(Part 4) Testing the limits of Windows Azure Service Bus

In the last post related to this subject we discover how we can process millions of messages over a network using Windows Azure Service Bus Topic. For the problem that we want to resolve we find out that 4 worker roles of Medium size is the perfect configuration for our case.
In this post we will talk about costs. We will see what the costs to process 10.000.000 messages are. We assume that messages are send to Service Bus from our on-premise servers. If the messages are send from our the same data center, the cost related to bandwidth would be 0.
Sending messages to Service Bus

• Sending messages: 10$
• Bandwidth cost: 27.46$ 

Receive messages from Service Bus

• Receiving messages: 10$
• Bandwidth cost: 0$ (we are in the same data center)

Worker role costs:

• 4 medium worker roles: 8.64$

Remarks: From our results, we need 8h and 40 minutes to process 10.000.000 messages from Service Bus. In this time we included the warm up of each instance – 20 minutes. We calculate the cost of running 4 instances for 9 hours.
Windows Azure Table Storage:

• Storage: 5$ (cost per month)
•         Transactions:  2$

Remarks: In the transaction cost is included the cost of reading other data from Table Storage. Otherwise the cost of transaction to Table Storage would be maximum 1$ - because of batch support this value can be smaller.
The final cost of processing 10.000.000 messages and store them to Table Storage for 1 month is:

• 63.1$ (when messages are send from on-premise servers)
• 35.64$ (when messages are send from the same data center)

In theory, using this configuration, the price of each message processing would be 0.0003564$. What do you think about this cost? I would say that this is a pretty good price.
If we would need to process 10.000.000 every day, during a month, our final cost would be:

• 924.2$ (when messages are send from the same data center)

In this post we saw what the costs of processing 10.000.000 messages are. Even if the costs look pretty good, I still have one thing that I don’t like - the time processing of all messages is pretty high – over than 1 hour and a half. In the next post we will see what we can do decrease the time processing.