>Some thoughts I had recently on the 0.7 datastore: > - Preallocation. We can give freenet a completely fixed space usage, > which would be useful for some users, by creating a full size empty > datastore on initialization. This is also advantageous in that there > is no chronological information leaked, as we can allocate blocks > randomly from a large empty store. And we don't need free blocks > support. If a block is no longer needed, simply demote it to the > back end of the LRU. We don't need to overwrite blocks freed from > usage by temp files as they were encrypted with temporary keys > anyway. On creation, we can write random data, with a valid hash, > for each block, and insert it at a random point in the LRU list.
this would be great, combined with a possibility to allocate more than one of these one-file-datastores (avoiding the mono-word here *g*), possibly on different disks. so users could distribute their freenet datastore evenly on different partitions (or even nfs drives) without filling up one single partition with a massive datastore. additionaly this multi-file-datastore model would enable the possibility to easily add another datastore to the node without having to tamper with an existing datastore or maybe even without shutting down the node (hook'n'use). this way even different datastore flavours are possible
