Commit Graph

622 Commits

Author SHA1 Message Date
Sebastian Messmer
a4ce49aef4 Fix unique_ref test cases 2016-07-22 14:07:06 +02:00
Sebastian Messmer
64d26c2f6e Fix compiler error on older gcc 2016-07-22 12:58:59 +02:00
Sebastian Messmer
f069733650 Fix warning on older gcc versions 2016-07-22 02:57:03 +02:00
Sebastian Messmer
e38c0f1e8b Fix isRightBorderNode when growing trees 2016-07-17 10:32:59 +02:00
Sebastian Messmer
bbdd8a548b Special case treatment for traversals until exactly the last leaf has to load the last leaf to grow it. Before this commit, we just always loaded the last traversed leaf and checked its size. Now we only do so if it's the right border leaf. This saves us loading some leaves. 2016-07-17 10:16:09 +02:00
Sebastian Messmer
b339261924 Merge branch 'next' into lessio 2016-07-16 17:14:07 +02:00
Sebastian Messmer
f6141f515b TODOs 2016-07-16 17:08:53 +02:00
Sebastian Messmer
dc40d61595 Merge branch 'next' into lessio 2016-07-16 16:47:36 +02:00
Sebastian Messmer
e3690e2f6b Add comment 2016-07-16 16:47:27 +02:00
Sebastian Messmer
9c42703c70 Merge branch 'next' into lessio 2016-07-16 16:10:03 +02:00
Sebastian Messmer
143c3b3b4f Use large reads and large write 2016-07-16 16:09:47 +02:00
Sebastian Messmer
eab7cb1df4 Implement BlockStore::overwrite(). This is the last step in ensuring that the write() call doen't have to load leaves if they're only overwritten anyhow. 2016-07-16 11:42:06 +02:00
Sebastian Messmer
eb792daefc More preparations for not loading blocks in write() calls 2016-07-15 21:06:41 +02:00
Sebastian Messmer
d80b95514e Traversal doesn't load leaves, but returns the key. This way, the write() call can (after a future commit) write to full leaves without loading them beforehand. 2016-07-15 20:12:24 +02:00
Sebastian Messmer
582917c1f5 Improve code decreasing tree depth 2016-07-15 16:23:00 +02:00
Sebastian Messmer
b1b90b8c3d Further reduce number of nodes loaded when deleting a tree and write test cases for it 2016-07-15 15:42:34 +02:00
Sebastian Messmer
c428d5642a Minimize number of Block::write() calls when creating new nodes in a traversal. Also add test cases for it. 2016-07-15 12:48:20 +02:00
Sebastian Messmer
d626349802 Finish MockBlockStore to also collect other performance metrics, and implement the standard block store tests for it. 2016-07-15 10:23:09 +02:00
Sebastian Messmer
f42e08a5f6 Fix DataTree::key() when called while traversing 2016-07-14 17:56:47 +02:00
Sebastian Messmer
6ff0839e90 Add locking to blob 2016-07-14 16:36:30 +02:00
Sebastian Messmer
457ca52eb3 Add assertions to unique_ref 2016-07-14 16:29:02 +02:00
Sebastian Messmer
7ea08fc2b0 Add test cases for removing nodes 2016-07-14 16:12:51 +02:00
Sebastian Messmer
91bf6c3647 Keep resource destructor in ParallelAccessStore under mutex 2016-07-14 15:10:52 +02:00
Sebastian Messmer
3f04a7411c If tree depth increases, and we already traversed the old tree, don't go into it again to grow its last leaf. 2016-07-14 14:23:15 +02:00
Sebastian Messmer
cbe122cb5f Add assertions 2016-07-14 13:23:36 +02:00
Sebastian Messmer
0bb6b320d4 Add asserts 2016-07-14 12:58:42 +02:00
Sebastian Messmer
0dbfdf3e2b When shrinking a tree, LeafTraverser takes care of decreasing tree depth, not DataTree anymore. This way, LeafTraverser is now responsible for increasing and decreasing depth as necessary. 2016-07-14 12:55:54 +02:00
Sebastian Messmer
35a4e42ba5 Small fixes in traversal algorithm 2016-07-14 12:38:25 +02:00
Sebastian Messmer
fce81f47ff Simplify code a bit and add assertions 2016-07-14 12:03:15 +02:00
Sebastian Messmer
9e9369b9ed Performance optimization: When removing or shrinking a tree, we don't load/read leaves anymore. Loading inner nodes is enough to get all block IDs and then we can delete the leaves from their IDs without loading them. 2016-07-13 11:27:31 +02:00
Sebastian Messmer
7a68757599 Remove unneeded subtrees when shrinking tree 2016-07-13 07:08:53 +02:00
Sebastian Messmer
cf38eb0eb3 Use tree traversal to remove data tree. 2016-07-13 00:29:36 +02:00
Sebastian Messmer
508766d2f2 Fix traversal and add test cases 2016-07-13 00:22:35 +02:00
Sebastian Messmer
845b0b5239 Use LeafTraversor for resizing blobs 2016-07-12 20:36:12 +02:00
Sebastian Messmer
a19d79361c If a tree is grown by a traversal, the tree is kept in a balanced state as long as possible during the traversal 2016-07-12 09:57:56 +02:00
Sebastian Messmer
1bf20277e4 Removed unnecessary code 2016-07-12 01:10:27 +02:00
Sebastian Messmer
58f2212043 Don't query size on traversal 2016-07-12 01:09:30 +02:00
Sebastian Messmer
582c9c1a4c Fix traversal 2016-07-12 01:04:33 +02:00
Sebastian Messmer
98b85ea8b6 Started rewriting traversal 2016-07-10 22:57:39 +02:00
Sebastian Messmer
2bc3b641aa Allow Data::FillWithZeroes on rvalue 2016-07-10 22:43:07 +02:00
Sebastian Messmer
dde89da556 Improve traverseLeaves 2016-07-10 08:38:02 +02:00
Sebastian Messmer
0d860fa4f0 Speed up block loading/storing by issuing only one read/write syscall to the base file system 2016-07-06 15:35:51 -07:00
Sebastian Messmer
7348f7c64d Add checks to Data::LoadFromFile() and Data::StoreToFile() 2016-07-06 15:28:14 -07:00
Sebastian Messmer
e85019e95b Cache value of DataTree.numLeaves(). This should make read()/write() faster. 2016-07-05 23:56:44 -07:00
Sebastian Messmer
a242076ecc Performance improvement: Allow more parallelity when getting read()/write() syscalls to the same blob 2016-07-05 23:02:59 -07:00
Sebastian Messmer
ff0ba06846 Add mutex to fix race condition 2016-06-29 16:42:43 -07:00
Sebastian Messmer
b7b5dfba2a Fix generation of new file systems 2016-06-27 22:31:06 -07:00
Sebastian Messmer
3425760daf Implement migration for old filesystems that don't have parent pointers. 2016-06-27 20:03:44 -07:00
Sebastian Messmer
bfb397f44e Tell user to not interrupt migration process 2016-06-27 20:01:02 -07:00
Sebastian Messmer
397de9372f Each blob stores a parent pointer (i.e. the ID of the directory that contains this blob). This stores the directory structure in a conflict-proof way and can be used to resolve such conflicts. 2016-06-27 18:22:13 -07:00