Datastage Performance Tuning

Hi All,

Here is some common process for tuning Datastage jobs to improve the performance.

1	Staged the data coming from ODBC/OCI/DB2UDB stages or any database on the server using Hash/Sequential files for optimum performance
2	Tuned the OCI stage for 'Array Size' and 'Rows per Transaction' numerical values for faster inserts, updates and selects.
3	Tuned the 'Project Tunables' in Administrator for better performance
4	Used sorted data for Aggregator
5	Sorted the data as much as possible in DB and reduced the use of DS-Sort for better performance of jobs
6	Removed the data not used from the source as early as possible in the job
7	Worked with DB-admin to create appropriate Indexes on tables for better performance of DS queries
8	Converted some of the complex joins/business in DS to Stored Procedures on DS for faster execution of the jobs.
9	If an input file has an excessive number of rows and can be split-up then use standard logic to run jobs in parallel.
10	Before writing a routine or a transform, make sure that there is not the functionality required in one of the standard routines supplied in the sdk or ds utilities categories.Constraints are generally CPU intensive and take a significant amount of time to process. This may be the case if the constraint calls routines or external macros but if it is inline code then the overhead will be minimal.
11	Try to have the constraints in the 'Selection' criteria of the jobs itself. This will eliminate the unnecessary records even getting in before joins are made.
12	Tuning should occur on a job-by-job basis.
13	Use the power of DBMS.
14	Try not to use a sort stage when you can use an ORDER BY clause in the database.
15	Using a constraint to filter a record set is much slower than performing a SELECT … WHERE….
16	Make every attempt to use the bulk loader for your particular database. Bulk loaders are generally faster than using ODBC or OLE.
17	Minimize the usage of Transformer (Instead of this use Copy modify Filter Row Generator
18	Use SQL Code while extracting the data
19	Handle the nulls
20	Minimize the warnings
21	Reduce the number of lookups in a job design
22	Try not to use more than 20stages in a job
23	Use IPC stage between two passive stages Reduces processing time
24	Drop indexes before data loading and recreate after loading data into tables
25	Check the write cache of Hash file. If the same hash file is used for Look up and as well as target disable this Option.
26	If the hash file is used only for lookup then  enable Preload to memory . This will improve the performance. Also check the order of execution of the routines.
27	Don't use more than 7 lookups in the same transformer; introduce new transformers if it exceeds 7 lookups.
28	Use Preload to memory option in the hash file output.
29	Use Write to cache in the hash file input.
30	Write into the error tables only after all the transformer stages.
31	Reduce the width of the input record - remove the columns that you would not use.
32	Cache the hash files you are reading from and writing into. Make sure your cache is big enough to hold the hash files.
33	Use ANALYZE.FILE or HASH.HELP to determine the optimal settings for your hash files.
34	Ideally, if the amount of data to be processed is small, configuration files with less number of nodes should be used while if data volume is more , configuration files with larger number of nodes should be used.
35	Partitioning should be set in such a way so as to have balanced data flow i.e. nearly equal partitioning of data should occur and data skew should be minimized.
36	In DataStage Jobs where high volume of data is processed, virtual memory settings for the job should be optimized. Jobs often abort in cases where a single lookup has multiple reference links. This happens due to low temp memory space. In such jobs $APT_BUFFER_MAXIMUM_MEMORY, $APT_MONITOR_SIZE and $APT_MONITOR_TIME should be set to sufficiently large values.
37	Sequential files should be used in following conditions. When we are reading a flat file (fixed width or delimited) from UNIX environment which is FTP ed from some external system
38	When some UNIX operations has to be done on the file Don’t use sequential file for intermediate storage between jobs. It causes performance overhead, as it needs to do data conversion before writing and reading from a UNIX file
39	In order to have faster reading from the Stage the number of readers per node can be increased (default value is one).
40	Usage of Dataset results in a good performance in a set of linked jobs. They help in achieving end-to-end parallelism by writing data in partitioned form and maintaining the sort order.
41	Look up Stage is faster when the data volume is less. If the reference data volume is more, usage of Lookup Stage should be avoided as all reference data is pulled in to local memory
42	Sparse lookup type should be chosen only if primary input data volume is small.
43	Join should be used when the data volume is high. It is a good alternative to the lookup stage and should be used when handling huge volumes of data.
44	Even though data can be sorted on a link, Sort Stage is used when the data to be sorted is huge.When we sort data on link ( sort / unique option) once the data size is beyond the fixed memory limit , I/O to disk takes place, which incurs an overhead. Therefore, if the volume of data is large explicit sort stage should be used instead of sort on link.Sort Stage gives an option on increasing the buffer memory used for sorting this would mean lower I/O and better performance.
45	It is also advisable to reduce the number of transformers in a Job by combining the logic into a single transformer rather than having multiple transformers.
46	Presence of a Funnel Stage reduces the performance of a job. It would increase the time taken by job by 30% (observations). When a Funnel Stage is to be used in a large job it is better to isolate itself to one job. Write the output to Datasets and funnel them in new job.
47	Funnel Stage should be run in “continuous” mode, without hindrance.
48	A single job should not be overloaded with Stages. Each extra Stage put in a Job corresponds to lesser number of resources available for every Stage, which directly affects the Jobs Performance. If possible, big jobs having large number of Stages should be logically split into smaller units.
49	Unnecessary column propagation should not be done. As far as possible, RCP (Runtime Column Propagation) should be disabled in the jobs
50	Most often neglected option is “don’t sort if previously sorted” in sort Stage, set this option to “true”. This improves the Sort Stage performance a great deal
51	In Transformer Stage “Preserve Sort Order” can be used to maintain sort order of the data and reduce sorting in the job
52	Reduce the number of Stage variables used.
53	The Copy stage should be used instead of a Transformer for simple operations
54	The “upsert” works well if the data is sorted on the primary key column of the table which is being loaded.
55	Don’t read from a Sequential File using SAME partitioning
56	By using hashfile stage we can improve the performance. In case of hashfile stage we can define the read cache size & write cache size but the default size is 128M.B.
57	By using active-to-active link performance also we can improve the performance. Here we can improve the performance by enabling the row buffer, the default row buffer size is 128K.B.

Regards,
Venkat....