I came across something on our Oracle 9.2 on Compaq TRU64 unix server:
diablo:/oracle/product/9.2.0/dbs>ps gvx | head
PID TTY S TIME SL PAGEIN VSZ RSS %CPU %MEM COMMAND
1294222 ?? S 5:36.53 2 2540 3.07G 928M 0.5 22.7 ora_dbw0_dbx2
1294609 ?? S 1:42.58 0 2 3.05G 717M 0.2 17.5 ora_i101_dbx2
1294635 ?? S 1:35.58 0 3 3.05G 703M 0.1 17.2 ora_i103_dbx2
1294633 ?? S 1:39.52 0 1076 3.05G 699M 0.0 17.1 ora_i104_dbx2
1294632 ?? S 1:35.30 0 2 3.05G 697M 0.0 17.0 ora_i102_dbx2
1433763 ?? S 6:54.39 0 1103 3.05G
The columns are not aligned correctly, but the 4 dbwr_io_slave process are taking up around 690-700M each ???
What does this mean ??
My first question is that why would you want to set the I/O slaves when the present day system supports the ASYNC I/O?
Life is a journey, not a destination!
Thanks for the reply Sam.
Can you point me to a doc that explains the async i/o function and how that relates to dbwr_io_slaves-thanks.
We are in the middle of moving from v815 to v92, and this param is something that we did not change.
Thanks for doc Shestakov.
I just talked to our Sys Admin guys, and they tell me that we do have controller cache, so, in theory, if Oracle does an i/o, the system call will be returned immediately, if the cache is not full.
Assuming the cache is usually not full, would you guys recommend turning dbwr_io_slaves to 0 and db_write_processes to 1 ?
Would it not be better to have 4 io slaves to do the writting ?? What is bad about that ??
[Edited by khussain on 10-02-2002 at 11:27 AM]
Click Here to Expand Forum to Full Width