Mixed Workload Design Priority Guidelines White Paper

Hewlett-Packard Company—528875-001

Mixed Workload Design

Priority Guidelines

Summary

This white paper briefly describes the high level design for the Mixed Workload Design and presents some

general guidelines to control mixed workload applications using priority settings.

MWE Basic Design

The HP NonStop Kernel operating system uses a priority-based design to control process dispatch. The

highest priority executable process on the ready-list is dispatched.

Applications rely on a variety of system services, including communication, security, transaction, and data

services. These services are typically provided using a client-server model, using non-stop process pairs.

The heart of the database engine is the data access manager, more commonly referred to as DP2.

By definition, DP2 runs at system priority (220), which ensures that the operating system and transaction

services function properly.

Definition: Priority Inversion

When a service process executes at system priority, a low priority transaction can utilize high

priority services.

Low priority workload requests for services can delay service for higher priority workloads. This

behavior is termed “priority inversion”.

To prevent priority inversion, the service process must not service lower priority requests while

higher priority workloads are waiting for any resources, AND must preempt active service when

higher priority workloads require new resources.

The use of message priority by the high priority system service (DP2) to prevent priority inversion includes

both CPU and I/O resource contention. The methods used to resolve this contention include the following:

1. CPU context switch, which compares the priority of the currently active request with the priority

of processes on the ready-list. This is used to resolve conflicts for the CPU resource.

2. Request context switch, which compares the priority of the currently active request with the

priority of the message queue. This is used to resolve conflicts for the data services.

3. Reduction in query read ahead, which compares the priority of the currently active request with

the priority of any active disk I/O. This is used to resolve conflicts for I/O services.

4. Query deferral, which compares the priority of the current active request with the priority of any

active disk I/O. This is used to resolve conflicts for I/O services when parallel query workloads

are active.

PAGE 1
Mixed Workload Design Priority Guidelines Summary This white paper briefly describes the high level design for the Mixed Workload Design and presents some general guidelines to control mixed workload applications using priority settings. MWE Basic Design The HP NonStop Kernel operating system uses a priority-based design to control process dispatch. The highest priority executable process on the ready-list is dispatched.
PAGE 2
Contention for CPU Resources Various techniques are used to ensure high priority workloads are not impacted by CPU intensive low priority workloads such as SQL query activity. The message queue for DP2 services is priority ordered, and the DP2 process will wait on the ready-list at the request priority until it reaches the head of the ready-list. Once DP2 is dispatched, it will process the request at the normal DP2 priority (220).
PAGE 3
Priority Guidelines Prior to determining guidelines for setting priorities of SQL queries, it is necessary to measure and establish response-time requirements for high-priority transactions. It is then possible to fine-tune the priority of SQL query workloads to determine the best priority. If response time requirements cannot be measured or established, then the only recourse is to run the SQL query at a very low priority to ensure there is no impact to higher priority workloads.