Optimizing database IO using AIX strip

I. background

The production environment IBM P560 is currently used on the standby machine of Oracle universal uard. When oracle applies archive logs, check the server I/O and find that the I/O on Hdisk0 and 1 can sometimes reach 100%. There are very few I/O activities on Hdisk2 and 3.

After analysis, the oracle data file storage directory is stored in/oradata, which is created on hdisk0 and 1. Therefore, when the database reads and writes data files, only hdisk0 and 1 have IO. The other two disks have less I/O activity. Therefore, we want to migrate database files from rootvg and create database archive files in rootvg to evenly deploy the server IO.

Ii. theoretical support

Generally, the storage managed by LVM may have many PVS in one vg. Similarly, one lv may span multiple PVS. In order to speed up hard disk storage, the strip technology is used to divide continuous data into data blocks of the same size and store them on various PVS in sequence. Similar to RAID0, the storage speed is faster. But it does not make data as dangerous and easy to lose as RAID0, because in formal use, it will not make multiple hard disks into a vg as it is tested at this time without any guarantee, it is a commonly connected back-end storage. RAID5 or RAID1 has been prepared on the physical hard disk before the LUN is divided, and multiple Luns have been divided based on RAID5 or RAID1, that is, the pv on the system. Even if the hard disk where the pv is located is damaged, the underlying hard RAID redundancy does not cause data loss.

Stripe unit size: the size of each stripe unit in a striped LV, corresponding to the size of data blocks in I/O. For Oracle, db_block_size is the set data block size. While db_file_multiblock_read_count is the maximum number of parallel data blocks during one read, and db_block_size and db_file_multiblock_read_count are multiplied by the total I/O size. The size cannot exceed the maximum I/O (max_io_size) value of the operating system. In ORACLE applications, the lv strip size is generally set to twice or more than twice the size of the Oracle block, because if it is set to the same size as the Oracle data block, it cannot be ensured that the boundary of Oracle data blocks exactly corresponds to the boundary of the Strip unit. Otherwise, a large number of I/O segments are processed by two strip units.

The principle of strip size: in the case of high concurrency and small I/O requests, a physical hard disk processes multiple I/O requests, with low concurrency but large I/O requests, multiple hard disks may be required to process one I/O Request. In principle, one I/O request can be processed at a time.

This is probably the concept of striping.

Iii. Parameter Extraction

1. P560A:/# lspv
2. Hdisk0 00c3ee9e3439bc67 rootvg active
3. Hdisk1 00c3ee9e5033384d rootvg active
4. Hdisk2 00c3ee9eae48cc48 rootvg active
5. Hdisk3 00c3ee9eae48df75 rootvg active
6.  
7. P560A:/# lspv-l hdisk0
8. Hdisk0:
9. LVNAMELPs PPs DISTRIBUTION MOUNT POINT
10. Hd10opt 8 8 00... 00... 08... 00 .. 00/opt
11. Hd3 40 40 00... 00... 40... 00... 00/tmp
12. Hd1 4 4 00... 00... 04... 00... 00/home
13. Hd2 16 16 00... 00... 16... 00... 00/usr
14. Hd9var 4 4 00... 00... 04... 00 .. 00/var
15. Hd8 1 1 00... 00... 01... 00... 00 N/
16. Hd4 4 4 00... 00... 04... 00... 00/
17. Hd5 1 1 01... 00... 00... 00... 00 N/
18. Hd6 32 32 00... 00... 32... 00... 00 N/
19. Tsmdb 30 30 20... 10... 00... 00 .. 00/tsmdb
20. Oradatalv 278 278 49 .. 11 .. 00 .. 109 .. 109/oradata
21. Oraclelv 40 40 40... 00... 00... 00 .. 00/home/oracle
22. Weblogiclv 40 40 00... 40... 00... 00 .. 00 .. 00/weblogic
23. Weblogic9lv 40 40 00... 40... 00... 00 .. 00/weblogic9
24. Lg_dumplv 8 8 00... 08... 00... 00... 00 N/
25.  
26. P560A:/# lspv-l hdisk1
27. Hdisk1:
28. LVNAMELPs PPs DISTRIBUTION MOUNT POINT
29. Hd10opt 8 8 00... 00... 08... 00 .. 00/opt
30. Hd3 40 40 00... 00... 40... 00... 00/tmp
31. Hd1 4 4 00... 00... 04... 00... 00/home
32. Hd2 16 16 00... 00... 16... 00... 00/usr
33. Hd9var 4 4 00... 00... 04... 00 .. 00/var
34. Hd8 1 1 00... 00... 01... 00... 00 N/
35. Hd4 4 4 00... 00... 04... 00... 00/
36. Hd5 1 1 01... 00... 00... 00... 00 N/
37. Hd6 32 32 00... 00... 32... 00... 00 N/
38. Tsmdb 30 30 20... 10... 00... 00 .. 00/tsmdb
39. Oradatalv 324 324 89 .. 17 .. 00 .. 109 .. 109/oradata
40. Fwdump 2 2 00... 02... 00... 00 .. 00/var/adm/ras/platform
41. Weblogiclv 40 40 00... 40... 00... 00 .. 00 .. 00/weblogic
42. Weblogic9lv 40 40 00... 40... 00... 00 .. 00/weblogic9
43.  
44. P560A:/# lspv-l hdisk2
45. Hdisk2:
46. LVNAMELPs PPs DISTRIBUTION MOUNT POINT
47. Oradatalv 598 598 152... 223... 223... 00... 00/oradata
48.  
49. P560A:/# lspv-l hdisk3
50. Hdisk3:
51. LVNAMELPs PPs DISTRIBUTION MOUNT POINT
52. Archlog_lv 400 400 00 .. 223 .. 177 .. 00 .. 00/archivelog
53.  
54. P560A:/# lslv-l oradatalv
55. Oradatalv:/oradata
56. PV COPIESINBAND DISTRIBUTION
57. Hdisk0 278: 000: 000 3% 049: 011: 000: 109: 109
58. Hdisk1 324: 000: 000 5% 089: 017: 000: 109: 109
59. Hdisk2 598: 000: 000 37% 152: 223: 223: 000: 000
60.  
61. P560A:/# lslv-l oraclelv
62. Oraclelv:/home/oracle
63. PV COPIESINBAND DISTRIBUTION
64. Hdisk0 040: 000: 000 0% 040: 000: 000: 000: 000
65.  
66. P560A:/# lslv-l archlog_lv
67. Archlog_lv:/archivelog
68. PV COPIESINBAND DISTRIBUTION
69. Hdisk3 400: 000: 000 55%: 223: 177: 000: 000

1 2 Next Page