E
EldertGrootenboer
In 2016 we announced the general availability of Service Bus Premium, which offers dedicated resources and consistent performance for mission-critical workloads. At this point we also published a blogpost that showed the results of the performance benchmarks we ran on this SKU.
Since then, we have made a variety of improvements to Service Bus Premium, providing even better performance for our customers’ workloads. In this blogpost, we will revisit the performance benchmarks from 2016 and compare them with the current performance of Service Bus Premium in 2024.
We ran this benchmark in the same configuration as for the previous blogpost. We used a performance benchmark tool, which can be found on GitHub. The benchmark tool was run on D48 VMs located in the same region as the namespace.
The benchmarks tested three scenarios: a single queue, a single topic with a single subscription, and a single topic with five subscriptions. The tests used 1, 2, and 4 messaging units (MU), which are the units of scale and performance in the premium SKU. Furthermore, the tests used 1 KB message size, the AMQP protocol, and implemented batching.
For each test we measured the average throughput (messages per second) for our send and receive operations.
Below are the results for the send and receive operations, and how these compare to those found in 2016. This indicates the number of messages that came through Service Bus per second, so higher numbers mean better performance.
Below are the numbers for each of these tests, and the improvement in percentages compared to the same tests in 2016.
Below are the numbers for each of these tests, and the improvement in percentages compared to the same tests in 2016.
These results show that there has been a significant performance improvement since we first introduced Service Bus Premium, with an average increase of ~50%, and even over 150% for some scenarios. Of course, we are going to continue improving our service, where performance is always on top of mind.
Continue reading...
Since then, we have made a variety of improvements to Service Bus Premium, providing even better performance for our customers’ workloads. In this blogpost, we will revisit the performance benchmarks from 2016 and compare them with the current performance of Service Bus Premium in 2024.
Methodology
We ran this benchmark in the same configuration as for the previous blogpost. We used a performance benchmark tool, which can be found on GitHub. The benchmark tool was run on D48 VMs located in the same region as the namespace.
The benchmarks tested three scenarios: a single queue, a single topic with a single subscription, and a single topic with five subscriptions. The tests used 1, 2, and 4 messaging units (MU), which are the units of scale and performance in the premium SKU. Furthermore, the tests used 1 KB message size, the AMQP protocol, and implemented batching.
For each test we measured the average throughput (messages per second) for our send and receive operations.
Results
Below are the results for the send and receive operations, and how these compare to those found in 2016. This indicates the number of messages that came through Service Bus per second, so higher numbers mean better performance.
Send
Below are the numbers for each of these tests, and the improvement in percentages compared to the same tests in 2016.
| # of MUs | Improvement % from 2016 to 2024 | |||
| Entity Type | Year | 1MU | 2MU | 4MU |
| Queue | 2016 | 4401 | 9070 | 16782 |
| 2024 | 6790 | 12770 | 25000 | |
| 1 Subscription | 2016 | 2223 | 4499 | 8960 |
| 2024 | 5000 | 10300 | 13200 | |
| 5 Subscriptions | 2016 | 1348 | 1676 | 3844 |
| 2024 | 1600 | 3100 | 4400 |
[td]
1MU
[/td]
[td]
2MU
[/td]
[td]
4MU
[/td]
1MU
[td]
2MU
[td]
4MU
[td]
54%
[/td]
[td]
41%
[/td]
[td]
49%
[/td]
54%
[td]
41%
[td]
49%
[td]
125%
[/td]
[td]
129%
[/td]
[td]
47%
[/td]
125%
[td]
129%
[td]
47%
[td]
19%
[/td]
[td]
85%
[/td]
[td]
14%
[/td]
19%
[td]
85%
[td]
14%
Receive
Below are the numbers for each of these tests, and the improvement in percentages compared to the same tests in 2016.
| # of MUs | Improvement % from 2016 to 2024 | |||
| Entity Type | Year | 1MU | 2MU | 4MU |
| Queue | 2016 | 4401 | 9069 | 16782 |
| 2024 | 6790 | 12770 | 25000 | |
| 1 Subscription | 2016 | 2220 | 4493 | 8956 |
| 2024 | 5000 | 10300 | 13200 | |
| 5 Subscriptions | 2016 | 2994 | 8331 | 19202 |
| 2024 | 8000 | 15300 | 22000 |
[td]
1MU
[/td]
[td]
2MU
[/td]
[td]
3MU
[/td]
1MU
[td]
2MU
[td]
3MU
[td]
54%
[/td]
[td]
41%
[/td]
[td]
49%
[/td]
54%
[td]
41%
[td]
49%
[td]
125%
[/td]
[td]
129%
[/td]
[td]
47%
[/td]
125%
[td]
129%
[td]
47%
[td]
167%
[/td]
[td]
84%
[/td]
[td]
15%
[/td]
167%
[td]
84%
[td]
15%
Conclusion
These results show that there has been a significant performance improvement since we first introduced Service Bus Premium, with an average increase of ~50%, and even over 150% for some scenarios. Of course, we are going to continue improving our service, where performance is always on top of mind.
Continue reading...