Export (0) Print
Expand All

Best Practices for Queued Communication

This topic provides recommended practices for queued communication in Windows Communication Foundation (WCF). The following sections discuss recommended practices from a scenario perspective.

For scenarios that require separation that queued messaging provides and fast, high-performance messaging with best-effort assurances, use a non-transactional queue and set the ExactlyOnce property to false.

In addition, you can choose not to incur the cost of disk writes by setting the Durable property to false.

Security has implications on performance. For more information, see Performance Considerations.

The following sections describe recommended practices for scenarios that require end-to-end reliable messaging.

ms731093.collapse_all(en-us,VS.110).gifBasic Reliable Transfer

For end-to-end reliability, set the ExactlyOnce property to true to ensure transfer. The Durable property can be set to true or false depending on your requirements (the default is true). Generally, the Durable property is set to true as part of end-to-end reliability. The compromise is a performance cost, but makes the message durable so that the message is not lost if a queue manager crashes.

ms731093.collapse_all(en-us,VS.110).gifUse of Transactions

You must use transactions to ensure end-to-end reliability. ExactlyOnce assurances only ensure that messages are delivered to the target queue. To ensure that the message is received, use transactions. Without transactions, if the service crashes, you lose the message that is being delivered but is actually delivered to the application.

ms731093.collapse_all(en-us,VS.110).gifUse of Dead-letter Queues

Dead-letter queues ensure that you are notified if a message fails to be delivered to the target queue. You can use the system-provided dead-letter queue or a custom dead-letter queue. In general, using a custom dead-letter queue is best because it enables you to send dead-letter messages from one application into a single dead-letter queue. Otherwise, all dead-letter messages that occur for all applications running on the system are delivered to a single queue. Each application must then search though the dead-letter queue to find the dead-letter messages that are relevant to that application. Sometimes, using a custom dead-letter queue is not feasible, such as when using MSMQ 3.0.

Turning off dead-letter queues for end-to-end reliable communication is not recommended.

For more information, see Using Dead-Letter Queues to Handle Message Transfer Failures.

ms731093.collapse_all(en-us,VS.110).gifUse of Poison-Message Handling

Poison-message handling provides the ability to recover from the failure to process messages.

When using the poison-message handling feature, ensure that the ReceiveErrorHandling property is set to the appropriate value. Setting it to Drop means the data is lost. On the other hand, setting it to Fault faults the service host when it detects a poison message. Using MSMQ 3.0, Fault is the best option to avoid data loss and move the poison message out of the way. Using MSMQ 4.0, Move is the recommended approach. Move moves a poisoned message out of the queue so the service can continue to process new messages. The poison-message service can then process the poison message separately.

For more information, see Poison Message Handling.

To achieve high throughput on a single endpoint, use the following:

  • Transacted batching. Transacted batching ensures that many messages can be read in a single transaction. This optimizes transaction commits, increasing overall performance. The cost of batching is that if a failure occurs in a single message within a batch, then the entire batch is rolled back and the messages must be processed one at a time until it is safe to batch again. In most cases, poison messages are rare, so batching is the preferred way to increase system performance, particularly when you have other resource managers that participate in the transaction. For more information, see Batching Messages in a Transaction.

  • Concurrency. Concurrency increases throughput, but concurrency also affects contention to shared resources. For more information, see Concurrency.

  • Throttling. For optimal performance, throttle the number of messages in the dispatcher pipeline. For an example of how to do this, see Throttling.

When using batching, be aware that concurrency and throttling translate to concurrent batches.

To achieve higher throughput and availability, use a farm of WCF services that read from the queue. This requires that all of these services expose the same contract on the same endpoint. The farm approach works best for applications that have high production rates of messages because it enables a number of services to all read from the same queue.

When using farms, be aware that MSMQ 3.0 does not support remote transacted reads. MSMQ 4.0 does support remote transacted reads.

For more information, see Batching Messages in a Transaction and Differences in Queuing Features in Windows Vista, Windows Server 2003, and Windows XP.

In some scenarios a group of messages in a queue may be related and, therefore, the ordering of these messages is significant. In such scenarios, process a group of related messages together as a single unit: either all of the messages are processed successfully or none are. To implement such behavior, use sessions with queues.

For more information, see Grouping Queued Messages in a Session.

Though queues are typically one-way, in some scenarios you may want to correlate a reply received to a request sent earlier. If you require such correlation, it is recommended that you apply your own SOAP message header that contains correlation information with the message. Typically, the sender attaches this header with the message, and the receiver, upon processing the message and replying back with a new message on a reply queue, attaches the sender's message header that contains the correlation information so that the sender can identify the reply message with the request message.

Use MsmqIntegrationBinding when integrating WCF services or clients with non-WCF services or clients. The non-WCF application can be an MSMQ application written using System.Messaging, COM+, Visual Basic, or C++.

When using MsmqIntegrationBinding, be aware of the following:

  • A WCF message body is not the same as a MSMQ message body. When sending a WCF message using a queued binding, the WCF message body is placed inside of a MSMQ message. The MSMQ infrastructure is oblivious to this extra information; it sees only the MSMQ message.

  • MsmqIntegrationBinding supports popular serialization types. Based on the serialization type, the body type of the generic message, MsmqMessage<T>, takes different type parameters. For example, ByteArray requires MsmqMessage<byte[]> and Stream requires MsmqMessage<Stream>.

  • With XML serialization, you can specify the known type using the KnownTypes attribute on the <behavior> of <serviceBehaviors> element that is then used to determine how to deserialize the XML message.

Show:
© 2014 Microsoft