483759a3de
by triggering the prefetching decoder path (which used to be dedicated to long-range offsets only). Figures on my laptop : no content prefetch : ~300 MB/s (for reference) full content prefetch : ~325 MB/s (before this patch) new prefetch path : ~375 MB/s (after this patch) The benchmark speed is already significant, but another side-effect is that this version prefetch less data into memory, since it only prefetches what's needed, instead of the full dictionary. This is supposed to help highly active environments such as active databases, that can't be properly measured in benchmark environment (too clean). Also : fixed the largeNbDict test program which was working improperly when setting nbBlocks > nbFiles. |
||
|---|---|---|
| .. | ||
| .gitignore | ||
| largeNbDicts.c | ||
| Makefile | ||
| README.md | ||
largeNbDicts
largeNbDicts is a benchmark test tool
dedicated to the specific scenario of
dictionary decompression using a very large number of dictionaries.
When dictionaries are constantly changing, they are always "cold",
suffering from increased latency due to cache misses.
The tool is created in a bid to investigate performance for this scenario, and experiment mitigation techniques.
Command line :
largeNbDicts [Options] filename(s)
Options :
-r : recursively load all files in subdirectories (default: off)
-B# : split input into blocks of size # (default: no split)
-# : use compression level # (default: 3)
-D # : use # as a dictionary (default: create one)
-i# : nb benchmark rounds (default: 6)
--nbDicts=# : set nb of dictionaries to # (default: one per block)
-h : help (this text)