11 * Better on-disk placement to reduce seek times
13 * "Free copies" - cache recently read/written block CRCs, on write check CRC and on match verify that the blocks
14 are the same. Then just create a btree link to the original data. This could also be done in a hammer cleanup
15 pass, effectively de-duplicating the file system.
18 ## Memory Management (Kernel)
22 Overal goal: reduce malloc() use and replace with object cache allocations. This should increase performance.
27 * Make free() nonblocking. At the moment a call to free() might block because it might need to
29 aquire the BGL and maybe needs to wait for other locks (needs to be checked).
33 This allows us to use free() with dragonfly matching semantic, i.e. it can be used in critical sections
35 or with a spinlock held/in an interrupt.
39 For this to work, we need to:
44 * push locking into vm_map_* routines
49 * make sure no code path from free() might need to block
56 * polish up the objcache to work like bonwick described it, most important: make magazines adjust
58 their size depending on depot lock contention
63 * implement a real slab allocator (bonwick) and switch the objcache to use it. It could us
65 per-cpu (not so good, i think) or per-numa-cluster slabs, but in the beginning it for sure is possible
67 to use one slab for all, as most allocations should be satisfied by the objcache (and be serialized
69 anyways already by the depot lock)
74 * find and change code paths assuming power-of-2 allocations are power-of-2 allocated
79 * switch malloc() and free() to use the objcache for reasonable sized allocations
81 use kmem for larger ones
89 Investigate why the system does not boot within one second. I have a fast box, this should be possible. Somewhere are delays adding up. Maybe use kernel threads for probing/initialisation.
97 Port [pmc](http://people.freebsd.org/~jkoshy/projects/perf-measurement/) (and maybe something like oprofile), so that we can find problematic areas.