2 min read
Microsoft Azure is pleased to announce a series of performance optimizations supporting the latest distributions of Linux (Ubuntu, Red Hat, CentOS) and Windows for all virtual machine (VM) sizes, providing up to 25 Gbps of networking throughput. 25Gbps bandwidth is currently the fastest published speed between VMs in the public cloud.
These optimizations have been deployed to the entire Azure computing fleet in coordination with the latest Linux operating systems being published to the Azure Marketplace. Other popular Linux operating systems plan to incorporate these optimizations through their regular updates.
In order to help our customers architect high performance solutions, Azure is also publishing the expected network performance for VMs on our website. This will make it easier for our customers to reduce solution costs while maintaining optimal performance. See expected network performance per VM size for further information.
Some examples of today's expected performance metrics are highlighted below:
|vCPU||Memory: GB||Local SSD: GB||Max data disks||Max cached and local disk throughput: IOPS / MBps (cache size in GB)||Max uncached disk throughput: IOPS / MBps||Max NICs / Expected network performance: Mbps|
|Standard_M64ms||64||1792||2048||32||80,000 / 800 (6348)||40,000 / 1,000||8 / 16000|
|32||448||896||64||160,000 / 1,600 (4,224)||80,000 / 2,000||8 / 20000|
|20||140||280||40||80,000 / 640 (720)||64,000 / 960||8 / 20000|
|128||2048||4096||64||160,000 / 1,600 (12,696)||80,000 / 2,000||8 / 25000|
Maximize your VM’s Performance with Accelerated Networking (AN) – now widely available for 8+ core virtual machines
We are also pleased to announce the General Availability (GA) of Accelerated Networking (AN) for Windows with an expanded Public Preview for Linux.
AN provides very low latency and jitter networking performance via Azure's in-house programmable hardware and technologies such as SR-IOV. Also, by moving much of the SDN stack into hardware, compute cycles are reclaimed by end user applications putting less load on the VM.
As an example of AN’s performance advantage, a growing number of Azure SQL services in production today have achieved an amazing 70% improvement in several benchmarks. This clearly demonstrates real world performance benefits for customers looking to run latency sensitive workloads in the cloud.
AN for Linux continues its Public Preview and is supported by the latest operating systems published in the Azure Marketplace (Ubuntu, Red Hat, CentOS, and SLES). This preview is quickly expanding to more regions over the coming weeks. Additional operating systems such as FreeBSD will be supported with updates coming soon.
Instructions on how to sign up and participate in Azure's AN Linux Public Preview are found here, Accelerated Networking Overview and Deployment Instructions, and are supplemented by a list of current limitations found here, Accelerated Networking Service Update.
Expected network performance is the maximum aggregated bandwidth allocated per VM size across all NICs for all destinations. Upper limits are not guaranteed, but are intended to provide guidance for selecting the right VM size for a specific application. Actual network performance will depend on a variety of factors including number of TCP connections, network congestion, application workloads, and network settings. For more information on optimizing network throughput, see Optimizing Network Throughput for Linux and Windows. To achieve network performance on Linux or Windows VMs, it may be necessary to select specific Azure recommended versions. To produce comparable results see How to Reliably Test for Virtual Machine Network Performance.