Let’s talk about Windows Azure Service Bus and the numbers of subscriptions that a topic can have. I started to prepare a POC for a possible client and it is possible to end up with hundreds of subscriptions on each topic.
Looking over the documentation from MSDN, each topic supports 2000 subscriptions. I wanted to see what is happening with the latency of each subscription if we have over 1000 of them per topic.
To find an answer to my question, I started to write a worker role that:
Creates a new subscription
Sends a message to the topic
Receives a message using the first subscription
I measured how long it takes to send a message to a topic that has 1, 2, … , 2000 subscriptions. The same think I’ve done for the first subscription of the topic. I measured how long it takes to read a message from it when we have 1,2, … , 200 subscriptions.
The code is extremely simple but the results are very interesting.
The first diagram shows how long the send and receive commands took. The blue color is for the send request to the topic and the orange one is for receive command for the first subscription of the topic. The AX represent the number of subscriptions per topic and the OY represent how long a command took (in milliseconds).
As you can see, the command latency doesn’t increase too much. Even for 2000 subscriptions the latency is under 100 milliseconds.
The next diagram show the average latency when we have 500, 1000, 1500 and 2000 subscriptions. Based on this result we observe that the latency increases with a factor under 1.5X.
We can say that we can use this service in our business scenarios for 2000 subscribers without a problem. The latency increases only with a factor that is under 1.5X. We could have a problem if the latency would increase with a factor of 10X or 100X. But we are in a safe zone.
It seems that Service Bus is a real cloud service that scales and works excellent.
Looking over the documentation from MSDN, each topic supports 2000 subscriptions. I wanted to see what is happening with the latency of each subscription if we have over 1000 of them per topic.
To find an answer to my question, I started to write a worker role that:
Creates a new subscription
Sends a message to the topic
Receives a message using the first subscription
I measured how long it takes to send a message to a topic that has 1, 2, … , 2000 subscriptions. The same think I’ve done for the first subscription of the topic. I measured how long it takes to read a message from it when we have 1,2, … , 200 subscriptions.
The code is extremely simple but the results are very interesting.
The first diagram shows how long the send and receive commands took. The blue color is for the send request to the topic and the orange one is for receive command for the first subscription of the topic. The AX represent the number of subscriptions per topic and the OY represent how long a command took (in milliseconds).
The next diagram show the average latency when we have 500, 1000, 1500 and 2000 subscriptions. Based on this result we observe that the latency increases with a factor under 1.5X.
It seems that Service Bus is a real cloud service that scales and works excellent.
Comments
Post a Comment