Wednesday, September 24, 2008

Oracle World 2008, 9/24, 1pm

SARM BEST PRACTICES

Siebel Application Response Management
Should help address these questions:
*where is the bottleneck for transaction x by user y
*which screen/view/bizcomp/script/wf were involved
*how much cpu and memory was consumed
*how many times were vaious components invoked
*how has transaction processiona nd resource consumption changed from release to release

SARM consists of framework, instrumentation, and tools
Oracle Enterprise Manager - management of the entire stack (app, middle, db) became available for Siebel 7.7 and up

Siebel Transaction Diagnostic Tool
*guidance: turn on SARM, level 1 w/ no overhead
*sarm is saved into a memory buffer to alleviate disk IO
*sarm data is flushed into the siebel file system in binary which can be converted to csv post processing (command line tool)
*sarm takes 3 to 4 % cpu
*sarm calls are < 40 microseconds per call
*based on benhcmark with 500 concurrent users, 30 seconds think time, on call center

sarm parameters set on siebel server, environment variables set on web and app server
0-off
1-monitoring
2-diagnostic
buffer default: 5,000,000 bytes
flush period: 3 (mins) default
max files: 4 default
file size: default 15,000,000 bytes

level 1 is considered best practice, you want it turned on for the obj manager that end users interact with

siebel 7.5, requires a bounce of servers.. later versions do not require
siebel 8 allows you to turn it on for a specific user

Best practices:
Use SARM throughout app lifecycle
Use SARM during pre-prod load tests
Use SARM for monitoring in production
Take SARM snapshots periodically (before a change to the prod environment) for comparisons

DEMO: featuring oracle enterprise manager, to find slow performance by user CCHENG
*report generated based on filtering on start/end dates showing summary of performance of each transaction request
*each sarm request can be drilled down to get to the bottleneck but cannot get to the sql statement by drilldown, but its in progress for a future release.
*for now, use sarm id to correlate to generated sql.
*if a scripting issue, the SARM tool will indicate script name!

No comments: