The title blew a bit big, buffer cache in the different situation in fact, in front of each small piece have, did not repeat the paste.
1.buffer Cache Size:
such as 16cpu,32g memory:
Initial allocation:
SGA_TARGET=16G--Half of memory
Buffer cache=12g---SGA 40%, can be set to 70%-80%--conservative point can also be set to 10G, the latter can be based on system operation combined with buffer pool advisory modified.
shared_pool=3g
When the parameter: Statistics_level is typical or all, or the Db_cache_advice parameter is set to On,oracle, the optimal buffer cache size for the current instance is automatically estimated based on the system load-through V$db_ Cache_advice or ASR report: Buffer pool Advisory column. You can then set a reasonable value based on the recommended size combined with the current system memory usage.
The following is my virtual machine environment data:--size_for_estimate Unit is M, the system is currently buffer_cache size 32M.
sys@ Bys3>select size_for_estimate,buffers_for_estimate,estd_physical_read_factor,estd_physical_reads from V$db _cache_advice;
Size_for_estimate buffers_for_estimate Estd_physical_read_factor estd_physical_reads
----------------- -------------------- ------------------------- -------------------
4 496 2.4576 527548
8 992 2.3653 507728
12 1488 2.2055 473442
16 1984 1.887 405072
20 2480 1.3526 290349
24 2976 1.1332 243258
28 3472 1.0613 227818
32 3968 1 214661
36 4464.9342 200530
40 4960.8871 190428
44 5456.7987 171447
48 5952.7203 154614
52 6448.66 141675
56 6944.6248 134123
60 7440.5859 125766
64 7936.5529 118684
As you can see from this view: The current buffer_cache size of the system is 32m,estd_physical_reads field is the number of physically read blocks, the Estd_physical_read_factor field is the ratio of physical read/logical reading, the smaller the better.
As you can see, the larger the Buffer_cache, the less physical reading block, the smaller the ratio of physical read/logical reading, so my experimental environment is to increase buffer_cache. You can add to 64M first, then run for a period of time, and then continue to combine the V$db_cache_advice view, finally set an optimal value.