Optimize fast compaction memory

[choubenson]

Optimize the memory usage of FastCompaction from the following two aspects:

• Release compressed chunk data buffer when deserializing chunk into page queue.
• Release compressed page data buffer when deserializing page into points.

Experiment

Experimental Scenario:
There is one sequential file and two unsequential files, all of which completely overlap. Each file contains 5 devices, with each device having 80 time series. Each time series has only one chunk, and each chunk has one page, with 250,000 data points per page. Each timeseries adopts GZIP compression and PLAIN encoding. The experiment simulates a compaction scenario of three-way compacting, with 80 sub threads simultaneously compacting 80 timeseries.

Experimental Procedure:
Using binary search method, adjust the MAX_HEAP_SIZE parameter of the system gradually until the difference between the upper and lower bounds is less than 2M, and then determine the memory usage of the compaction algorithm.

Experimental Results:
Before optimization, FastCompaction algorithm occupies 1327M, and after optimization, it occupies 1187M.