until all the moves have happened, only actually writing back a single update a short while after the operation is completed. Of course, if someone pulls the plug halfway through, you'll have an inconsistent FileSystem.