Customers and partners regularly ask me questions around capacity of traffic on their Oracle Utilities products implementations and how to best handle their expected volumes.
The key to answering this question is to under a number of key concepts:
- Capacity is related to the number of users, threads etc, lets call them actors to be generic, are actively using the system. As the Oracle Utilities Application Framework is stateless, then actorsonly consume resources when they are active on any part of the architecture. If they are idle then they are not consuming resources. This is important as the number of logged on users does not dictate capacity.
- The goal of capacity is to have enough resource to handle peak loads and to minimize capacity when the load drops to the minimum expected. This makes sure you have enough for the busy times but also you are not wasting resource.
- Capacity is not just online users it is also batch threads, Web Service Clients, REST clients and mobile clients (for mobile application interfaces). It is a combination for each channel. Each channel can be monitored individually to determine capacity for each channel.
This is the advice I tend to give customers who want to monitor capacity:
- For channels using Oracle WebLogic you want to use Oracle WebLogic Mbeans such as ThreadPoolRuntimeMbean (using ExecuteThreads) for protocol level monitoring. If you want to monitor each server individually to get an idea of capacity then you might want to try ServerChannelRuntimeMBean (using ConnectionsCount). In the latter case, look at each channel individually to see what your traffic looks like.
- For Batch, when using it with Oracle Coherence, then use the inbuilt Batch monitoring API (via JMX) and use the sum of NumberOfMembers attribute to determine the active number of threads etc running in your cluster. Refer to the Server Administration Guide shipped with the Oracle Utilities product for details of this metric and how to collect it.
- For database connections, it is more complex as connection pools (regardless of the technique used) rely on a maximum size limit. If this limit is exceeded then you want to know of how many pending requests are waiting to detect how bigger the pool should be. The calculations are as follows:
- Oracle WebLogic JDBC Data Sources: Use the Mbean JDBCDataSourceRuntimeMBean with the CurrCapacity + WaitingForConnectionCurrentCount attributes.
- Oracle Universal Connection Pool (12.x): Use the UCP MBean UniversalConnectionPoolMBean with the (maxPoolSize– availableConnectionsCount) + PendingRequestsCount attributes.
Note: You might notice that the database active connections are actually calculations. This is due to the fact that the metrics capture the capacity within a limit and need to take into account when the limit is reached and has waiting requests.
The above metrics should be collected at peak and non-peak times. This can be done manually or using Oracle Enterprise Manager.
Once the data is collected it is recommended to be used for the following:
- Connection Pool Sizes– The connection pools should be sized using the minimum values experienced and with the maximum values with some tolerances for growth.
- Number of Servers to setup– For each channel, determine the number of servers based upon the numbers and the capacity on each server. Typically at a minimum of two servers should be setup for the minimum high availability solutions. Refer to Oracle Maximum Availability Architecture for more advice.