Home > Blog > For Architects
Customers in industries like finance, telecommunications, and e-commerce demand messaging solutions that can sustain over a million guaranteed messages per second with exceptionally low latency. Achieving such performance in on-premises environments is challenging enough (often requires purpose-built appliances), but doing so in the cloud presents a whole new set of challenges.
In this blog post, we’ll explore how we fine-tuned PubSub+ Event Broker to push the boundaries of performance on Microsoft Azure Boost-enabled virtual machines. Our journey uncovered critical insights into optimizing network, storage, compute, and configuration to deliver the sustained throughput customers demand — all in a cloud-native setup.
Subscribe to Our Blog
Get the latest trends, solutions, and insights into the event-driven future every week.
Thanks for subscribing.
The Challenge: Sustaining Extreme Throughput in the Cloud
High-throughput event messaging in the cloud requires a combination of:
- Sufficient network bandwidth to handle massive data flow.
- Low-latency, high-IOPS storage to support guaranteed delivery.
- Efficient CPU utilization to avoid bottlenecks during intense workloads.
To meet these requirements, we tested Solace PubSub+ Event Broker on Azure Ebdsv5 virtual machines with Azure Boost, a technology that offloads virtualization processes onto specialized software and hardware. Our testing revealed the potential of this configuration to support sustained messaging at scale.
Foundations for Cloud-Native Performance
Our performance exploration focused on optimizing three key components: VMs, disk configuration, and HA architecture.
Virtual Machines
We used the Ebdsv5 series, leveraging Azure Boost to achieve:
- 40,000 Mbps network bandwidth for high throughput.
- 400,000 IOPS and 10,000 MBps disk throughput, enabled by NVMe controllers.
Testing spanned VM sizes from E16bds_v5 (16 vCPUs) to E112ibds_v5 (112 vCPUs).
Disk Configuration
Solace message spool storage requires a high-performance disk to meet guaranteed delivery demands. We tested:
- Premium SSD v2: A cost-effective option with tunable IOPS and throughput.
- Ultra Disks: For extreme performance, configured to 300,000 IOPS/10,000 MBps.
High-Availability Architecture
A Solace PubSub+ redundancy group (active/standby brokers with monitoring, see diagram below) was deployed across three availability zones in Azure’s East US 2 region to ensure resilience.
Testing Methodology
We tested both 1 KB and 1 MB guaranteed delivery messages with a 1:5 fan-out pattern. 40 publishing and 200 subscribing clients were distributed across 40 Azure Kubernetes Service (AKS) worker nodes to generate sufficient message load, simulating real-world scenarios. (We grouped the publishers and subscribers together for ease of deployment.)
Results: Breaking Throughput Barriers
The tests showcased the scalability and efficiency of Solace PubSub+ on Azure.
Here’s how different configurations performed:
VM Type | Specifications | Disk |
1 KB msg/s Ingress |
1 KB msg/s Egress |
1 MB msg/s Ingress |
1 MB msg/s Egress |
E112ibds_v5 |
112 vCPUs |
Ultra |
225,000 | 1,125,000 | 480 | 2,400 |
E64bds_v5 |
64 vCPUs |
Ultra |
222,000 | 1,110,000 | 373 | 1,865 |
E32bds_v5 |
32 vCPUs |
Premium SSD v2 |
217,000 | 1,085,000 | 300 | 1,500 |
E16bds_v5 |
16 vCPUs |
Premium SSD v2 |
186,000 | 930,000 | 240 | 1,200 |
Key Insights and Optimizations
Compute
- Idle CPU cores: Larger VMs often had idle CPU cores, suggesting limited impact from additional vCPUs.
- Hyperthreading: Enabling/disabling hyperthreading made no measurable difference.
Disk Performance
- A dedicated NVMe data disk for Solace message spool storage was crucial.
- Premium SSD v2 offered the best cost-to-performance ratio, while Ultra disks were more suitable for extreme edge cases.
Network Configuration
- Message throughput scaled with VM network bandwidth, especially for larger messages.
- Cross-zone deployments showed no performance degradation, enabling enhanced resilience without sacrificing throughput.
Solace-Specific Tuning
Adjustments to Solace broker and client settings unlocked higher performance.
- Broker optimizations: Increased max connections and file limits.
- Pub/sub settings: Adjusted window size and acknowledgment thresholds for efficient message flow.
Conclusion: Cloud-Native Scalability with Solace on Azure
With a little fine tuning, we demonstrated how Solace PubSub+ on Azure Boost-enabled VMs delivers exceptional performance for cloud-based event-driven architectures. Whether processing millions of financial transactions or scaling e-commerce platforms during peak seasons, this solution can meet the demands of even the most performance-critical use cases.
Are you ready to unlock high-throughput messaging in the cloud?
Reach out to explore how Solace can power your event-driven system on Azure.
Explore other posts from categories: For Architects | For Developers
Emerson Chen is a senior solutions architect at Solace, with over 35 years of experience in middleware and real-time integration. Specializing in event-driven architecture, digital transformation, and enterprise integration, Emerson empowers organizations to unlock the full potential of their technology. With a proven track record of leading diverse teams, he has successfully delivered cutting-edge middleware solutions across industries such as capital markets, energy, and commercial lending.
Marin Marinov is a senior technical specialist at Microsoft, specializing in Microsoft Azure cloud infrastructure and high-performance computing. He advises large enterprises on Azure environments design, migration, and solutions architecture throughout their cloud transformation journey.
Subscribe to Our Blog
Get the latest trends, solutions, and insights into the event-driven future every week.
Thanks for subscribing.