Home » Infrastructure » Unix » How to identify Asynchronous I/O?
| How to identify Asynchronous I/O? [message #287721] |
Thu, 13 December 2007 04:48  |
trantuananh24hg
Messages: 744
Registered: January 2007
Location: Ha Noi, Viet Nam
|
Senior Member |
|
|
Hi all!
I am newbie on Solaris.
With the server has 4 CPU, 8GB, Solaris 10, SPARC V890, I want to use db_writer_process initialization parameter.
However, according to the Mr.Tom's guide
| Quote: |
Normally, DBWR would do the IO. If the OS supports async io, it would use that. Async
IO just means that DBWR would make a request to the OS to write something to disk and
instead of waiting for it to be written, the write call would return immediate.
However, DBWR does not consider the write completed yet -- rather, DBWR will wait for the
OS to asyncronously signal it (call back) to say "write completed"
You can get much better throughput this way. DBWR might have 100 blocks to write out and
instead of write/wait/write/wait..... it will write write write.... and then wait for all
of the writes to complete -- the OS will notify it.
Now, some OS's/configurations don't support that -- so you can use IO slaves to emulate
it. DBWR will ask an IO slave to do the IO for it and the slave will notify DBWR upon
success (or failure of course) of the write just like the OS would have.
the number of slaves is more directly related to the number of disks, not cpus
|
and, according to the Oracle document..
| Quote: |
Database writer (DBWn)
The database writer writes modified blocks from the database buffer cache to the datafiles. Oracle Database allows a maximum of 20 database writer processes (DBW0-DBW9 and DBWa-DBWj). The DB_WRITER_PROCESSES initialization parameter specifies the number of DBWn processes. The database selects an appropriate default setting for this initialization parameter or adjusts a user-specified setting based on the number of CPUs and the number of processor groups.
|
| Quote: |
The DB_WRITER_PROCESSES initialization parameter lets you configure multiple database writer processes (from DBW0 to DBW9 and from DBWa to DBWj). Configuring multiple DBWR processes distributes the work required to identify buffers to be written, and it also distributes the I/O load over these processes. Multiple db writer processes are highly recommended for systems with multiple CPUs (at least one db writer for every 8 CPUs) or multiple processor groups (at least as many db writers as processor groups).
Based upon the number of CPUs and the number of processor groups, Oracle either selects an appropriate default setting for DB_WRITER_PROCESSES or adjusts a user-specified setting.
|
And would you like to answer my question:
- Does the server SPARC V890 with Solaris 10, 4CPU support Asyn IO?
- If it does, can I alter the parameter db_writer_processes=4 (<=CPUs) or I do nothing?
There is brief of my STATSPACK
| Quote: |
Load Profile Per Second Per Transaction
~~~~~~~~~~~~ --------------- ---------------
Redo size: 20,000.49 69,871.28
Logical reads: 1,139.38 3,980.41
Block changes: 82.66 288.79
Physical reads: 0.93 3.24
Physical writes: 2.67 9.34
User calls: 23.71 82.83
Parses: 7.10 24.79
Hard parses: 2.15 7.50
Sorts: 1.93 6.74
Logons: 0.05 0.19
Executes: 6.77 23.64
Transactions: 0.29
% Blocks changed per Read: 7.26 Recursive Call %: 59.72
Rollback per transaction %: 1.13 Rows per Sort: 479.96
|
and, especially
| Quote: |
Instance CPU
~~~~~~~~~~~~
% of total CPU for Instance: 24.57
% of busy CPU for Instance: 95.94
%DB time waiting for CPU - Resource Mgr:
Memory Statistics Begin End
~~~~~~~~~~~~~~~~~ ------------ ------------
Host Mem (MB): 8,129.7 8,129.7
SGA use (MB): 4,336.0 4,336.0
PGA use (MB): 125.9 170.3
% Host Mem used for SGA+PGA: 54.9 55.4
-------------------------------------------------------------
|
[Updated on: Thu, 13 December 2007 04:50] Report message to a moderator |
|
|
|
|
|
| Re: How to identify Asynchronous I/O? [message #287736 is a reply to message #287727] |
Thu, 13 December 2007 05:24  |
trantuananh24hg
Messages: 744
Registered: January 2007
Location: Ha Noi, Viet Nam
|
Senior Member |
|
|
Thank you, Michel!
If the OS support ASYN IO, Do I not need to chance the parameter dbwr_writer_process, because the OS does it?
| Quote: |
Just see if you have DBWR wait events or some "write complete" waits or "local write" waits
|
Is it the tacit talking that "Recommend the user client often commit"?
Avg
%Time Total Wait wait Waits
Event Waits -outs Time (s) (ms) /txn
--------------------------------- ------------ ------ ---------- ------ --------
log file parallel write 60,121 0 3,608 60 1.2
db file parallel write 63,648 0 3,006 47 1.2
log buffer space 2,690 74 2,274 845 0.1
log file sync 43,125 4 2,174 50 0.8
SQL*Net message from dblink 1,408,863 0 2,081 1 27.2
control file parallel write 60,849 0 1,226 20 1.2
os thread startup 2,993 9 430 144 0.1
log file switch completion 519 50 350 674 0.0
buffer busy waits 755 28 221 293 0.0
I really do not understand "buffer busy waits's values". Would you like to explain?
Thank for your time!
[Updated on: Thu, 13 December 2007 05:27] Report message to a moderator |
|
|
|
| Re: How to identify Asynchronous I/O? [message #287738 is a reply to message #287736] |
Thu, 13 December 2007 05:27 |
 |
Michel Cadot
Messages: 68716
Registered: March 2007
Location: Saint-Maur, France, https...
|
Senior Member
Account Moderator |
|
|
| Quote: |
Is it the tacit talking that "Recommend the user client often commit"?
|
I don't understand this sentence.
Client must commit ONLY when a transaction is over, no more.
Client should never have to commit because of other thing than its functional need.
Regards
Michel
|
|
|
|
Goto Forum:
Current Time: Sat Nov 23 21:11:32 CST 2024
|
] 
Blog
Search
Help
Members
Register
Login
Home
