Optimizing database Io with AIX stripe (stripe)

I. BACKGROUND

Production environment IBM P560 is currently used for Oracle DataGuard standby machines. When Oracle Apply Archive log, look at the server IO situation, found that Hdisk0, 1 io sometimes reach to 100%. and HDISK2, 3 IO activities are very few.

After analysis, Oracle's data file storage directory is located in/oradata, which is based on Hdisk0, 1. Therefore, when the database occurs to read and write data files, only hdisk0,1 have IO. The other 2 disk IO activities are less. Therefore, it is considered that the database files should be migrated from ROOTVG, and the archive files of the database should be built in the ROOTVG to achieve the average distributed server IO.

Second, theoretical support

Typically stored in LVM management, a VG may have a lot of PV, similarly, a LV may span more than one block of PV, in order to make the hard disk storage speed, you will use the stripe technology, that is, the continuous data into the same size block of data, and then stored in each PV. Similar to RAID0, which speeds up storage. But it does not make the data as dangerous as RAID0 easily lost, because in the formal use, not as a test at this time without any security to the multiple hard disk into a VG, but generally connected to the background storage, before the LUN, has been done on the physical hard drive RAID5 or RAID1, RAID5 or RAID1 on the basis of the partition of more than one LUN, that is, PV on the system, even if the PV on the hard disk damage, but the underlying hard raid redundancy, and will not lose data.

Stripe unit Size: The Strip-band LV, the size of each stripe unit, corresponds to the size of the data block in I/O. For Oracle, Db_block_size is the size of the data block that is set. The number of data blocks that are db_file_multiblock_read_count in parallel for one read, db_block_size and Db_file_multiblock_read_count multiplied by the total I/O size. This size cannot exceed the operating system's maximum I/O (max_io_size) value. In Oracle applications, the LV stripe size is generally set to twice times or more than twice times the size of the Oracle block, because if set to the same size as the Oracle data block, it is not guaranteed that the boundaries of the Oracle data block exactly correspond to the boundaries of the stripe unit, if not, there will be a large number of an I /O is handled by two strip units.

The principle of stripe size: For high concurrency and small IO requests, a physical hard disk handles multiple I/O requests, with low concurrency but large I/O requests, which may require multiple hard drives to handle an I/O request. The requirement in principle is that an I/O request can be processed at one time.

The idea of a general stripe is this.