Transcript Understanding and Improving Server Performance Alan Doerhoefer Timothy Hanson

Understanding and Improving Server
Performance
Alan Doerhoefer
Timothy Hanson
Tableau Software Inc.
All rights reserved. © 2009 Tableau Software Inc.
How To Improve Server Performance
Steps to Great Performance
1.
2.
3.
4.
5.
6.
7.
8.
9.
Learn the architecture
Use appropriate hardware
Use distributed server
Use server extracts
Tune
Monitor
Troubleshoot
Learn the types of views
Understand the load test results
All rights reserved. © 2009 Tableau Software Inc.
Understand Server Architecture
Tableau Server
Load Balancer
Tableau
Professional
All rights reserved. © 2009 Tableau Software Inc.
Application State
Data
Sources
Web
Web
VizQL
App
App
Processes
Repository
Application
Web
Web
(WG)
App
App
Processes
Use Appropriate Server Hardware
Good Server for performance:
• Windows Server 2003 or 2008 64 bit
• 16 GB RAM
• Fast RAID-5 Drives
• Gigabit Network Card
• 8 Fast CPU cores
All rights reserved. © 2009 Tableau Software Inc.
Use Appropriate Server Hardware
But Tableau Server will work on lower
class hardware:
• Windows Server 2003 32 bit
• 2 GB RAM
• Any type of drive
• 100 Mb Network Card
• 2 slow CPU cores
All rights reserved. © 2009 Tableau Software Inc.
Use Distributed Server
Worker 1
VizQL
oror
VizQL
VizQL
or
WG
WG
Application
Processes
Processes
Processes
Main
Load Balancer
Worker N
Application State
Repository
Worker 2
VizQL
oror
VizQL
VizQL
or
WG
WG
Application
Processes
Processes
Processes
All rights reserved. © 2009 Tableau Software Inc.
Use Distributed Server
Installation is fast and easy
All rights reserved. © 2009 Tableau Software Inc.
Use Distributed Server
Installation
All rights reserved. © 2009 Tableau Software Inc.
Use Distributed Server
Determining the number of server processes
Use 2 X the number of processor cores
Example
3 server cluster (2 workers)
8 cores per machine
•Use 2 x 8 = 16 App processes on one worker
•Use 2 x 8 = 16 VizQL processes on the other worker
All rights reserved. © 2009 Tableau Software Inc.
Use Server Extracts
• New in 5.0
• A database that stores “extracts” of user data locally and efficiently
• Performance is improved since data is local and indexed
• Data can be filtered and aggregated as part of the extract
• Can be set to refresh data sources automatically on a schedule
• The following extract database types are supported:
– Built-In
– Postgres
– MySQL
– Microsoft SQL Server
• Please attend Extracts: Fast Data To Go Thurs. 11 am for more
extract details
All rights reserved. © 2009 Tableau Software Inc.
Tune the OS
• Windows 2003 Server (SP1 or Later)
TCP Setting
Not an issue in Windows 2008 Server
Set SynAttackProtect to zero:
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\
Parameters] "SynAttackProtect"=dword:00000000
• Disable other processes
All rights reserved. © 2009 Tableau Software Inc.
Tune the Server
Session Timeouts
Can be changed in the configuration file
Tableau Server\config\tabsvc.yml
• WG Session
Default is 4 hours
To change it to 1 hour add this to tabsvc.yml:
wgserver.session.idle_limit: 60
• VizQL Session
Default is 2 hours
To change it to 30 minutes add this to
tabsvc.yml:
vizqlserver.session.expiry.timeout: 30
All rights reserved. © 2009 Tableau Software Inc.
Monitor Server
Server Status
Demo
All rights reserved. © 2009 Tableau Software Inc.
Monitor Server
Server Activity
Demo
All rights reserved. © 2009 Tableau Software Inc.
Monitor Server
User Activity
Demo
All rights reserved. © 2009 Tableau Software Inc.
Troubleshoot Server
Apache logs
Contains information about every request to server at the HTTP
request level
All rights reserved. © 2009 Tableau Software Inc.
Troubleshoot Server
WG logs and VizQL logs
Contains details of each operation performed by the server
All rights reserved. © 2009 Tableau Software Inc.
Troubleshoot Server
tabadmin ziplogs command
collects all logs – even on a distributed cluster – into one archive
All rights reserved. © 2009 Tableau Software Inc.
Understand View Types and Performance
Three Types Of Views:
1. Fast
- Fast query
- Fast rendering
2. Query-bound
- Slow query
- Fast Rendering
- Same as Fast once query is cached
3. CPU-bound
- Render-Intensive
- Uses the most CPU
- Example: text tables with thousands of cells
All rights reserved. © 2009 Tableau Software Inc.
Understand the Load Test Results
Capabilities of:
• 1 machine
• 3 machines
• 5 machines
* Note that these tests show results from a specific test
configuration and should not be taken as a guarantee of
client response times. These benchmark results were
returned in a controlled lab environment, without other
applications running during execution. Actual results
will vary based on load type, hardware, network speed,
browser settings, and database performance.
All rights reserved. © 2009 Tableau Software Inc.
Understand the Load Test Results
Initial View Load Times
Effect of caching
All rights reserved. © 2009 Tableau Software Inc.
5 servers in our
load tests served
2,250 named
users, or 225
concurrent users
Understand the Load Test Results
Caching
- Pre-load views whenever possible to improve initial load times
All rights reserved. © 2009 Tableau Software Inc.
Summary
Keys to Performance:
- Understand view types
- Understand and utilize the distributed architecture
- Use server extracts
- Learn how to monitor and investigate
All rights reserved. © 2009 Tableau Software Inc.