1. AOBD 07/08H. Galhardas DBMS Performance Monitoring

2. Performance Monitoring Goals Monitoring should check that the performance- influencing database parameters are correctly set and if they are not, it should point to where the problems are Alternatively, you can sit tight and wait for the inevitable angry users’ complaints OR Monitoring (preventive method) Correct problems when detected (reactive method)

3. How does one monitor a DBMS? By extracting relevant performance indicators, (counters, gauges and details of internal DBMS’s activities) By comparing the obtained values of these indicators against ideal values But… there are many indicators!

4. A Consumer-Producer Chain of a DBMS’s Resources High Level Consumers Intermediate Resources/ Consumers Primary Resources PARSER OPTIMIZER EXECUTION SUBSYSTEM DISK SYBSYSTEM CACHE MANAGER LOGGING SUBSYSTEM LOCKING SUBSYSTEM NETWORK DISK/ CONTROLLER CPUMEMORY sqlcommands probing spots for indicators

5. Recurrent Patterns of Problems An overloading high-level consumer  Ex: query accessing too many rows A poorly parameterized subsystem  Ex: disk subsystem that stores tables, indexes, logging in a single disk An overloaded primary resource  Ex: CPU busy because non-DB processes are using it at the same time Effects are not always felt first where the cause is!

6. A Systematic Approach to Monitoring Question 1: Are critical queries being served in the most efficient manner? Question 2: Are DBMS subsystems making optimal use of resources? Question 3: Are there enough primary resources available and are they configured adequately, given the current and expected workload? Extract indicators to answer the following questions

7. Investigating High Level Consumers: Critical query monitoring Find critical queries Found any? Investigate lower levels Answer Q1 over them Overcon- sumption? Tune problematic queries yes no

8. Investigating Intermediate and Primary Resources: routine monitoring Answer Q3 Problems at OS level? Answer Q2 Tune low-level resources yesno Problematic subsystems? Tune subsystems Investigate upper level yesno

9. Tools used to gather information Event monitors Query plan explainers Performance monitors

10. Investigating High Level Consumers Answer question 1: “Are critical queries being served in the most efficient manner?” 1. Identify the critical queries 2. Analyze their access plans 3. Profile their execution

11. Identifying Critical Queries Critical queries are usually those that:  Take a long time Consume a great amount of resources (read a lot of data, perform heavy aggregation or sorting, lock an entire table, etc)  Are frequently executed  Have to be answered in a short time Often, a user complaint will tip us off.

12. Event Monitors to Identify Critical Queries If no user complains... Record DBMS’s performance measurements, but only when a system event happens  Ex: new user connection, or beginning/end of the execution of an SQL statement Logs the event time, duration and associated performance indicator values Less overhead than other type of tools because indicators are usually by-product of operations monitored and are accessed in a convenient time  Particularly appropriate for capturing extraordinary conditions, ex: deadlocks. Typical measures include CPU used, IO used, locks obtained etc.

13. What should be measured and evaluated Interested in the end-of-statement or end-of- transaction event  Carry relevant info that indicate the resources the executed query consumed: SQL of the statement, duration of execution, CPU, IO, cache,lock consumption statistics  Sort the collected data over a given resource consumption statistic and we can find the most expensive queries by that criterion

14. An example Event Monitor Several CPU indicators sorted by Oracle’s Trace Data Viewer Similar tools: DB2’s Event Monitor and MSSQL’s Server Profiler

15. What to do in case of problems? Make sure expensive queries are tuned in the best possible way May have to drill down through the consumption chain to search for the real cause of the problem

16. Investigating High Level Consumers Answer question 1: “Are critical queries being served in the most efficient manner?” 1. Identify the critical queries 2. Analyze their access plans 3. Profile their execution

17. Diagnose Expensive Queries: analyzing access plans SQL commands are translated into an internal executable format before they are processed After parsing, the optimizer enumerates and estimates the cost of a number of possible ways to execute that query The best choice, according to the existing statistics, is chosen  But this choice may not be the best one! sql parserewrite enumerate/ cost plans generate chosen plan plan PARSER/OPTIMIZER

18. Query Plan Explainers to Analyze Plans Explainers usually depict an access plan as a (graphical) single-rooted tree in which sources at the leaf end are tables or indexes, and internal nodes are operators These form an assembly line (actually tree) of tuples! Most explainers will let you see the estimated costs for CPU consumption, I/O, and cardinality of each operator. If you see something strange, change an index.

19. An example Plan Explainer Access plan according to MSSQL’s Query Analyzer Similar tools: DB2’s Visual Explain and Oracle’s SQL Analyze Tool

20. Finding Strangeness in Access Plans What to pay attention to on a plan  Access paths for each table For every table involved in the query, determine how it is accessed: scan, index?  Sorts or intermediary results Check if the plan includes a sorting (because of DISTINCT, ORDER BY, GROUP BY) and see if it can be eliminated Materialization of temporary results should be avoided  Order of operations May want to force an ordering if we otice the order does not favor the early reduction in the number of tuples  Algorithms used in the operators

21. To Index or not to index? select c_name, n_name from CUSTOMER join NATION on c_nationkey=n_nationkey where c_acctbal > 0 Which plan performs best? (nation_pk is an non-clustered index over n_nationkey, and similarly for acctbal_ix over c_acctbal)

22. Non-clustering indexes can be trouble Each access to the index generates a random access to the table – possibly duplicate! If !36,308 out of 150,000 customers have positive balance, it ends up that the number of pages read from the table is greater than its size, i.e., a table scan is way better Table ScanIndex Scan 5 sec 143,075 pages 6,777 pages 136,319 pages 7 pages 76 sec 272,618 pages 131,425 pages 273,173 pages 552 pages CPU time data logical reads data physical reads index logical reads index physical reads

23. Investigating High Level Consumers Answer question 1: “Are critical queries being served in the most efficient manner?” 1. Identify the critical queries 2. Analyze their access plans 3. Profile their execution

24. Profiling a Query’s Execution A query was found critical but its plan looks okay. What’s going on? Put it to run. Profiling it would determine the quantity of the resources used by a query and assessing how efficient this use was  Profile of a query execution consists of detailed information about the duration and resource consumption Resources  DBMS subsystems: cache, disk, lock, log  OS raw resources: CPU

25. Query profiling indicators Duration information involves 3 indicators:  Elapsed time for the query: time it took to process it as perceived by the user  CPU time: time the CPU was actually used to process the query  Wait time: time the query was not processing and waiting for a resource to become available Resource consumption information includes:  (I/O) Physical and logical reads/writes  (Locking) Maximum nb locks held, nb lock escalations,nb deadlocks/timeouts, total time spent waiting for locks  (SQL activity) Nb sorts and temporary area usage: gives an idea of the time spent in expensive overhead activity

26. Two common scenarios 1. Elapsed query time close CPU time, wait time negligible. Consumption fair: reasonable nb pages being accesses, most of them are logical accesses; nb locks low or inexistent and no deadlocks or lock escalations; if there were sorts, they didn’t augment the nb of logical/physical reads/writes Access plan is executed in the best way possible; if the rest of the chain is balanced, this is the best performance the system can deliver for this query 2. Noticeable discrepancy between elapsed and CPU time, wait time fills the gap. So there must be a problem in resource consumption: contention pb (probably waiting for locks) or poorly performing resource (ex. If area allocated for sorting is small, then additional I/O).  To distinguish between contention and physical resource consumption problem, run the query in isolation

27. Performance Monitors to Profiling Queries Access or compute performance indicators’ values at any time Many, many flavors  Scope of the indicators: Generic (all indicators from OS and DBMS) or specific (correlated indicators of a given subsystem or a given query) Choose the generic one when interested in how several distinct indicators behave together  Frequency of data gathering: snapshot perf indicators: allow to freeze a situation at a specific moment in time Continuous perf. Indicator: snaphotting at regular intervals Alarm modes: allows us to enter thresholds beyond which the system is acting abnormally  Presentation of data: textual or graphical Textual good for seeing the precise values of indicators Graphical chosen when wants to see the trends

28. An example Performance Monitor (query level) Details of buffer and CPU consumption on a query’s report according to DB2’s Benchmark tool Similar tools: MSSQL’s SET STATISTICS switch and Oracle’s SQL Analyze Tool Statement number: 1 select C_NAME, N_NAME from DBA.CUSTOMER join DBA.NATION on C_NATIONKEY = N_NATIONKEY where C_ACCTBAL > 0 Number of rows retrieved is: 136308 Number of rows sent to output is: 0 Elapsed Time is: 76.349 seconds … Buffer pool data logical reads = 272618 Buffer pool data physical reads = 131425 Buffer pool data writes = 0 Buffer pool index logical reads = 273173 Buffer pool index physical reads = 552 Buffer pool index writes = 0 Total buffer pool read time (ms) = 71352 Total buffer pool write time (ms) = 0 … Summary of Results ================== Elapsed Agent CPU Rows Rows Statement # Time (s) Time (s) Fetched Printed 1 76.349 6.670 136308 0 Statement number: 1 select C_NAME, N_NAME from DBA.CUSTOMER join DBA.NATION on C_NATIONKEY = N_NATIONKEY where C_ACCTBAL > 0 Number of rows retrieved is: 136308 Number of rows sent to output is: 0 Elapsed Time is: 76.349 seconds … Buffer pool data logical reads = 272618 Buffer pool data physical reads = 131425 Buffer pool data writes = 0 Buffer pool index logical reads = 273173 Buffer pool index physical reads = 552 Buffer pool index writes = 0 Total buffer pool read time (ms) = 71352 Total buffer pool write time (ms) = 0 … Summary of Results ================== Elapsed Agent CPU Rows Rows Statement # Time (s) Time (s) Fetched Printed 1 76.349 6.670 136308 0

29. An example Performance Monitor (system level) An IO indicator’s consumption evolution (qualitative and quantitative) according to DB2’s System Monitor Similar tools: Window’s Performance Monitor and Oracle’s Performance Manager

30. Investigating Intermediate and Primary Resources: routine monitoring Answer Q3 Problems at OS level? Answer Q2 Tune low-level resources yesno Problematic subsystems? Tune subsystems Investigate upper level yesno

31. Investigating Primary Resources Answer question 3: “Are there enough primary resources available for a DBMS to consume?” Primary resources: CPU, disk/controllers, memory, and network Analyze specific OS-level indicators to discover bottlenecks. A system-level performance monitor is the right tool here

32. CPU Consumption Indicators at the OS Level 100% CPU % of utilization 70% time Sustained utilization over 70% should trigger the alert. System utilization shouldn’t be more than 40%. DBMS (in a non- dedicated machine) should be getting a decent time share. total usage system usage

33. Disk Performance Indicators at the OS Level Wait queue Average Queue Size New requests Disk Transfers /second Should be close to zero Wait times should also be close to zero Idle disk with pending requests? Check controller contention. Also, transfers should be balanced among disks/controllers

34. What to do in case of problem? If there are long waiting queues and long service times, check the number of bytes transferred to confirm the disk is busy.  Idle disks with pending requests imply controller problems Also compare the wait times from each disk, if we have a group of disks serving the same purpose Only reorganize data if we are sure the I/o problems are not dur to high-level consumer inefficiency

35. Memory Consumption Indicators at the OS Level pagefile real memory virtual memory Page faults/time should be close to zero. If paging happens, at least not DB cache pages. % of pagefile in use (it’s used a fixed file/partition) will tell you how much memory is “lacking”.

36. What to do in case of problems? If the DB buffer size was properly configured, paging should come from non-DBMS sources. Checking memory at the OS level should be done whenever we change the memory allocation parameters of the DBMS or there are significant changes in the nb of users

37. Investigating Intermediate and Primary Resources: routine monitoring Answer Q3 Problems at OS level? Answer Q2 Tune low-level resources yesno Problematic subsystems? Tune subsystems Investigate upper level yesno

38. Investigating Intermediate Resources/Consumers Answer question 2: “Are subsystems making optimal use of resources?” Main subsystems: Cache Manager, Disk subsystem, Lock subsystem, and Log/Recovery subsystem Extract and analyze relevant perf. indicators A Performance Monitor is usually useful, but sometimes specific tools apply

39. Cache Manager Performance Indicators Table scan readpage () Free Page slots Page reads/ writes Pick victim strategy Data Pages Cache Manager If page is not in the cache, readpage (logical) generate an actual IO (physical). Ratio of readpages that did not generate physical IO (cache hit ration) should be 90% or more Pages are regularly saved to disk to make free space. # of free slots should always be > 0

40. What to do in case of problem? If indicators have bad values, the buffer manager is not saving as many disk accesses as it should  One or a small nb of queries may be responsible  May have slow disk subsystem  May have insufficient memory Tuning parameters accordingly

41. Disk Manager Performance Indicators rows page extent file Storage Hierarchy (simplified) disk Row displacement: should kept under 5% of rows Free space fragmentation: pages with few space should not be in the free list Data fragmentation: ideally files that store DB objects (table, index) should be in one or few (<5) contiguous extents File position: should balance workload evenly among all disks

42. What to do in case of problem? Bad indicator values usually mean inadequately chosen storage parameters  To choose good storage parameters, should be aware of rate at which data grows and with what mix of inserts/updates/deletes that growth rate happens  Also, identifiy occasional but predicatble fluctuations in the DB activity  Then tune disk subsystem

43. Lock Manager Performance Indicators Lock request ObjectLock TypeTXN ID Lock List Locks pending list Deadlocks and timeouts should be virtually inexistent or extremely low (no more then 1% of the transactions) Lock wait time for a transaction should be a small fraction of the whole transaction time. Number of locks on wait should be a small fraction of the number of locks on the lock list, under 20%

44. What to do in case of problem? The lock subsystem is often a reflection of how well the transactions were designed and seldom a source of a problem itself Can move up the consumption chain and find expensive queries, including data locking ones.

45. Logging Subsystem Performance Indicatores Log is used by every transaction that alters, inserts or deletes data Well tuned logging subsyst. Can write log records at a pace that doesn’t slow active transactions Indicatores to measure:  Number of log waits to confirm that the disks storing the log files are keeping pace with the transaction workload – should be zero  Nb of log expansions or log archives due to lack of space – if bigger than zero indicates a configuration problem  Log cache-hit ratio to check whether the size of the log buffer is adequate

46. Conclusion Monitoring a DBMS’s performance can be done in a systematic way  The consumption chain helps distinguishing problems’ causes from their symptoms  Existing tools help extracting relevant performance indicators  The three questions guide the whole monitoring process