Friday, August 24, 2007

Cluster ....

It appears that there is a significant change that I had overlooked in the new OBI 10.1.3.2, compared to the Analytics 7.8 version.

I had noticed that there was a new webcat structure, that was easy, but one of our contacts spotted that there is a new way of clustering the new Webcat.

Previously, each Webcat server (BI Presentation server) had it's own copy of the webcat file, held within the Data folder.
If you had two or more webcat servers they would syncronise the contents of the webcat using a utility - something that did not work very well.

Now you have one webcat, irrespective of how many web servers you use. They all point to the same webcat.
This neatly solves the problems you get with Syncoronisation, but introduces one significant flaw - a Single Point of Failure.

What do you do? Put your webcat on a SAN, which has all the RAID you can get. Back up the whole webcat folder every night - obviously!
Now take snapshots of the webcat throughout the day, but be very careful how many you take. Too many will slow the system down.

One thing I have always felt though is that you should cluster for Customers' High Availability requirements NOT for performance.

The majority of performance issues should be dealt with by database design, prudent Caching (with seeding), adequate memory, and ETL design. Only when these have been worked through should you look for that extra 5% from clustered servers.

1 comment:

Anonymous said...

Hi,
Following is the OBI environment we currently have:
1 Instance of OBI 10.1.3.2 server component installed on Server Machine
1 Instance of OBI 10.1.3.2 presentation services installed on seperate Web Server machine
OBI Scheduler services installed but not configured and not in use.

We now have a requirement to configure OBI for high availability with BigIP, a load balancer server which will be entry point for End User web requests. This would load balance between the two web servers which would connect to one BI Server instance. Thus, webcat file will be placed on a shared network drive where it saves changes received from any one of the two web servers.
We are not clustering the BI Server, thus, we have only one BI Server machine that hosts Admin Tools/respository.

I have read chapters 3 and 4 of OBIEE Deployment Guide that explain the clustering and load balancing for high availability, however, if you can answer/explain a procedure that would specifically apply to my above explained scenario, it would really help me clear some confusions.

1. Using IIS, does BI Presentation Services Plug-in (Java Servlet)or ISAPI Plug-in install on server machine that will act as Load Balancer only or does it need to be installed on the 2 web servers (where presentation services are installed)? or all three?

2. Since we have only one BI Server, do we need to install and configure BI Cluster Controller, configure Scheduler or change NQClusterConfig.INI file?

The guide mentions "Since communications to BI Presentation Services instances in a cluster are not controlled via the Cluster Controllers, the list of BI Presentation Services instances must be specified. This is done
either through Job Manager or the schconfig utility. When more than one BI Presentation Services instance is specified, load balancing of requests to the multiple BI Presentation Services instances automatically takes effect."

3. Does this mean, BI Cluster Controller installing and enabling need not required, simply install ISAPI on IIS hosts, configure isapiconfig.xml file and we are done?

I would really appreciate your help. Also if you have come across any blog or any other documents that explain clustering/load balancing of OBI, it would be a great help.

Thank you,
Regards,
Afia Tahir

The Cowes

The Cowes
Cowes Racing

Sudoku