Performance

[kjanes]

I believe the preload option uses the Hash File Stage Read/Write Cache sizes to determine whether or no it will preload the hash table.

Using DataStage Admin, check the project properties under Tunables for the Hash File Read/Write Cache size. By default, it is 128MB.
If the server has enough cache, it will preload. Check your DataStage log for the job to see if the cache is turned on or off.

In addition, under DataStage 6, you can possibly use Row Buffering to help speed up performance. This can be turned on at the Job level (Job Properties) or the Server level (properties - performance). I have found Row Buffering to be a good performance enhancer more so that preloading hash files into memory. There are two different types of Row Buffering. In Process and Inter Process. Check help to see which may be more applicable for you.

Kevin Janes

[ray.wurlod]

For a sufficiently small hashed file (say up to 16K, and in a well designed DataStage job you'd be surprised just how many are this small), there is no real value is using the cache, since the hashed file itself will tend to be resident in memory.
The larger the hashed file, the more likely you are to benefit from using the write cache. The reason is that writes to a hashed file tend to be to random locations within the file; writing to random locations on disk involves a lot of waiting for the heads to move (seek activity), whereas writing to random locations in memory involves no such delay. Once the load is complete, the memory cache can be flushed to disk in group order, minimizing real I/O.

Whether this represents a gain in "performance" depends on how you define "performance".