This method is used in packages such as PNG, Zip, and gzip. Calculation and data loading are faster with direct I/O and zlib compression than with buffered I/O and zlib compression. If data storage is your greatest limiting factor, use zlib, but be aware that, under some circumstances, data loads may be up to 10% slower than bitmap compression. The size of the database, however, is generally significantly smaller when you use zlib compression.
In contrast to bitmap compression, which uses an algorithm to track which values are missing and does not interact with any other type of data, zlib compression builds a data dictionary based on the actual data being compressed (including any missing values). Therefore, zlib compression should provide greater compression ratios over bitmap compression, given extremely dense data. However, because the effectiveness of the zlib algorithm is dependent (at the bit level) on the actual data being compressed, general guidelines about when zlib compression provides greater compression than bitmap compression based solely on density are not available. Unlike other compression methods, the storage space saved has little or no relationship to the number of missing cells or the number of contiguous cells of equal value. Generally, the more dense or heterogeneous the data is, the better zlib will compress it in comparison to bitmap or RLE compression. However, under some circumstances, it is possible that zlib will not yield better results than using bitmap or RLE compression. It is best to test with a representative data sample.
To estimate the storage savings you may obtain with zlib, create a small database using your usual compression technique (bitmap or RLE) with a small sampling of real data and shut down Essbase Server. Note the size of the created data files. Then clear the data in the sample database, change the compression setting to zlib, reload the same sample data, and shut down Essbase Server again. Now note the difference in the storage used. You can also use the small sample database to estimate any changes in calculation or data loading speed.