Feature Specifications

If the TPC-C policy is enabled but scheduling is disabled, existing services are not affected.
If the TPC-C compression policy is disabled, existing services are not affected.
If the ILM policy is set by using TPCC.bmsql_order_line (only orders that have been delivered are identified as cold rows) but scheduling is disabled, the tpmC deterioration is not higher than 2% (taking the 56-core CPU, 370 GB memory, 3 TB SSD, and 350 GB shared buffer as an example).
If the ILM policy is set by using TPCC.bmsql_order_line (only orders that have been delivered are identified as cold rows) but parameter scheduling is enabled by default, the tpmC deterioration is not higher than 5% (taking the 56-core CPU, 370 GB memory, 3 TB SSD, and 350 GB shared buffer as an example).
The processing rate of a single-thread ILM job is about 100 MB/s (taking the 56-core CPU, 370 GB memory, 3 TB SSD, and 350 GB shared buffer as an example).
The processing rate can be measured based on the start time and end time of compression and the number of compressed pages.
When GET is used to query data, the performance of accessing compressed data deteriorates compared with that of accessing non-compressed data. The performance deterioration on the driver side is not higher than 10%, and that on the PL/SQL side is not higher than 15% (taking the 32 MB shared buffer and 60,000 data pages as an example).
When multi-get is used to query data, the performance of accessing compressed data deteriorates compared with that of accessing non-compressed data. The performance deterioration on the driver side is not higher than 30%, and that on the PL/SQL side is not higher than 40% (taking the 32 MB shared buffer and 60,000 data pages as an example).
When table-scan is used to query data, the performance of accessing compressed data deteriorates compared with that of accessing non-compressed data. The performance deterioration on the driver side is not higher than 30%, and that on the PL/SQL side is not higher than 40% (taking the 32 MB shared buffer and 60,000 data pages as an example).
The compression ratio of the TPCH.lineitem table (all cold rows) is greater than or equal to 2:1.
Tests on the Orderline table of TPC-C and the Lineitem, Orders, Customer, and Part tables of TPC-H show that the compression ratio is higher than that of LZ4 and ZLIB when there are many numeric columns; when there are a large number of text columns, the compression ratio is between that of the compression algorithms of the LZ class and the LZ+Huffman class.