More Examples of Interpreting Wait Events To Boost System Performance Roger Schrag and Terry Sutton Database Specialists, Inc. www.dbspecialists.com

Session Objectives Briefly introduce wait events: Define wait events Discuss how to use the wait event interface Walk through five examples of how wait event information was used to troubleshoot production problems

“Wait Event” Defined We say an Oracle process is “busy” when it wants CPU time. When an Oracle process is not busy, it is waiting for something to happen. There are only so many things an Oracle process could be waiting for, and the kernel developers at Oracle have attached names to them all. These are wait events.

Wait Event Examples An Oracle process waiting for the client application to submit a SQL statement waits on a “SQL*Net message from client” event. An Oracle process waiting on another session to release a row-level lock waits on an “enqueue” event.

Wait Event Interface Each Oracle process identifies the event it is waiting for each time a wait begins. The instance collects cumulative statistics about events waited upon since instance startup. You can access this information through v$ views and a wait event tracing facility. These make up the wait event interface.

Viewing Wait Events http://dbrx.dbspecialists.com/pls/dbrx/view_report

Why Wait Event Information Is Useful Wait events touch all areas of Oracle—from I/O to latches to parallelism to network traffic. Wait event data can be remarkably detailed. “Waited 0.02 seconds to read 8 blocks from file 42 starting at block 18042.” Analyzing wait event data will yield a path toward a solution for almost any problem.

Important Wait Events There were 158 wait events in Oracle 8.0. There are 363 wait events in Oracle 9i Release 2 (9.2.0). Most come up infrequently or are rarely significant for troubleshooting performance. Different wait events are significant in different environments, depending on which Oracle features have been deployed.

A Few Common Events buffer busy waits library cache load lock control file parallel write library cache pin control file sequential read log buffer space db file parallel read / write log file parallel write db file scattered read log file sequential read db file sequential read log file switch completion direct path read / write log file sync enqueue undo segment extension free buffer waits write complete waits latch free

Idle Events Sometimes an Oracle process is not busy simply because it has nothing to do. In this case the process will be waiting on an event that we call an “idle event.” Idle events are usually not interesting from the tuning and troubleshooting perspective.

Common Idle Events client message PX Deq: Execute Reply dispatcher timer PX Deq: Execution Msg gcs for action PX Deq: Signal ACK gcs remote message PX Deq: Table Q Normal ges remote message PX Deque wait i/o slave wait PX Idle Wait jobq slave wait queue messages lock manager wait for remote message rdbms ipc message null event slave wait parallel query dequeue smon timer pipe get SQL*Net message from client PL/SQL lock timer SQL*Net message to client pmon timer SQL*Net more data from client PX Deq Credit: need buffer virtual circuit status PX Deq Credit: send blkd wakeup time manager

Accounted for by the Wait Event Interface Time spent waiting for something to do (idle events) Time spent waiting for something to happen so that work may continue (non-idle events)

Not Accounted for by the Wait Event Interface Time spent using a CPU Time spent waiting for a CPU Time spent waiting for virtual memory to be swapped back into physical memory Time spent on CPU-intensive activities: Logical reads Spinning while waiting for latches Statement parsing

Timed Statistics The wait event interface will not collect timing information unless timed statistics are enabled. Enable timed statistics dynamically at the instance or session level: ALTER SYSTEM SET timed_statistics = TRUE; ALTER SESSION SET timed_statistics = TRUE; Enable timed statistics at instance startup by setting the instance parameter: timed_statistics = true

The Wait Event Interface Dynamic performance views v$system_event v$session_event v$event_name v$session_wait Wait event tracing

The v$system_event View Shows one row for each wait event name, along with cumulative statistics since instance startup. Wait events that have not occurred at least once since instance startup do not appear in this view. Column Name Data Type -------------------------- ------------ EVENT VARCHAR2(64) TOTAL_WAITS NUMBER TOTAL_TIMEOUTS NUMBER TIME_WAITED NUMBER AVERAGE_WAIT NUMBER TIME_WAITED_MICRO NUMBER

Columns In v$system_event EVENT: The name of a wait event TOTAL_WAITS: Total number of times a process has waited for this event since instance startup TOTAL_TIMEOUTS: Total number of timeouts while waiting for this event since instance startup TIME_WAITED: Total time waited for this wait event by all processes since startup (in centiseconds) AVERAGE_WAIT: The average length of a wait for this event since instance startup (in centiseconds) TIME_WAITED_MICRO: Same as TIME_WAITED but in microseconds (Oracle 9i)

Sample v$system_event Query SQL> SELECT event, time_waited 2 FROM v$system_event 3 WHERE event IN ('smon timer', 4 'SQL*Net message from client', 5 'db file sequential read', 6 'log file parallel write'); EVENT TIME_WAITED --------------------------------- ----------- log file parallel write 159692 db file sequential read 28657 smon timer 130673837 SQL*Net message from client 16528989

The v$session_event View Shows one row for each wait event name within each session, along with cumulative statistics since session start. Column Name Data Type -------------------------- ------------ SID NUMBER EVENT VARCHAR2(64) TOTAL_WAITS NUMBER TOTAL_TIMEOUTS NUMBER TIME_WAITED NUMBER AVERAGE_WAIT NUMBER MAX_WAIT NUMBER TIME_WAITED_MICRO NUMBER

Columns In v$session_event SID: The ID of a session (from v$session) EVENT: The name of a wait event TOTAL_WAITS: Total number of times this session has waited for this event TOTAL_TIMEOUTS: Total number of timeouts while this session has waited for this event TIME_WAITED: Total time waited for this event by this session (in centiseconds) AVERAGE_WAIT: The average length of a wait for this event in this session (in centiseconds) MAX_WAIT: The maximum amount of time the session had to wait for this event (in centiseconds)

Sample v$session_event Query SQL> SELECT event, total_waits, time_waited_micro 2 FROM v$session_event 3 WHERE SID = 4 (SELECT sid FROM v$session 5 WHERE audsid = 6 USERENV ('sessionid') );   EVENT WAITS TIME_WAITED_MICRO --------------------------- ----- ----------------- db file sequential read 552 2409173 db file scattered read 41 315928 SQL*Net message to client 73 347 SQL*Net message from client 72 3397382712

Oracle 9i Bug #2429929 SQL> SELECT event, total_waits, time_waited_micro 2 FROM v$session_event 3 WHERE SID + 1 = 4 (SELECT sid FROM v$session 5 WHERE audsid = 6 USERENV ('sessionid') );   EVENT WAITS TIME_WAITED_MICRO --------------------------- ----- ----------------- db file sequential read 552 2409173 db file scattered read 41 315928 SQL*Net message to client 73 347 SQL*Net message from client 72 3397382712

The v$event_name View Shows one row for each wait event name known to the Oracle kernel, along with names of up to three parameters associated with the wait event. Column Name Data Type -------------------------- ------------ EVENT# NUMBER NAME VARCHAR2(64) PARAMETER1 VARCHAR2(64) PARAMETER2 VARCHAR2(64) PARAMETER3 VARCHAR2(64)

Columns In v$event_name EVENT#: An internal ID NAME: The name of a wait event PARAMETERn: The name of a parameter associated with the wait event

Sample v$event_name Query SQL> SELECT * 2 FROM v$event_name 3 WHERE name = 'db file scattered read';   EVENT# NAME ---------- ------------------------------ PARAMETER1 PARAMETER2 PARAMETER3 ------------- ------------- ------------- 95 db file scattered read file# block# blocks

The v$session_wait View Shows one row for each session, providing detailed information about the current or most recent wait event. Column Name Data Type -------------------------- ------------ SID NUMBER SEQ# NUMBER EVENT VARCHAR2(64) P1TEXT VARCHAR2(64) P1 NUMBER P1RAW RAW(4) P2TEXT VARCHAR2(64) P2 NUMBER P2RAW RAW(4) P3TEXT VARCHAR2(64) P3 NUMBER P3RAW RAW(4) WAIT_TIME NUMBER SECONDS_IN_WAIT NUMBER STATE VARCHAR2(19)

Columns In v$session_wait SID: The ID of a session SEQ#: A number that increments by one on each new wait STATE: An indicator of the session status: ‘WAITING’: The session is currently waiting, and details of the wait event are provided. ‘WAITED KNOWN TIME’: The session is not waiting, but information about the most recent wait is provided. ‘WAITED SHORT TIME’ or ‘WAITED UNKNOWN TIME’: The session is not waiting, but partial information about the most recent wait is provided.

Columns In v$session_wait (cont.) EVENT: The name of a wait event PnTEXT: The name of a parameter associated with the wait event Pn: The value of the parameter in decimal form PnRAW: The value of the parameter in raw form WAIT_TIME: Length of most recent wait (in centiseconds) if STATE = ‘WAITED KNOWN TIME’ SECONDS_IN_WAIT: How long current wait has been so far if STATE = ‘WAITING’

Sample v$session_wait Query SQL> SELECT * FROM v$session_wait WHERE sid = 16;   SID SEQ# EVENT ---- ----- ------------------------------ P1TEXT P1 P1RAW P2TEXT P2 P2RAW ------ ---- -------- ------ ---- -------- P3TEXT P3 P3RAW WAIT_TIME SECONDS_IN_WAIT ------ ---- -------- --------- --------------- STATE ------------------- 16 303 db file scattered read file# 17 00000011 block# 2721 00000AA1 blocks 8 00000008 -1 0 WAITED SHORT TIME

Tracing Wait Event Activity Using the dbms_support package or setting debug event 10046 enables SQL trace, and can optionally include wait event information and bind variable data in trace files as well. Methods for setting debug events: ALTER SESSION SET events oradebug dbms_system.set_ev

Activating Wait Event Tracing dbms_support is missing from many releases of Oracle 8i, but is available as a patch. dbms_support is not installed by default; run dbmssupp.sql in ?/rdbms/admin to install it. dbms_system.set_ev is not supported by Oracle Corporation because it lets you set any debug event and some can put your database at risk. Tracing imposes serious system overhead, so trace only what you need.

Debug Event 10046 Settings Value of N Effect 1 ALTER SESSION SET events '10046 trace name context forever, level N'; Value of N Effect 1 Enables ordinary SQL trace 4 Enables SQL trace with bind variable values included in trace file 8 Enables SQL trace with wait event information included in trace file 12 Equivalent of level 4 and level 8 together

Sample Oracle 8i Trace Output ===================== PARSING IN CURSOR #1 len=80 dep=0 uid=502 oct=3 lid=502 tim=2293771931 hv=2293373707 ad='511dca20' SELECT /*+ FULL */ SUM (LENGTH(notes)) FROM customer_calls WHERE status = :x END OF STMT PARSE #1:c=0,e=0,p=0,cr=0,cu=0,mis=1,r=0,dep=0,og=0,tim=2293771931 BINDS #1: bind 0: dty=2 mxl=22(22) mal=00 scl=00 pre=00 oacflg=03 oacfl2=0 size=24 offset=0 bfp=09717724 bln=22 avl=02 flg=05 value=43 EXEC #1:c=0,e=0,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=4,tim=2293771931 WAIT #1: nam='SQL*Net message to client' ela= 0 p1=675562835 p2=1 p3=0 WAIT #1: nam='db file scattered read' ela= 3 p1=17 p2=923 p3=8 WAIT #1: nam='db file scattered read' ela= 1 p1=17 p2=931 p3=8 WAIT #1: nam='db file scattered read' ela= 2 p1=17 p2=939 p3=8 WAIT #1: nam='db file sequential read' ela= 0 p1=17 p2=947 p3=1 WAIT #1: nam='db file scattered read' ela= 3 p1=17 p2=1657 p3=8

Wait Event Tracing Enhancements In Oracle 9i The dbms_support package is provided for easier trace activation. Elapsed times in the trace file are shown in microseconds instead of centiseconds. A “waits=yes” option has been added to TKPROF to include wait event statistics in the TKPROF report.

Using Wait Event Information Five examples of how wait event information was used to diagnose production problems

Example #1: Buffer Busy Waits A magazine publisher has a website that displays content stored in a database. At times the website would get bogged down—response time would become poor and the database server would become extremely busy (near-zero idle time).

Viewing Wait Events Statistics With Statspack Collect Statspack snapshots at regular intervals. Statspack report shows top wait events for entire instance during snapshot interval. Oracle 9i Statspack also shows CPU time used during the interval.

Statspack Report Output Snap Id Snap Time Sessions ------- ------------------ -------- Begin Snap: 61 11-Dec-02 13:00:52 145 End Snap: 71 11-Dec-02 14:00:26 145 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits Time (cs) Wt Time ------------------------ ---------- ---------- ------- buffer busy waits 1,962,372 1,278,649 50.03 db file sequential read 1,336,870 1,050,878 41.12 db file scattered read 47,717 49,326 1.93 direct path write 8,070 40,574 1.59 latch free 38,220 31,012 1.21

What We See in the Statspack Report Dominant wait events: buffer busy waits db file sequential read Over 23,000 seconds of wait time on these two events in a one hour period (over 6 seconds of waiting per elapsed second)

Understanding the Buffer Busy Waits Event SQL> SELECT parameter1, parameter2, parameter3 2 FROM v$event_name 3 WHERE name = 'buffer busy waits';   PARAMETER1 PARAMETER2 PARAMETER3 ------------ ------------ ------------ file# block# id file#: Data file containing the desired data block block#: Block within the data file that is desired id: Reason the buffer in the buffer cache is busy (see Metalink bulletin #34405.1)

Finding Which Data Blocks Are Experiencing Buffer Contention SQL> SELECT sid, event, state, seconds_in_wait, 2 wait_time, p1, p2, p3 3 FROM v$session_wait 4 WHERE event = 'buffer busy waits' 5 ORDER BY sid;   SID EVENT STATE SEC TIME P1 P2 P3 --- ----------------- ----- --- ---- ----- ----- ----- 12 buffer busy waits WAITE 1 0 30 62157 130 31 buffer busy waits WAITE 1 0 30 23558 130

Finding Which Data Blocks Are Experiencing Buffer Contention SQL> SELECT owner, segment_name, segment_type 2 FROM dba_extents 3 WHERE file_id = &absolute_file_number 4 AND &block_number BETWEEN block_id 5 AND block_id + blocks -1; Enter value for absolute_file_number: 30 Enter value for block_number: 62157 OWNER SEGMENT_NAME SEGMENT_TYPE ----------------- ------------------- ------------ PRODMGR SAMPLES TABLE

Reason Codes from Metalink Bulletin #34405.1

What We Have Learned So Far A buffer containing a data block of the SAMPLES table is experiencing contention. The buffer in the buffer cache is busy because another session is reading the same data block from disk.

Understanding the DB File Sequential Read Event SQL> SELECT parameter1, parameter2, parameter3 2 FROM v$event_name 3 WHERE name = 'db file sequential read';   PARAMETER1 PARAMETER2 PARAMETER3 ------------ ------------ ------------ file# block# blocks file#: Data file containing the desired data block block#: Block within the data file that is desired blocks: How many blocks are being read (typically 1 for db file sequential read)

Finding Which Data Blocks Are Being Read SQL> SELECT sid, event, state, seconds_in_wait, 2 wait_time, p1, p2, p3 3 FROM v$session_wait 4 WHERE event = 'db file sequential read' 5 ORDER BY sid; SID EVENT STATE SEC TIME P1 P2 P3 --- ----------------- ----- --- ---- ----- ----- ----- 17 db file sequentia WAITE 1 0 30 62042 1 19 db file sequentia WAITE 1 0 30 61731 1 33 db file sequentia WAITI 0 0 30 57292 1

Finding Which Data Blocks Are Being Read SQL> SELECT owner, segment_name, segment_type 2 FROM dba_extents 3 WHERE file_id = &absolute_file_number 4 AND &block_number BETWEEN block_id 5 AND block_id + blocks -1; Enter value for absolute_file_number: 30 Enter value for block_number: 62042 OWNER SEGMENT_NAME SEGMENT_TYPE ----------------- ------------------- ------------ PRODMGR SAMPLES TABLE

The SAMPLES Table Contained a LONG column with very large values Excessive row chaining Most queries did not retrieve the LONG data Table assigned to KEEP pool, but too large to fit entirely in memory

Long-Term Problem Resolution Convert the LONG column to a CLOB. Large CLOB data will be stored in a separate LOB segment. Row chaining will be reduced or eliminated. The table segment will be much smaller and more likely to fit in memory.

Short-Term Problem Resolution Added index on most columns of SAMPLES table Allowed most queries to avoid table segment Enlarged KEEP pool Allowed index segment to fit in memory

Statspack Report Output Snap Id Snap Time Sessions ------- ------------------ -------- Begin Snap: 1192 20-Dec-02 13:00:49 102 End Snap: 1202 20-Dec-02 14:00:18 102 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits Time (cs) Wt Time ------------------------ ---------- ---------- ------- direct path write 6,467 13,545 30.61 log file sync 4,914 7,493 16.93 library cache pin 1,175 6,090 13.76 direct path read 5,488 3,428 7.75 latch free 14,528 2,931 6.62

What We See in the Statspack Report Now No db file sequential read or buffer busy waits. All data was already in the buffer cache. Physical reads reduced by over 90%. Total wait time on all non-idle events reduced by over 98%. Before: 12,786.49 / 0.5003 = 25,557.65 After: 135.45 / 0.3061 = 442.50

What We Learned from Wait Event Information Large amounts of time were being spent waiting on single block disk reads and buffer contention in the buffer cache. Random samples showed the disk reads and contention involved the SAMPLES table. The buffer contention was the result of multiple sessions needing the same block from disk. Wait events pointed us directly to the problem segment.

Example #2: More Buffer Busy Waits, Plus Latch Contention A genetic research company stored their data in Oracle. Applications running concurrently on many workstations would fetch raw data, process it, and put the data back in the database. But throughput bogged down as they added more workstations.

Activating Wait Event Tracing Added to application code on workstation #30: ALTER SESSION SET events '10046 trace name context forever, level 8'; Could have used dbms_support if it was installed: dbms_support.start_trace; Modified application code to exit after 500 iterations

TKPROF Wait Events Reporting in Oracle 9i tkprof prodgen_ora_16466.trc report_16466.prf waits=yes

TKPROF Report Output UPDATE processing_stations SET status = 'ACTIVE', status_date = SYSDATE, data_set_id_being_processed = :b1 WHERE station_id = 30 call count cpu elapsed disk query current rows ------- ------ ------- --------- ----- ----- ------- ----- Parse 1 0.01 0.00 0 0 0 0 Execute 500 0.23 10.14 0 3616 1010 500 Fetch 0 0.00 0.00 0 0 0 0 total 501 0.24 10.14 0 3616 1010 500 Elapsed times include waiting on following events: Event waited on Times Max. Wait Total Waited --------------------------- Waited ---------- ------------ buffer busy waits 26 0.71 7.87 latch free 17 0.57 2.08 log file switch completion 3 0.09 0.20

What We See In the TKPROF Report 500 trivial updates took 10.14 seconds Most of that time was spent waiting Dominant wait events: buffer busy waits latch free CPU time plus wait time does not add up to elapsed time due to round-off errors  

Waits In the Trace File Elapsed times are in microseconds in Oracle 9i WAIT #2: nam='buffer busy waits' ela= 527727 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 498765 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 137611 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 124165 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 5237 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 264050 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 270177 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 330912 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 156317 p1=18 p2=10 p3=220 WAIT #2: nam='buffer busy waits' ela= 710696 p1=18 p2=10 p3=220 Elapsed times are in microseconds in Oracle 9i

Finding Which Data Blocks Are Experiencing Buffer Contention SQL> SELECT owner, segment_name, segment_type 2 FROM dba_extents 3 WHERE file_id = &absolute_file_number 4 AND &block_number BETWEEN block_id 5 AND block_id + blocks -1; Enter value for absolute_file_number: 18 Enter value for block_number: 10 OWNER SEGMENT_NAME SEGMENT_TYPE ----------------- ------------------- ------------ GEN PROCESSING_STATIONS TABLE

Reason Codes from Metalink Bulletin #34405.1

What We Have Learned So Far A buffer containing a data block of the PROCESSING_STATIONS table is experiencing contention. The buffer in the buffer cache is busy because another session has the buffer in an incompatible mode. All 26 buffer busy waits totaling 7.87 seconds involved the same data block.

The PROCESSING_STATIONS Table SQL> SELECT SYSDATE - last_analyzed, blocks, 2 avg_row_len, avg_space, num_rows 3 FROM user_tables 4 WHERE table_name = 'PROCESSING_STATIONS'; SYSDATE- AVG_ AVG_ LAST_ANALYZED BLOCKS ROW_LEN SPACE NUM_ROWS ------------- ------ ------- ----- -------- 2.132118056 1 62 1686 100

Two Important Observations There were 100 workstations running the processing application concurrently. The trace we ran on workstation #30 completed in just under one minute.

Lots of Updates! Workstation #30 updated the PROCESSING_STATIONS table 500 times in less than one minute. If all 100 workstations do similar things: More than 50,000 updates to one data block every minute by 100 concurrent sessions!

Why So Many Updates? Workstations use the PROCESSING_STATIONS table to track which workstation is processing which data set. Processing one data set takes between 0.1 second and 20 minutes. Workstations update the table frequently to keep the timestamp current. This would be helpful in the event of a workstation crash.

Long-Term Problem Resolution Modify the application to update the PROCESSING_STATIONS table less frequently—once per data set or once per second for larger data sets: Will reduce updates by over 80% Buffer busy waits will disappear or dramatically decrease

Short-Term Problem Resolution Rebuilt the PROCESSING_STATIONS table with PCTFREE set to 99: Oracle reserved 99% of each data block for future row expansion. Each row got its own data block. Each workstation session now updates a separate data block.

The Rebuilt PROCESSING_STATIONS Table SQL> SELECT SYSDATE - last_analyzed, blocks, 2 avg_row_len, avg_space, num_rows 3 FROM user_tables 4 WHERE table_name = 'PROCESSING_STATIONS'; SYSDATE- AVG_ AVG_ LAST_ANALYZED BLOCKS ROW_LEN SPACE NUM_ROWS ------------- ------ ------- ----- -------- .130868056 100 62 8014 100

TKPROF Report Output UPDATE processing_stations SET status = 'ACTIVE', status_date = SYSDATE, data_set_id_being_processed = :b1 WHERE station_id = 30 call count cpu elapsed disk query current rows ------- ------ ------- --------- ----- ----- ------- ----- Parse 1 0.00 0.00 0 0 0 0 Execute 500 0.20 2.22 0 500 1009 500 Fetch 0 0.00 0.00 0 0 0 0 total 501 0.20 2.22 0 500 1009 500 Elapsed times include waiting on following events: Event waited on Times Max. Wait Total Waited --------------------------- Waited ---------- ------------ latch free 2 0.35 0.61

What We See in the TKPROF Report Now 500 updates took 2.22 seconds, down from 10.14 seconds No more buffer busy waits Waited 0.61 seconds on latches, down from 2.08 seconds CPU time was 0.20 seconds, down from 0.23 seconds 1.41 seconds unaccounted for—likely a mix of waiting for CPU and round-off error

Understanding the Latch Free Event SQL> SELECT parameter1, parameter2, parameter3 2 FROM v$event_name 3 WHERE name = 'latch free';   PARAMETER1 PARAMETER2 PARAMETER3 ------------ ------------ ------------ address number tries         address: Join to addr in v$latch       number: Join to latch# in v$latchname tries: Number of times the session has waited while trying to acquire the latch

Waits In the Trace File WAIT #2: nam='latch free' ela= 47004 p1=15113593728 p2=97 p3=0 WAIT #2: nam='latch free' ela= 14629 p1=15113593728 p2=97 p3=1 WAIT #2: nam='latch free' ela= 20652 p1=15113593728 p2=97 p3=2 WAIT #2: nam='latch free' ela= 37737 p1=15113593728 p2=97 p3=3   Four consecutive waits for one acquisition of the latch

Finding Which Latches Are Experiencing Contention SQL> SELECT latch#, name 2 FROM v$latchname 3 WHERE latch# = &latch_number;   Enter value for latch_number: 97 LATCH# NAME ---------- -------------------- 97 cache buffers chains

What We Learned from Wait Event Information Much time was spent waiting on latch contention and buffer contention in the buffer cache. The buffer contention was all for one data block. The buffer contention was the result of multiple sessions needing to update the same data block. The latch contention involved the latch that protects the buffer cache chains data structure. Wait events pointed us directly to the hot buffer in the buffer cache.

Example #3: Log File Waits A data warehouse loader application was tuned in a test environment until it met user acceptance. The production server was larger and more powerful, but the data loads actually took longer in production than in the test environment.

Summarizing Wait Events During A Period of Time v$system_event shows wait event totals since instance startup. v$session_event shows wait event totals since the beginning of a session. You can capture view contents at different points in time and compute the delta in order to get wait event information for a specific period of time. Statspack and many third-party tools can do this, but a simple script has less overhead and can be quicker to deploy.

Simple Script to See Wait Events During a 30 Second Period CREATE TABLE previous_events AS SELECT SYSDATE timestamp, v$system_event.* FROM v$system_event; EXECUTE dbms_lock.sleep (30); SELECT A.event, A.total_waits - NVL (B.total_waits, 0) total_waits, A.time_waited - NVL (B.time_waited, 0) time_waited FROM v$system_event A, previous_events B WHERE A.event NOT IN (list of idle events) AND B.event (+) = A.event ORDER BY time_waited;

Wait Events During 30 Seconds of Data Loading EVENT TOTAL_WAITS TIME_WAITED ---------------------------- ----------- ----------- control file sequential read 61 1 latch free 2 1 db file sequential read 6 7 control file parallel write 41 31 log file single write 6 164 db file parallel write 13 220 enqueue 6 486 log buffer space 24 2007 log file sequential read 30 2655 log file switch completion 33 2883 log file parallel write 19 3561 log file sync 113 10249

What We See in the Script Output Over 215 seconds spent waiting on log-related events: Sessions waited 102 seconds for LGWR to flush the log buffer to disk for a commit Sessions waited 48 seconds for LGWR to make space for more redo LGWR waited 37 seconds for disk writes ARCH waited 26 seconds for disk reads

Investigate the Redo Log Check production online redo log location for contention and slow hardware: All log files located on one disk Same disk held hot files for another database Compare to test environment: Log files located on a striped volume Minimal other activity on the volume Database in NOARCHIVELOG mode

Problem Resolution Sped up online redo log file performance: Moved log files to dedicated disks Spread log files over multiple disks

What We Learned from Wait Event Information The disk I/O speed writing and reading the online redo log files was the bottleneck slowing down the data warehouse load. Wait events pointed us directly to the area within Oracle that was holding up the works.

Example #4: Direct Path I/O Waits Analysts in a customer service unit were satisfied with the response time when they queried individual customer orders from their data warehouse. However, queries involving summarizations of multiple orders were unacceptably slow.

Database Rx Wait Event Report

What We See in the Database Rx Report Dominant wait events: direct path write db file scattered read direct path read Above account for 99% of non-idle event wait time Insignificant db file sequential read waits

What We Have Learned So Far Large amount of direct path I/O activity Usually involves temporary segments Significant multi-block I/O reads Full table scans are common in a data warehouse environment Insignificant single-block I/O reads Frequently accessed data blocks probably in buffer cache

Understanding the Direct Path I/O Events SQL> SELECT name, parameter1, parameter2, parameter3 2 FROM v$event_name 3 WHERE name LIKE 'direct path%'; NAME PARAMETER1 PARAMETER2 PARAMETER3 ----------------- ----------- ---------- ---------- direct path read file number first dba block cnt direct path write file number first dba block cnt file number: File containing data block first dba: First block within the file to be accessed block cnt: Number of blocks to be accessed

Finding Which Files Are Being Accessed SQL> SELECT sid, event, state, seconds_in_wait, 2 wait_time, p1, p2, p3 3 FROM v$session_wait 4 WHERE event = 'direct path write' 5 ORDER BY sid; SID EVENT STATE SEC TIME P1 P2 P3 --- ----------------- ----- --- ---- ----- ----- ----- 39 direct path write WAITI 0 0 201 65 7 47 direct path write WAITI 0 0 201 2248 7

Finding Which Files Are Being Accessed SQL> SELECT tablespace_name, file_id "AFN" 2 FROM dba_data_files 3 WHERE file_id = 201; no rows selected SQL> SELECT tablespace_name, file_id + value "AFN" 2 FROM dba_temp_files, v$parameter 3 WHERE name = 'db_files' 4 AND file_id + value = 201; TABLESPACE_NAME AFN ------------------------------ ---------- TEMP 201

Problem Resolution Increased sort_area_size: Was set to default of 65536 Increased to 10485760 (few concurrent sessions) If that had not solved the problem: Tune application code to reduce sorting Check for other active files on disks holding temp files Move temp files to a striped volume

What We Learned from Wait Event Information Direct path I/O accounted for 75% of the non-idle event wait time on the system. Multi-block reads accounted for 24% of the non-idle event wait time—not unusual in a data warehouse environment. Random samples showed direct path I/O involved the temporary tablespace. Wait events pointed us directly to the area within Oracle that needed adjustment.

Logical vs. Physical Reads

Logical vs. Physical Reads During the Database Rx sample interval there were more physical reads than logical reads. Direct path reads count as physical reads but not logical reads. Be careful how you compute your buffer cache hit ratios—in this example you might come up with a negative figure!

Example #5: Database Link Wait Events A company had five Oracle databases, one per region. Due to human error, the same customer transactions would sometimes get loaded into multiple databases. A report was built to identify these duplicates, but it took 30 minutes to run.

Isolating a Query and Analyzing Its Wait Events Start a new database session in SQL*Plus or a similar tool. Run the query. Monitor the session’s wait events and statistics from another session: v$session_event v$sesstat This is a handy technique when you know which statement is the bottleneck.

Query Output from v$session_event SQL> SELECT event, total_waits, time_waited, max_wait 2 FROM v$session_event 3 WHERE sid = 47 4 ORDER BY event; EVENT TOTAL_WAITS TIME_WAITED MAX_WAIT --------------------------- ----------- ----------- ---------- SQL*Net message from client 32 4435 2432 SQL*Net message from dblink 1525516 104919 31 SQL*Net message to client 33 0 0 SQL*Net message to dblink 1525516 466 9 db file sequential read 27199 8025 28 latch free 40 5 4 log file sync 1 2 2

Query Output from v$sesstat SQL> SELECT A.name, B.value 2 FROM v$statname A, v$sesstat B 3 WHERE A.statistic# = 12 4 AND B.statistic# = A.statistic# 5 AND B.sid = 47; NAME VALUE ------------------------------ ---------- CPU used by this session 67937

What We See In the v$ Data 1.5 million waits on network roundtrips through a database link: 1053 seconds Network latency Time for the remote database to respond to each request 27,000 waits for single-block disk reads: 80 seconds

The Query We Are Studying SELECT customer_id, batch_serial_number, batch_date, load_date, batch_comment, control_total FROM customer_xfer_batches A WHERE exists (SELECT 1 FROM customer_xfer_batches@prdwest B WHERE B.customer_id = A.customer_id AND B.batch_serial_number = A.batch_serial_number) ORDER BY customer_id, batch_serial_number;

The Query We Are Studying Execution Plan ---------------------------------------------------------- 0 SELECT STATEMENT 1 0 FILTER 2 1 TABLE ACCESS (BY INDEX ROWID) OF 'CUSTOMER_XFER_BATCHES' 3 2 INDEX (FULL SCAN) OF 'CUST_XFER_BAT_PK' (UNIQUE) 4 1 REMOTE* PRDWEST 4 SERIAL_FROM_REMOTE SELECT "CUSTOMER_ID","BATCH_SERIAL_NUMBER" FROM "CUSTOMER_XFER_BATCHES" "B" WHERE "BATCH_SERIAL_NUMBER"=:1 AND "CUSTOMER_ID"=:2

CUSTOMER_XFER_BATCHES SQL> SELECT blocks, num_rows 2 FROM user_tables 3 WHERE table_name = 4 'CUSTOMER_XFER_BATCHES'; BLOCKS NUM_ROWS ------ -------- 21825 1526003

What We Have Learned So Far Oracle is doing a full scan of the index on the local table and fetching each row one at a time This does avoid a sort Very high price to pay to skip sorting a few rows Oracle is doing one remote query for each row fetched from the local table

Problem Resolution - Part 1 SELECT customer_id, batch_serial_number, batch_date, load_date, batch_comment, control_total FROM customer_xfer_batches WHERE (customer_id, batch_serial_number) IN (SELECT customer_id, batch_serial_number FROM customer_xfer_batches INTERSECT SELECT customer_id, batch_serial_number FROM customer_xfer_batches@prdwest) ORDER BY customer_id, batch_serial_number;

Query Output from v$session_event SQL> SELECT event, total_waits, time_waited, max_wait 2 FROM v$session_event 3 WHERE sid = 49 4 ORDER BY event; EVENT TOTAL_WAITS TIME_WAITED MAX_WAIT --------------------------- ----------- ----------- ---------- SQL*Net message from client 46 3680 2481 SQL*Net message from dblink 24 31 18 SQL*Net message to client 47 0 0 SQL*Net message to dblink 24 0 0 SQL*Net more data from dbli 5978 1337 13 db file scattered read 3430 675 8 db file sequential read 182 60 2 direct path read 148 233 11 direct path write 920 3572 33

Query Output from v$sesstat SQL> SELECT A.name, B.value 2 FROM v$statname A, v$sesstat B 3 WHERE A.statistic# = 12 4 AND B.statistic# = A.statistic# 5 AND B.sid = 49; NAME VALUE ------------------------------ ---------- CPU used by this session 3227

What We See in the v$ Data Now 24 network roundtrips through a database link instead of 1.5 million: 14 seconds (down from 1053) Fewer, larger network packets Fewer requests to remote database 3,600 waits on mostly multi-block disk reads instead of 27,000 waits on single-block disk reads: 7 seconds (down from 80) Fewer multi-block reads instead of many single-block reads

What We See in the v$ Data Now 1100 waits on direct path I/O: 38 seconds (new) Sorting to implement the INTERSECT operation 32 seconds of CPU time (down from 679) Fewer logical reads and network roundtrips Elapsed time: 92 seconds (down from 31 minutes)

Iterative Tuning Curing one bottleneck often reveals or creates another, smaller bottleneck. Repeat the wait event evaluation process after each change until performance goals are met. In this situation, a 95% reduction in runtime from 31 minutes to 92 seconds still did not meet the performance goal.

What We Have So Far Rewritten query completes in 92 seconds: 32 CPU seconds 38 seconds of wait on direct path I/O 14 seconds of wait on network roundtrips 7 seconds of wait on multi-block and single-block reads

Problem Resolution - Part 2 Eliminating or speeding up direct path I/O seems like the logical next step: sort_area_size set to 1 Mb Try dynamically changing it to 100 Mb?

Query Output from v$session_event SQL> SELECT event, total_waits, time_waited, max_wait 2 FROM v$session_event 3 WHERE sid = 46 4 ORDER BY event; EVENT TOTAL_WAITS TIME_WAITED MAX_WAIT --------------------------- ----------- ----------- ---------- SQL*Net message from client 47 442 287 SQL*Net message from dblink 25 25 14 SQL*Net message to client 48 0 0 SQL*Net message to dblink 25 0 0 SQL*Net more data from dbli 6050 1378 26 db file scattered read 3430 945 8 db file sequential read 191 59 1 log file sync 1 3 3

Query Output from v$sesstat SQL> SELECT A.name, B.value 2 FROM v$statname A, v$sesstat B 3 WHERE A.statistic# = 12 4 AND B.statistic# = A.statistic# 5 AND B.sid = 46; NAME VALUE ------------------------------ ---------- CPU used by this session 3296

What We See in the v$ Data Now Waits on network roundtrips through a database link, multi-block reads, and single-block reads unchanged CPU time used unchanged Direct path I/O waits eliminated completely Entire sort now performed in memory Elapsed time: 55 seconds (down from 92)

What We Learned from Wait Event Information A query ran slowly due to excessive network roundtrips and single-block reads. After these problems were corrected, 40% of the query execution time was devoted to sorting to disk. Wait events showed us how Oracle was spending its time while executing the query, helping us improve the query’s performance in an iterative fashion.

A Summary Of Wait Event Techniques Using Statspack snapshots and reports to analyze wait events at the instance level Polling v$session_wait to determine which buffers or latches have contention Enabling wait event tracing in a session Using Oracle 9i TKPROF to tabulate waits at the statement level within one session

A Summary Of Wait Event Techniques (continued) Collecting wait event data for a session or the entire instance at two different times and computing the difference to find the wait events during a specific period of time Ranking cumulative wait event data in order to see which wait events account for the most wait time Isolating a statement and analyzing its wait events

Send Us Your Wait Event Puzzles We are always looking for interesting wait event situations to learn from! If you are trying to diagnose a problem using the wait event interface, feel free to email us wait events data and a problem description. We’ll do our best to look over what you send us and share our thoughts with you.

The White Paper A companion white paper to this presentation is available for free download from our company’s website at: www.dbspecialists.com/presentations.html

Resources from Database Specialists The Specialist newsletter www.dbspecialists.com/specialist.html Database Rx® dbrx.dbspecialists.com/guest Provides secure, automated monitoring, alert notification, and analysis of your Oracle databases

In Conclusion The wait event interface gives you access to a detailed accounting of how Oracle processes spend their time. Wait events touch all aspects of the Oracle database server. The wait event interface will not always give you the answer to every performance problem, but it will just about always give you insights that guide you down the proper path to problem resolution.

Contact Information Roger Schrag rschrag@dbspecialists.com Terry Sutton tsutton@dbspecialists.com Database Specialists, Inc. 388 Market Street, Suite 400 San Francisco, CA 94111 Tel: 415/344-0500 Web: www.dbspecialists.com