17
Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Embed Size (px)

Citation preview

Page 1: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 1

Workload Distribution for the Enterprise

Mark Nesson, Vashti Ragoonath June, 2008

Page 2: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 2

Workload Distribution for the Enterprise Overview

WebFOCUS installations supports hundreds, even thousands, of users. Today, it’s becoming more and more difficult to secure skilled systems administrators to manage systems: So what’s the answer

We need a self-managing or autonomic solution that will deliver service 24 hours a day, seven days a week

An autonomic system must be: ■ Self-Aware – The system must know what resources it has

available, and how these resources are functioning ■ Self-Healing – The system must be able to use its awareness to

correct problems in the delivery of service ■ Self-Optimizing – The system must be continuously adjusting its

resources to keep within the limits of performance needed

Page 3: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 3

Workload Distribution for the Enterprise WebFOCUS Workload Distribution

WebFOCUS contains an autonomic server called CLM (Cluster Load Manager) that has all three characteristics.

Self Aware because it: Knows which servers are running and it continuously measures

their response time Monitors all of the servers within the cluster and provides a

central point of management through its own Console CLM also maintains metadata describing the capacity of each

computer and the amount of memory available to the servers It starts up the servers when it needs to and begins to monitor

their performance

Page 4: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 4

Workload Distribution for the Enterprise WebFOCUS CLMSelf Healing because: It has the ability to bring servers up and down automatically It has the ability to monitor the cluster operation at a central

console resulting in less complexity and improved performance

Page 5: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 5

Workload Distribution for the Enterprise WebFOCUS CLM

Self Optimizing because: The CLM manages every instance of WebFOCUS. Load balancing

is dynamic. Should the load increase to the point that response times cannot be maintained within the limits.

The CLM logs statistics in a log file that can later be used to create custom reports that converts the data into meaningful information. The log is an invaluable resource, both for self-assessment and performance analysis.

Page 6: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 6

Workload Distribution for the Enterprise How do we use it?

Installation It is a licensed component that can be installed independent of

any other WebFOCUS software. Recommended to be installed ideally on its own machine If not,

with the WebFOCUS Client, but not on the Reporting Server machines.

For business critical systems, implement CLM Failover. CLM listens on a TCP/IP Port and can be started in the same way

as the Reporting Server. It has an HTTP Web Console from which one can define rules for

monitoring and dispatching requests to the Remote Servers.

Page 7: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 7

Workload Distribution for the Enterprise Configuring CLM

We will demonstrate the basics of configuring the CLM. From the CLM Console:

Enable Cluster Manager (enter License Code) and restart. Add Remote Servers to manage

Can flag Remote Servers as RESERVED Can specify startup command to start servers

Create one or more CLM Clusters Add Remote Servers to the Clusters Click on Configure the Cluster Manager to specify rules for

Remote Servers

Page 8: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 8

Workload Distribution for the Enterprise Specifying Rules for CLM

CLM rules apply to all Remote Servers independent of the cluster to which they were assigned. Polling History   Polling Interval     User Query Dispatch Method  Max Response Time Limit      Threshold for starting/stopping servers    Running Average Period      Cluster Manager Log  AUTO_CONFIG CLM_MAXWAIT

Page 9: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 9

Workload Distribution for the Enterprise Specifying Rules for CLM

An explanation of CLM rules: Polling History  

If CLM Logging is on, the number of records to keep for each Remote Server.

Polling Interval   How often should CLM poll the Remote Servers to do a health

check? User Query Dispatch Method 

What dispatch algorithm to use to determine which server the request should be routed to?

Page 10: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 10

Workload Distribution for the Enterprise Specifying Rules for CLM

An explanation of Dispatch Algorithms: Smart ( Default) – based on Response Times Random - random Round Robin – alternate between nodes in a cluster Weighted – each server is weighted according to response times

and then randomly chosen if more than 1 server has the same weight.

Page 11: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 11

Workload Distribution for the Enterprise Specifying Rules for CLM

An explanation of CLM rules: Max Response Time Limit 

Average Response time from Remote Server that is used as the threshold for starting/stopping servers      

Threshold for starting/stopping servers  Number of poll periods to determine if response times are

consistent before starting/stopping RESERVED servers.

Page 12: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 12

Workload Distribution for the Enterprise Specifying Rules for CLM

An explanation of CLM rules: Running Average Period

Average Response time only for the number of polls set by the threshold for starting/stopping servers. It’s not the average response time for server lifetime.

Cluster Manager Log  Set to ON if logs are to be captured

Page 13: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 13

Workload Distribution for the Enterprise WebFOCUS Client and the CLM

The WebFOCUS Client (WFC) has to be configured to communicate to one or more clusters defined on the CLM. Cluster names must match on WFC and CLM Specify the hostname and port of the CLM      Example of WFC communicating to CLM cluster node CLUST1

where CLM on different machine from WFC OR CLM and WFC are using different odin.cfg files. NODE = CLUST1BEGIN CLASS = CLUSTER REMOTE_CLM_HOST = jetsetter REMOTE_CLM_PORT = 38121END

Page 14: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 14

Workload Distribution for the Enterprise Case Scenarios

Let’s demonstrate how we can put these rules to use. Case 1:I have three servers that are monitored and load balanced by Cluster

Manager. My requirement is whenever the Response Times for any of those servers exceed 5 seconds, I want to start a RESERVED server.

Page 15: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 15

Workload Distribution for the Enterprise Case Scenarios

Let’s demonstrate how we can put these rules to use. Case 2:I have three servers that are monitored and load balanced by Cluster

Manager. My business is mission critical so I need to implement CLM Failover.Example of WFC odin.cfg set up for CLM FAILOVER

NODE = CLUST1BEGIN CLASS = CLUSTER REMOTE_CLM_HOST = jetsetter,jetsetter2 REMOTE_CLM_PORT = 38121,48121END

Page 16: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 16

Workload Distribution for the Enterprise Benefits

Here is a list of how features translate to benefits.

■ Lower total cost of ownership

■ Higher system up-time

■ Less human expertise needed

■ More uniform response time

■ More insight into system operations

Page 17: Copyright 2007, Information Builders. Slide 1 Workload Distribution for the Enterprise Mark Nesson, Vashti Ragoonath June, 2008

Copyright 2007, Information Builders. Slide 17

Workload Distribution for the Enterprise Conclusion

Questions and Comments.