greg/pcompress
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
290 commits 3 branches 21 tags 7.6 MiB
Commit graph

79 commits

Author SHA1 Message Date
Moinak Ghosh
fc65111bae Fix issue #11. 
Increase default chunk size to 8MB.
Use default compression level of 1 (fast mode) for LZ4.
 2013-08-24 22:58:50 +05:30
Moinak Ghosh
3db5188445 Support for deduplication using 2KB block size. 2013-08-19 13:38:52 +05:30
Moinak Ghosh
d31c6433c2 Update free memory computation to include cached buffers. 
Fix a potential rare corner case.
 2013-08-17 11:31:44 +05:30
Moinak Ghosh
f35d0ff4ef Fix multiple crashes for some corner cases. 
Increase max block size for variable dedup block sizes greater than 16KB.
Update test cases and fix a test script bug.
 2013-08-09 21:55:06 +05:30
Moinak Ghosh
fe18afbcf4 Use wrapper script to set paths when launching pcompress from build directory. 
Use smaller max block size when doing global dedupe.
Fix init of executable name.
 2013-08-07 22:03:52 +05:30
Moinak Ghosh
f34cfb1aa6 Make data partitioning between threads more effective. 
Remove unnecessary computation to make Fixed block chunking faster.
 2013-07-21 09:31:59 +05:30
Moinak Ghosh
8b73303488 Some minor code cleanup. 2013-07-05 22:22:11 +05:30
Moinak Ghosh
e10a13ad94 Improve accuracy of the KMV sketch computation and speed it up. 2013-07-03 19:24:06 +05:30
Moinak Ghosh
ab1ced942d Update invalid environment variable handling to actually fail rather than auto-correct. 2013-05-28 21:38:35 +05:30
Moinak Ghosh
e9ce7a5ed2 Fix a crash with invalid PCOMPRESS_CHUNK_HASH_GLOBAL. 
Update testcase to correctly detect core files.
 2013-05-26 23:38:10 +05:30
Moinak Ghosh
ddaa3b6b6d Drastic simplification of Min-heap code and resultant Delta speedup. 2013-05-25 17:34:38 +05:30
Moinak Ghosh
41b036adac Fix issue #8. 2013-05-10 19:51:24 +05:30
Moinak Ghosh
c27317d7da Add SSE2 optimizations for Segmented Dedupe. 2013-05-05 23:34:26 +05:30
Moinak Ghosh
6ecc400571 Fix segment offset sorting. 
Get rid of incorrect duplicate checks in index.
 2013-05-05 18:50:52 +05:30
Moinak Ghosh
c6da2325e3 Allow SKEIN to be used as a Global Dedupe chunk lookup hash. 2013-05-04 15:59:29 +05:30
Moinak Ghosh
0cf94c308a Add a qsort variant optimized for integers and use in global dedupe. 
Cleanup LZMA CRC64/32 declarations and add a header.
Fix heapq header.
 2013-05-03 22:06:55 +05:30
Moinak Ghosh
c43e99f422 Use openmp parallelism always for chunk hash computation during Global Dedupe. 2013-05-02 23:24:43 +05:30
Moinak Ghosh
120877348c Use SHA256 for Global Dedupe chunk lookup hash by default. 
Allow changing Global Dedupe chunk lookup hash via env variable.
 2013-05-02 00:05:05 +05:30
Moinak Ghosh
b23b5789fb Fix bugs and improve accuracy in Segmented Dedupe. 
Fix segment hashlist size computation.
Remove unnecessary sync of segment hashlist file writes.
Pass correct number of threads to index creation routine.
Add more error checks.
Handle correct positioning of segment hashlist file offset on write error.
Add missing semaphore signaling at dedupe abort points with global dedupe.
Use closer min-values sampling for improved segmented dedupe accuracy.
Update proper checksum info in README.
 2013-04-30 19:35:18 +05:30
Moinak Ghosh
074e265f70 Fix sizing of similarity hash buffer. 2013-04-26 22:36:14 +05:30
Moinak Ghosh
75f62d6a36 Simplify segment lookup loop. 
Fix assertion.
 2013-04-26 10:56:29 +05:30
Moinak Ghosh
5bb028fe03 Change Segmented Dedupe flow to improve parallelism. 
Periodically sync writes to segcache file.
Use simple insertion sort for small numbers of elements.
 2013-04-25 23:42:32 +05:30
Moinak Ghosh
79a6e7f770 Capability to output data to stdout when compressing. 
Always use segmented similarity bases dedupe when using -G option in pipe mode.
Standardize on average 8MB segment size for segmented dedupe.
Fix hashtable sizing.
Some miscellaneous cleanups.
Update README with details of new features.
 2013-04-24 23:03:58 +05:30
Moinak Ghosh
6c5d8d9e18 Optimize index lookup for 8-byte keys. 
More cleanups.
 2013-04-24 19:49:43 +05:30
Moinak Ghosh
eabd670790 Improve segment similarity detection and drastically reduce index size. 2013-04-23 23:15:32 +05:30
Moinak Ghosh
b32f4b3f9a Improve duplicate segment match detection. 2013-04-23 20:51:12 +05:30
Moinak Ghosh
6b7d883393 Tweak percentage intervals computation to improve segmented dedupe ratio. 
Avoid repeat processing of already processed segments.
 2013-04-23 18:53:56 +05:30
Moinak Ghosh
2c4024792a Several bugfixes. 
Avoid matching with self during hash lookup.
 2013-04-22 22:07:07 +05:30
Moinak Ghosh
6b23f6a73a Several fixes and optimizations. 2013-04-22 19:52:18 +05:30
Moinak Ghosh
c0b4aa0116 Many optimizations and changes to Segmented Global Dedupe. 
Use chunk hash based similarity matching rather than content based.
Use sorting to order hash buffer rather than min-heap for better accuracy.
Use fast CRC64 for similarity hash for speed and lower memory requirements.
 2013-04-21 18:11:16 +05:30
Moinak Ghosh
3b8a5813fd Many optimizations to segmented global dedupe. 
Use chunk hash based cumulative similarity matching instead of chunk content.
 2013-04-19 22:51:51 +05:30
Moinak Ghosh
426c0d0bf2 Properly cleanup global dedupe state. 2013-04-18 21:36:36 +05:30
Moinak Ghosh
8ae571124d Complete implementation for Segmented Global Deduplication. 2013-04-18 21:26:24 +05:30
Moinak Ghosh
a22b52cf08 Work in progress changes for Segmented Global Deduplication. 2013-04-14 23:51:54 +05:30
Moinak Ghosh
50251107de Work in progress changes for Segmented Global Deduplication. 2013-04-09 22:23:51 +05:30
Moinak Ghosh
3d7a179a77 Work in progress changes for scalable segmented global deduplication. 
Allow user-specified environment setting to control in-memory index size.
 2013-04-06 15:15:27 +05:30
Moinak Ghosh
c357452079 Implement global dedupe in pipe mode. 
Update hash index calculations to use upto 75% memavail when file size is not known.
Use little-endian nonce format for Salsa20.
 2013-03-29 15:18:25 +05:30
Moinak Ghosh
19b304f30c Add global index cleanup function. 
Fix location of sem_wait().
More comments.
 2013-03-25 21:04:16 +05:30
Moinak Ghosh
fbf4658635 Implement Global Deduplication. 2013-03-24 23:21:17 +05:30
Moinak Ghosh
876796be5c Work in progress changes for global dedupe. 2013-03-21 22:00:38 +05:30
Moinak Ghosh
b7fdeb08bc Work in progress global dedupe changes. 2013-03-20 22:47:03 +05:30
Moinak Ghosh
f8f23e5200 Major License text cleanup. 2013-03-07 20:26:48 +05:30
Moinak Ghosh
f89473d29c Fixes for issues/warnings reported in issue #4. 2013-02-15 22:53:17 +05:30
Moinak Ghosh
3e1737b4ab Use OpenMP parallelism when computing xxHashes for chunks. 2013-02-02 09:27:58 +05:30
Moinak Ghosh
af4c6e1d84 Reduce dedupe hash table collisions by half. 2013-01-31 00:38:41 +05:30
Moinak Ghosh
3d8f3ada1c Improve Deduplication performance by another 95%. 
Start sliding window scanning near minimum chunk size boundaries to avoid scanning whole chunk.
 2013-01-30 22:41:13 +05:30
Moinak Ghosh
5c8704c5bb Improve Deduplication throughtput by 90%. 
Use SSE4 register as sliding window for default 16-byte window size.
Use local variable for sliding window position to avoid spurios memory access in non-SIMD case.
Avoid computing breakpoint check value if processed length < minimum block length.
 2013-01-22 15:54:42 +05:30
Moinak Ghosh
3888c8d316 Many optimization tweaks 
Optimize Rabin Deduplication and Bsdiff
Vectorize XXHash using SE4
 2013-01-20 22:02:26 +05:30
Moinak Ghosh
455c8107d5 Use pre-increment for shorter instruction length and slight speed. 2013-01-17 22:54:30 +05:30
Moinak Ghosh
39dbc4be43 Implement algo-specific minimum distance match for Delta Compression. 2013-01-14 13:20:07 +05:30