Informatica网格和原理.ppt

* The Integration Service uses both process memory and system shared memory to perform these tasks. It runs as a daemon on UNIX and as a service on Windows NT/2000. The Integration Service uses the following processes to run a session: The Load Manager process. Starts the session, creates the DTM process, and sends post-session email when the session completes. The DTM process. Creates threads to initialize the session, read, write, and transform data, and handle pre- and post-session operations. The Load Manager is the primary Integration Service process. It performs the following tasks: Manages session and batch scheduling. Locks the session and reads session properties. Reads the parameter file. Expands the server and session variables and parameters. Verifies permissions and privileges. Validates source and target code pages. Creates the session log file. Creates the Data Transformation Manager (DTM) process, which executes the session. LM memory: set to approximately 200K/concurrent session; don’t set too high. It is a function only of the maximum number of concurrent sessions. * Remember our rules… step 1) establish a baseline and part of the baseline is understanding the load on the computer when “idle” * Here’s a fairly complex mapping. Source data is read from a file and aggregated. The agg output is split to SCD type processing as well as through an in-line normalizer that pivots address information. * Our attention is immediately drawn to the fact that the transformation thread is 98% busy. Let’s look at re-partitioning.. NOTE: This isn’t exactly the default stages. Extra credit to anybody who can say why? OK, I’ll tell you – there should be an additional thread created by the aggregator since aggs, by default, get their own stage. However, with K2, you can remove these partition points and I thought this better illustrated the example since the default pipeline’s first stage basically does nothing (includes all transformation