Closed ncw closed 8 months ago
Before this change, we would spend an awful lot of CPU time reading and decompressing the same block to only read a small part from it.
This change caches the last block used in the file handle and for typical block sizes used in reads this makes a 10x difference in throughput.
This is great. Solid improvement.
Before this change, we would spend an awful lot of CPU time reading and decompressing the same block to only read a small part from it.
This change caches the last block used in the file handle and for typical block sizes used in reads this makes a 10x difference in throughput.