13.1.4 Reducing the HADB server startup time (by applying HugePages)

This section explains how to reduce the HADB server's startup time.

HADB's global buffers use the OS's shared memory. Consequently, if you specify in the HADB server definition that a large number of global buffers are to be allocated, it will occur frequently that a requested address will not be found in the TLB, which manages the OS's memory. This is called a TLB miss. As a result, the access performance of the shared memory will deteriorate, resulting in slowing down the HADB server's startup time.

In such a case, apply HugePages to HADB's shared memory. HugePages is a function that expands the TLB's page table. Expanding the TLB's page table reduces TLB misses and improves the access performance of the shared memory.

The following shows the procedure for applying HugePages to HADB's shared memory.

Setup method

1. Terminate the HADB server.

   Terminate the HADB server by executing the adbstop command.

2. To enable HugePages, specify the following kernel parameters:

   • vm.nr_hugepages parameter

   • vm.hugetlb_shm_group parameter

   • soft memlock parameter

   • hard memlock parameter

   For details about individual kernel parameters, see 6.2 Estimating the kernel parameters.

3. Restart the OS.

   If you changed the kernel parameter specifications, restart the OS.

4. Change the server definition.

   Specify the adb_sys_shm_huge_page_size operand in the server definition.

   If you changed the value of the vm.nr_hugepages parameter in step 2, change also the value of the adb_sys_memory_limit operand in the server definition.

5. Start the HADB server.

   Execute the adbstart command to start the HADB server.

Note

For details about the adb_sys_shm_huge_page_size and adb_sys_memory_limit operands in the server definition, see the adb_sys_shm_huge_page_size and adb_sys_memory_limit operands in 7.2.2 Operands related to performance (set format).

To Page Top