On July 15th, Microsoft announced the Azure Premium Messaging service has reached General Availability (GA). It has been ten months since Microsoft launched their public preview. InfoQ previously covered the Public Preview announcement in November 2015. Azure Premium Messaging is built on Microsoft's Platform as a Service (PaaS) messaging platform called Azure Service Bus. The premium service includes existing cloud-based messaging capabilities found in the Azure Service Bus Standard Messaging tier, such as Queues and Topics. However, Premium Messaging is not a multi-tenant messaging platform, like the Standard Messaging tier, and allows customers to take advantage of dedicated compute for their workloads. As a result of dedicated compute, the Service Bus team has increased the maximum message size that customers can process in the messaging service. Instead of the traditional 256KB limit, Premium Messaging supports messages up to 1MB.
Customer adoption of the service continues to accelerate. Justin Conway, program manager on the Azure Service Bus team, explains: “Premium Messaging has shown strong monthly growth with traffic volume nearly surpassing one billion operations per day. The general availability of Premium Messaging truly delivers on a state of the art messaging platform for those who want reliability, high throughput and an environment to call their own.”
Customer segments using the new service include financial services companies, pizza chains, insurance providers, media companies in addition to Microsoft services such as Dynamics CRM, Bing Maps, and analytics services.
Premium Messaging will be available in seven regions including: Southeast Asia, East Asia, West Europe, North Europe, Central US, West US, and East US. Additional regions will be added based upon customer demand.
InfoQ had a chance to discuss this release with Dan Rosanova, principal program manager on the Azure Service Bus team. Here is what he had to say:
InfoQ: A recent blog post written by your team mentions Premium Messaging traffic has nearly surpassed 1 billion operations a day. What is driving adoption for customers?
Dan Rosanova: Customers with sensitivity to latency or predictability tend to gravitate towards premium. Typical use cases are e-commerce and order processing where response times to a user is critical or assets and money are changing hands. We also have a group of ISV’s that need to provide a scaled platform for their customers who are adopting quickly.
InfoQ: For some customers, the current message size limitation (256KB) in Standard Messaging has become a constraint, in some scenarios. Why did you decide to allow larger messages in Premium Messaging and how did you come up with 1MB?
Rosanova: The primary concern for larger messages in a mutli-tenant platform like Standard Messaging is that you are taking away resources from the larger pool. Since Premium Messaging is not a multi-tenant platform we are not as concerned by this and since we no longer use SQL as a message store this also becomes less of a platform constraint. Each customer has their own message store. For message size, we were looking at a range of sizes. Our service is a messaging focused one; you have to decide if your scenario is really messaging or storage. In the future, we could raise the limit a bit higher, but not orders of magnitude. However, compared to other cloud messaging services 1MB is a fairly large message size.
InfoQ: Are there any best practices, or guidance, for customers who are interested in processing large messages in the new service?
Rosanova: Since this is a dedicated resource model, large messages will require more resources than small messages. This is true in our messaging service as well as your senders and receivers. The cost of serialization is heavier for your code than our code because we don’t open the message body. It’s important to be aware that the larger your messages become, the more CPU you will have to use in performing serialization and processing in your own code. Additionally, you should benchmark your service and namespace to see how it performs with these larger messages. The large messages will eat up IOPS and storage faster than small messages, but in Premium the performance will be consistent – always.
InfoQ: Your team recently published some performance statistics about the new service. Were you surprised by these numbers? Are there plans to increase the number of Messaging Units (MUs) that customers can provision for additional scale?
Rosanova: Coming from an on premise messaging background, durable messaging beyond a thousand messages per second was always fairly fast. So yes, we are pleased we are beyond that point. We did expect for the scale to be linear and the tests show it is. As for increasing the number of messaging units available, it’s always important to shard your workload at scale and we don’t want to encourage people to avoid making these important decisions. We feel that with 16,000 messages per second available, we should meet most customer demands, but are always looking for customer feedback. We are considering increasing available MUs, but we have no firm plans at this time.
InfoQ: At the recent World Partner Conference in Toronto, you shared that Event Hubs Dedicated Capacity is on its way. How will these two services be different?
Rosanova: We quietly released Event Hubs dedicated capacity in the late spring for EA customers. These are single tenant clusters of Event Hubs for very large and demanding workloads. Similar to Premium Messaging, but on a much larger scale, the pricing model is based on a fixed monthly price with no additional charges for usage. This does require a three month committed purchase and involvement with our product team. It is not a lightweight thing to just try out for a weekend. This really targets customers who need hundreds of thousands, or millions, of messages per second. We welcome anyone interested to reach out to their EA representative or Azure customer service.