Resource Mapping A Wait Time Based Methodology for Database Performance Analysis Prepared for Hotsos Symposium, 2005 Presented by Matt Larson Chief Technology Officer Confio Software
Presentation Agenda
2
Introduction Conventional Tuning vs. Wait-based Tuning Foundation: Resource Mapping Methodology 5 Key Steps of Applying RMM RMM Advantages Conclusion
Who am I?
Former DBA consultant specializing in Oracle performance tuning Co-author of three Oracle books (Oracle Development Unleashed, Oracle Unleashed 2 Edition, Oracle8 Server Unleashed) Co-author of two other database related books CTO and founder of Oracle performance software company nd
3
Problems with Conventional Tuning Tools: Like the Drunk Under the Streetlight
4
Conventional Tuning
5
Art, not a science Ratio-based (cache hit ratios, etc.) Sometimes fruitless It’s “tuned” (I guess?) Different tuning/investigation process for each DBA/DBA Team/Company
Problems with Conventional Tuning Tools
Optimize systems, not business results Conventional tools: • V$ Views: limited visibility & granularity • Statspack: averages across entire database • Explain Plan: deemphasizes how non-object resources affect performance
Incorrect Data hides real results • System-wide averages • Event counters • Incomplete visibility
6
What Problems are you Trying to Solve? • I spend the whole week monitoring and optimizing Oracle configurations, but I have no demonstrable results to show for it - why? • Will more hardware make my application run faster? By how much? • Will the new application run efficiently on the production server? • Why does one application keep impacting my SLA compliance? • If I could make one (or 2, 3, or 4) changes to my database to have the biggest impact, what would they be? 7
You know you are working on the wrong thing when…
After spending an agonizing week tuning Oracle buffers to minimize I/O operations, management typically rewards you with: • • • •
8
A. B. C. D.
An all expense paid vacation A free lunch A stale donut Reward? Nobody even noticed!
You know you have a visibility problem when…
You measure database performance based on: • • • •
9
A. B. C. D.
Increasing trends in user response time Increasing system down time Increasing help desk calls Increasing decibel levels from irate users
Your role is sub-essential to the business of your organization when…
Your role in the rollout of a new customer facing application results in: • • • •
10
A. B. C. D.
Keys to drive the CEO’s Porsche Keys to use the executive restroom A mop to use in the executive restroom Your office has been moved to the restroom
You know you are accustomed to measuring the wrong thing when…
You measure the commute time to work based on: • • • •
11
A. B. C. D.
The time it takes to get there Counting the times your wheels rotate Monitoring your tachometer The number of speeding tickets
Wait-based Performance Tuning
12
Emerging best-practice for database tuning Proponents include leading consultants, trainers and authors Oracle is starting to build wait-based tuning tools into the database particularly in 10g Tune by determining where processing time is spent
Oracle 10g - Moving towards wait-based
Adding wait-based columns to existing views New wait-based views
Example: v$session_wait_history • • •
13
Provides the last 10 wait events for a session Session ID, Username, Event, Wait_Time, etc. Used to provide wait_time for only a few events
DBA Success Stories using RMM
14
DBA solves a “Cold Case”. Problem unresolved for 1 year with traditional tools; Solution identified in 10 minutes during hands-on training DBA ends “Crit Sit” 2 week situation ends quickly after identification of Library Cache pin wait and load locks. Metalink identifies Oracle bug, patch successfully applied DBA saves $700K. 90% CPU capacity initiates expansion from 12 to 24 CPU server. DBA identifies parallel queries across 16 parallel threads as source of bottleneck. CPU eliminated as constraint, no new server required.
RMM: Confio’s Underlying Methodology
Resource Mapping Methodology: Wait-Event Analysis General approachbest practice
Resource Mapping Methodology Rigorous, complete requirements
DBFlash Packaged product implementation
Three Key Principles of RMM 1. SQL View : All statistics at SQL statement level 2. Time View : Measure Time, not number of times a resource is utilized 3. Full View : Separately measure every resource to isolate source of problems
15
Confio’s Resource Mapping Methodology • The principles of RMM can be illustrated by using the analogy that data processing is like an assembly line. Data goes in one end, is subject to a series of changes, and comes out the other end as a finished product • The assembly line (or SQL Statement) must be observed at the lowest level where a unit of work is being performed (SQL View Principle) • Measurements are made with regard to time instead of counting how often an event occurs (Time View Principle) • All resources system-wide must be monitored to get a full view of potential bottlenecks i.e. no blind spots (Full View Principle) Counters CPU 74% Reads 1789327
145 seconds 16
Time
Counters Blind Spot
Follow a unit of work through every operation
CPU 38% Reads 4955
8726 seconds Time
Blind Spot
Track SQL Time, Not System Counters • Watching Counters leads to wrong conclusions: Time is more relevant • Total System Counters hide information: Need breakdown to individual SQLs Total System Counter SQL 1 SQL 2 SQL 3 Resource s 17
80K Reads
30 Minutes 5 R 15M 25 R 5M 50 Reads
I/O
5K Packets 125 Attempts
4 M
6 M 50 A
200 Minutes 10 M 35 A 5M
Network
216K Writes
4 M
200 Minutes
100 Minutes 5M 50 A
Locks
Redo
RMM-compliant Performance Tools
Oracle Tracing • RMM compliant when wait events are traced • Shows SQL level statistics (SQLView), all events (FullView) and events by time (TimeView) • Text-based, short-term technical reporting • Primarily used for reactive tuning
Confio DBFlash for Oracle • • • •
18
RMM compliant 24/7 proactive monitoring Graphical, long-term trend reporting RMM-based Alerting
Applying RMM for Business Results
Five Step Process focusing on what matters
1. Identify
19
2. Allocate
3. Quantify
4. Prioritize
5. Assign
Step 1: Identify
Identify SQL Statements having largest impact • (SQL View and Time View principles)
20
Longest wait times = most significant “pain points” for customers Conversely, low cache hit ratios or high latch usage may not impose high wait times for users (so why fix them?)
SQL statements prioritized by Total Wait Time
Step 2: Allocate
Allocate impact to real customers (internal or external) Allocate wait time to Program, Session, Machine • SQL View principle makes this connection
21
Understanding database customer and application
Programs Prioritized by Total Wait Time
Step 3: Quantify
How much is save in time/money if fixed? Enabled by Full View and Time View principles Soft dollar savings • Data entry clerks • DBA time spent in problem resolution
Hard dollar savings • Reduce hardware upgrades • Meet SLA’s avoiding penality • Ensure business isn’t lost due to poor performing or unavailable system
Quantifiable benefit of Tuning a specific statement 22
Step 4: Prioritize
23
If last step properly executed, this step is fairly straight forward Allow’s DBA to cut through the clutter of potential new projects, investigations, and trials. Better justification for priorities. (e.g. We aren’t working on your problem since this other has a higher demonstrable business impact)
Step 5: Assign
Assign the right people to the problem • Log_buffer waits • Network issues • Same query 10,000/hour
24
Enabled by Full View principle Avoid finger-pointing by accurately assigning quickly
Resource Mapping Methodology RMM
Wait Based Tuning
25
Network, Storage, Application, Web, etc.
Silo Monitoring Business Management
LIMITED VIEW
IT Management
LIMITED VIEW
Web Team
Sitescope
Custom App Team Network Team Database/OS Teams Storage/OS Teams
Often No Commercial Tools
Software Layers Web Server Custom Biz Logic Network
HP Openview Database Server
Wait-based tuning Storage Box
EMC Control Center
Each team uses their own tool to partially monitor their non-Oracle layers. No view across layers. Management has no clear view. 26
The Solution - Integrated Vision Business Management
RM M acr oss the stack
IT Management Web Team
Web Server
Custom App Team
Custom Biz Logic
Network Team
Network
Database/OS Teams
Database Server
Storage/OS Teams
Storage Box
All teams see a complete picture of all layers and dependencies. Enables more efficient “Umbrella� solution.
27
RMM Achieved Business Benefits
28
RMM Does:
Business Benefit:
35% reduction in database capacity requirement
Reduce capital investment Avoid unnecessary additions Recovers un-used capacity
Standardizes “expert� analysis ability across entire DBA team
Reduce training & consulting costs
Quantifies performance impact
Focus tuning efforts on biggest business impacts
Identifies problem Root Cause and resolution
Assign human resources and responsibility
Anticipates + resolves performance bottlenecks
Maintain SLA and end user performance
Example 1: Problem Observed
Critical situation: Secure Service Center application performance unsatisfactory • Response time between 2400 and 9000 seconds • Very high network traffic (3x—4x normal), indicating time-outs and user refreshes • “CritSit” declared: major effort to resolve problem
29
Observations using Resource Mapping Methods  
1: Identify accumulated Waits 2: Identify specific resources used
Lib cache pin wait
Lib cache load lock 30
Notice scale: > 8000 secs
Results
Library cache pin nearly unobservable
Library cache load lock no longer observable
31
Notice scale: < 1400 secs max
Results
Response time improvement from 8000 seconds (worst case) to 900 seconds Variance improvement: • Before: response time 2400 - 8000 sec • After: response time 800 - 900 sec
32
Example 2: Performance Drain – Identify the Source
33
Slow response reported DBA and database focus of delays Database problem? No – SQL*Net Message identified as source of delay 2nd highest wait event
RMM Drill Down identifies source of problem
Single application generates all SQL*Net Messages App on same server as Oracle! Answer: Misconfiguration – TCP/IP used within server Change to IPC, eliminate NIC traffic and 30% of wait time
Solution requires knowing: Which SQL, What Wait Time, Which Resource 34
Example 3: Scattered Reads
Situation: LINS06 database - Hourly profile identifies high wait anomaly 3-10x higher than other periods – requires investigation
wait time 42,000 seconds 10:00-11:00
35
Drill Down to Key RMM Parameters
Db file scattered reads
Db file scattered reads 36
Notice scale: > 6000 secs
Conclusion
Look for what has an impact Resource Mapping is more that Wait Time – Analysis must include: • SQL level granularity • Full Resource granularity
37
Isolating the SQL and Resource allows you to find and fix the Root Cause DBAs can have an impact and be heroes!
Thank you for coming Matt Larson Contact Information • mattlarson@confio.com • 303-938-8282 ext. 110 • Company website www.confio.com
38