Vous êtes sur la page 1sur 2

AWR:

If we want to generate AWR report for a parcticular sql statement , then we


generate it for short duration, while for a non-specific problem , we do it for
long couple of hours

1. Elapsed Time - duration between snapshots


2. DB Time - The total time spent in the DB - sum of cpu time and non-idle wait
time(server process waiting for latches and locks to complete)

DB Time/Elapsed Time = DB load

Latches : lightweight serialization mechanisms which restrict access to sga data


structures. They are visible inside the local instance of Memory.
Locks: Heavy serialization mechanism which restricts access to the transactions.
They are visible for the transactions of the whole DB.

if DB Time > Elapsed Time, no of concurrent active sessions are more


No.of inactive sessions = Foreground Wait Events=>sql *net message from client=>
(Total Wait Time(s))/(elapsed time * 60)

3. if DB CPU > No.of cores, then it's a cpu bound system


otherwise it's an I/O bound system
4.Check the No.of Sessions = increasing/reducing/constant Load of the DB
5. Load Profile : Physical Reads, Logical Reads, Block Changes,Rollback,
Transactions LR > PR => good Trans are high and Block Changes are less => DSS
system which means few sql running Long time
User calls & Parses are low => DSS
if Sorting is high => DSS
if block changes>=Transactions => OLTP systems
6.Execute to Parse => How many Times Parsed sql statements are re-used
7.Parse CPU to Parse Elapsed : How long CPU Time was spent for parsing sql
statements
8.if the above ratio is lesser, then check for cursor_sharing and bind variables
usage
9.Instance Efficiency % - 95 to 100 %
10. Shared Pool Stats % - 70 to 80 %
11.Wait Events :

a)db file sequential read(index read)- reads single block -


not using specified index,High I/O on particular disk or mount point,bad
application design,fragmented indexes,wrong column order of index etc.,-Rebuild
Indexes,updated stats,table partition, force FTS

b)db file scattered read(FTS) - From disk to SGA multiple blocks to be


read,FTS,FFIS - Use Keep pool,Minimize the number of Physical and Logical Reads,
sql tuning/appn tuning

c)buffer busy wait - One Session is trying to access a block, while the other
session might try to access it-read/read,read/write/write/write- Reduce the number
of rows per block, increase the no of rollback segments, increase the number of
freelists and freelist groups or increase the extent size

d)logfile sync: Oracle Server Process is waiting to commit or rollback , too many
commits/rollback,more archivelog generation, table and index files might have been
kept in same disks-
Reduce no of commits or by batching transactions or keeping tables and indexes at
separate disks

Vous aimerez peut-être aussi