Domains of workload management

In a data server environment, you must also define goals. Sometimes the goals are clear, especially when they originate from service level agreement (SLA) objectives. For example, queries from a particular application can consume no more than 10% of the total processor resource. Goals can also be tied to a particular time of day. For example, an overnight batch utility might have to complete loading data by 8 a.m. so that the daily sales reports are on time. In other situations, the goals can be difficult to quantify. A goal might be to keep the database users satisfied and to prevent aberrant database activity from hampering their day-to-day work. Whether the goals are quantifiable or not, understanding them is critical when considering the following stages of workload management:

Identification

If you want to achieve a goal for some kind of work, you first must be able to identify details about the work. In the grocery store, you can identify shopper information through credit cards and debit cards, or an unpaid-for item through an active security tag on the item. For the data server, you need to decide how you want to identify the work that enters the system. You can use the name of the application that submits the work, the authorization ID that submits the work, or a combination of elements that provide some form of identification.

Management

The management phase includes mechanisms for making steady progress towards your goal, and actions to take if a goal is not being met. An example of a mechanism is managing price checks in fast checkout lanes. Fast checkout lanes should result in faster throughput and satisfied customers, but if a carton of milk has the wrong price and a price check is required, the fast checkout lane could slow down. The problem is managed by performing a fast price check, possibly opening up another checkout lane, and trying to fix the pricing problem so that it does not occur again. On the data server, you might find that overall performance is suffering when a few poorly written SQL statements are running, a surge in volume occurs during peak times, or there is too much competition between different applications for the same resources. The management phase includes mechanisms for assigning resources to achieve your goals and actions to take if a goal is not being met. These workload management mechanisms, which indirectly and directly control CPU resources, include the following:

1. Concurrency thresholds, applied with a work action set defined on the workload, to control the concurrency of incoming work
2. The ability to move work form one service class to another, currently only effective when workload management dispatcher is being used on those service classes
3. Workload management dispatcher to specifically allocate CPU resources for workloads assigned to service classes for more granular control of CPU resources when the first two workload management mechanisms are not adequate

Monitoring

Monitoring is important for a couple of reasons. First, to determine whether you are achieving a goal, you must have a mechanism to track progress toward that goal. Also, monitoring helps to identify the problems that might be preventing you from achieving your goal. In a store, the store manager can watch the flow of customers, automatically be alerted to problems such as shoplifting or dangerously low inventory of a particular sale item, or perform analysis on historical purchase patterns to determine optimal product placement in the store. For a data server, there are often explicit goals for response times of database activities and it is important to have a way to measure this metric, and watch for trends.