Blue Skies

Network Throughput Optimization

Network throughput is a critical factor in the performance of many Linux-based applications, especially those that handle large volumes of data or require high-speed data transfer. Azure provides several techniques to optimize network throughput for Linux VMs, ensuring that applications can handle their workloads efficiently.

One key technique is to select VM sizes that are optimized for network performance. Azure offers specialized VM sizes, such as the H-series for high-performance computing or the D-series for general-purpose workloads, that provide enhanced network capabilities. These VMs often come with higher network bandwidth limits and are ideal for applications requiring substantial data movement.

Another throughput optimization technique is to ensure that the underlying storage infrastructure is not a bottleneck. The use of Azure Premium SSDs or Ultra Disks can importantly improve data read/write speeds, which directly affects the network performance of data-intensive applications. Combining these techniques with Accelerated Networking can lead to substantial performance gains.

Configuring the Linux VM's network stack to handle high throughput, which includes tuning TCP settings such as window size, buffer sizes, and congestion-control algorithms, is also important. Tools such as sysctl can be used to adjust these parameters, allowing the Linux kernel to manage larger data flows more effectively, for example:

sysctl -w net.core.rmem_max=16777216
sysctl -w net.core.wmem_max=16777216
sysctl -w net.ipv4.tcp_window_scaling=1

These adjustments help in maximizing the throughput by allowing the Linux VM to process more data packets simultaneously, reducing delays caused by network congestion.

Latency Minimization Strategies

Minimizing latency is important for latency-sensitive Linux applications, such as those used in real-time data processing, financial transactions, or interactive user interfaces. Azure offers several strategies to reduce network latency, ensuring that Linux workloads remain responsive and efficient.

One of the primary strategies is to use Accelerated Networking, which, as mentioned earlier, reduces network latency by offloading processing to the NIC hardware. This method is particularly effective in scenarios where even microseconds of delay can have an important effect on application performance.

Keeping resources close to each other geographically reduces the physical distance data must travel, thereby minimizing latency. Additionally, the use of Azure proximity placement groups can further reduce latency by ensuring that VMs are physically located close to one another within the same data center.

For applications that require global reach, Azure Traffic Manager can be used to direct users to the nearest Azure region by network performance metrics. This approach helps minimize latency for end users by routing their requests to the closest available resource, reducing the time it takes for data to travel across the network.

Finally, optimizing the Linux kernel's networking stack for low latency is important and includes disabling features that introduce unnecessary processing delays, such as TCP slow start and tuning the interrupt coalescing settings on the NIC to reduce the time it takes to process network packets. Tools like ethtool,

ethtool -C eth0 rx-usecs 0

adjust these settings.

Summary

In this article, I explored the critical components of networking for Linux workloads on Azure, from architecting advanced VNet designs and customizing configurations for optimal IP management and routing to securing environments with NSGs and advanced firewall configurations.

The Author

Marcin Gastol is a Senior DevOps Engineer and Microsoft Certified Trainer with extensive experience in Azure technologies. He has taught various IT subjects, hosts a blog that addresses multiple IT areas (https://marcingastol.com/), and speaks at various conferences.