Fast Path buffer performance considerations

The performance considerations for Fast Path buffers differ depending on whether you are using the Fast Path 64-bit buffer manager.

If you are using the Fast Path 64-bit buffer manager to create and manage your Fast Path buffer pools, IMS optimizes many aspects of the buffer pool performance for you.

If you are not using the Fast Path 64-bit buffer manager, you might need to modify the buffer pool specifications yourself to maintain optimum performance of the buffer pools.

The following considerations apply to both buffers managed by the Fast Path 64-bit buffer manager and Fast Path buffers that are not managed by the Fast Path 64-bit buffer manager.

• An NBA value that is too large can increase the probability of contention (and delays) for other transactions. All CIs can be acquired at the exclusive level and be kept at that level until the buffer stealing facility is invoked. This occurrence happens after the NBA limit is reached. Therefore, an NBA that is too large can increase resource contention.
• A (NBA + OBA) value that is too small might result in more frequent unsuccessful processing. This means an 'FR' status code condition for BMP regions, or transaction abend for MD and MPP regions.
• Inquiry-only programs do not make use of an OBA specification, as buffers already allocated are reused when the NBA limit is reached.
• IMS logs information about buffers and their use to the X'5937' log. This information can be helpful in determining how efficiently the Fast Path buffers are being used.

The following considerations apply only when Fast Path buffers are not managed by the Fast Path 64-bit buffer manager.

• An incorrect specification of DBBF (too small) can result in the rejection of an area open or a region initialization. The system calculates the size of the buffer pool and rejects the open or initialization if the actual DBBF value is smaller.
• A DBFX value that is too small is likely to cause region waits and increase response time.
• An NBA value that is too small might cause the region processing to be serialized through the overflow buffer latch and again cause delays.