Transcript Title of Your Presentation

Performance and Tuning
Mark Nesson, Vashti Ragoonath
June 2008
Copyright 2007, Information Builders. Slide 1
Performance and Tuning
Overview
Performance Testing
 Used to eliminate bottlenecks and establish a baseline
 Should be conducted in a controlled environment
 Must have a clear set of expectations
 Number of concurrent users
 Response times for queries
 Meet users expectations and Service Level Response times
Tuning
Repetitive process of:
 Running tests to identify bottlenecks
 Tuning one component at a time and redoing tests
 Increasing load to identify all bottlenecks
..until objectives are met
Copyright 2007, Information Builders. Slide 2
Performance and Tuning
Overview
Bottlenecks can exist at different levels:
 Application Level
 Developers can look for inefficiencies in their code
 Database Level
 Use query optimizers and database profilers to check for
bottlenecks
 Operating System Level
 Administrators can use utilities tools like PerfMon on Windows
and on UNIX top, vmstat, iostat to track CPU, RAM, Disk Space
utilization
 Network Level
 Network Administrators can use utilities like netstat, packet
sniffers like tcpdump and others like ethereal
Copyright 2007, Information Builders. Slide 3
Performance and Tuning
Tuning Components
External Tuning Components
 Operating System
 Database Server
 Application Server
 Java Virtual Machine
 Network
WebFOCUS Tuning Components
 Application
 Data Adapters Dialect Specific or Generic
 Reporting Server Workspace Manager
 Reporting Server Java Services
 ReportCaster Distribution Server
Copyright 2007, Information Builders. Slide 4
Performance and Tuning
External Tuning Components - OS
 Operating System
 Work with the OS Administrators to identify bottlenecks. If the


system is maxed out on CPU and RAM use the system utilities
to identify which processes are taking up all the resources.
Maybe some jobs can be run at off-peak hours or deferred and
a lower CPU priority.
Check all configuration files to ensure they are in sync with the
current releases of installed software. Upgrades may require
updates to old configuration files that could improve
performance overall on the system.
Track usage patterns on the machines to identify problems and
failures which can be fixed easily.
Copyright 2007, Information Builders. Slide 5
Performance and Tuning
External Tuning Components – Database
Database Server
 Verify that the database optimizer runs as recommended to
generate good data statistics. Example: Oracle ANALYZE
command creates optimizer statistics to choose methods of fast
access to the data. If the statistics were not generated properly,
performance will be affected.
 Fix Indexes. Use EXPLAIN to see what the optimizer does with
SQL statements. Does it do a full table scan? Does the table have
appropriate indexes for the queries?
You may have to rebuild indexes.
 Spread the Database Over Disks and I/O Channels. e.g: Partition
tables and indexes over multiple disks
Copyright 2007, Information Builders. Slide 6
Performance and Tuning
External Tuning Components – App Server




Disable Auto-deployment
Use Pre-compiled JavaServer Pages
Disable Dynamic Application Reloading
Tune Thread Pool
 Initial Thread Pool size
 Minimum number of threads in pool
 Maximum Thread Pool size
 Maximum number of concurrent threads in pool
 Thread inactivity timeout
 Time spent waiting for client response before being
returned to pool
 Growable thread pool
 Number of threads the pool should increase by when it’s
maxed out
Copyright 2007, Information Builders. Slide 7
Performance and Tuning
External Tuning Components – JVM
Java Virtual Machine (JVM)
 Tune Heap Size so garbage collection time is minimized and
clients can still be processed.
 Initial Heap Size
 Maximum Heap Size
 Garbage Collection
 Use the -verbosegc option to track frequency and times of
garbage collection
Copyright 2007, Information Builders. Slide 8
Performance and Tuning
External Tuning Components – Network
Network
 Network administrators will be able to identify and tune
bottlenecks are with the use of:
 packet sniffers such as tcpdump
 network protocol analyzers such as ethereal
 utilities such as netstat, ping, traceroute and others
Copyright 2007, Information Builders. Slide 9
Performance and Tuning
Demo: JVM Tuning
 Let’s see a live demo of tuning the Tomcat thread pool
 Test Case 1 – Low thread pool
 Tomcat parameters: Low thread pool(15), 256MB Heap size,




25 concurrent users, request sleeps for 15 secs, 100 available
tscom agents
Capture baseline performance statistics
Tune the Tomcat JVM: Large thread pool(75), 256MB Heap
size, 25 concurrent users, request sleeps for 15 secs, 100
available tscom agents
Capture performance statistics
Response times improved by a little less than 50%!
Copyright 2007, Information Builders. Slide 10
Performance and Tuning
WebFOCUS Tuning Components – Application
Application
 Verify that requests are generating optimized SQL
 Enforcing all RDBMS efficiencies
Data Adapters
 Enable OPTIMIZATION flags ( SET OPTIMIZATION)
 Fetch data in larger blocks (SET FETCHSIZE)
 Set ISOLATION Level RU (Read Uncommitted)
There are many more documented in the Data Adapters Manuals
Copyright 2007, Information Builders. Slide 11
Performance and Tuning
Tuning Components –Reporting Server
Workspace Manager -Data Services
 maximum
 number_ready
 deployment * private connection_pooling
 Queuing
 idle_session_limit
 idle_agent_limit
 cpu_limit
 memory_limit
 connection_limit
 max_connections_per_user *
 agent_refresh
 sched_priority
Copyright 2007, Information Builders. Slide 12
Performance and Tuning
Tuning Components –Java Services
Workspace Manager –Configuration
 Java Services
 Most commonly used to:
 Rendering graphical images on the Reporting Server
 With the JDBC Adapter
 Tuning Parameters
 Initial Java Heap Size
 Maximum Java Heap Size
 Java Thread Stack Size
 JVM_OPTIONS
 Specify other performance type options
Copyright 2007, Information Builders. Slide 13
Performance and Tuning
Demo: Graph Processing



Let’s see a live demo of how Java Services is used for processing
graphs.
 JSCOM Listener under Java Services
 Can have a cluster of JSCOM Listeners
Graph request generates 25 images in the report. Using the
default JSCOM Listener we will run 25 concurrent users and look
at the response time.
We then add a second JSCOM Listener and rerun our test.
 Is there an improvement in the response times?
 Yes there is…
Copyright 2007, Information Builders. Slide 14
Performance and Tuning
Tuning Components –ReportCaster
ReportCaster Configuration
 General Tab
 Maximum Threads
 Data Server (NODE)
 Maximum Connections
 Data Server (CLUSTER)
Weight
Maximum Connection
Algorithm to determine which server to send request to:
 Weight * maximum connection = priority
 Request will be routed to server with highest priority
Copyright 2007, Information Builders. Slide 15
Performance and Tuning
Demo: ReportCaster Compression

Let’s see a live demo of how we can have a scalable distribution
server by using ReportCaster compression.
 Run request with embedded graphical images and distribute
in uncompressed PDF format to Report Library.
 Check to see what the original file size is in the BOTLIB table
and make sure it was not flagged as compressed in the
BOTSCHED table.
 Column AUDITOR in table BOTSCHED has a flag to
indicate if data is compressed.
 Column REPORTSIZE in table BOTLIB has file size
before compression.
 Column LIBFIELD1 in table BOTLIB has compressed file
size.
 Clone and schedule the same request and distribute in
compressed PDF format to Report Library.
 Check the file sizes.
 View both reports from Report Library
Copyright 2007, Information Builders. Slide 16
Performance and Tuning
Demo: ReportCaster Compression



You can distribute the same report via EMAIL and use the ZIP
option to minimize space allocation.
You can also ZIP and distribute Active Reports for your users and
they can slice and dice that data without being connected.
We now have Active Reports in Connected mode but more details
in another presentation that we will be doing.
Copyright 2007, Information Builders. Slide 17
Performance and Tuning
Conclusion

Questions and Comments.
Copyright 2007, Information Builders. Slide 18