Configure MTU size on Azure PaaS

Question

We're running a WCF based API on Windows Azure Web Roles in North Europe. Recently we have had issues with very unstable connections to a client running on AWS US East. A wireshark dump from our Azure instance showed that TCP packets was lost somewhere along the path with several TCP retransmission happening from our Azure instances.

AWS excellent support helped us nail down that this was probably being caused due to the Azure Instances use of Jumbo Frames with fragment sizes > 1500 bytes and the path MTU discovery not working properly. A note is that Usually the connections would get stuck already in the TLS handshake (our API runs over HTTPS) as the TLS Hello package would be around 5000 bytes.

My question is, what is the recommended way to configure the MTU size on Azure PaaS? Running an netsh command on the instances would not be persistent when the cloud fabric relocate instances etc.

Also, have anyone experienced similar issues? I'm worried about degraded network performance to other cloud resources (SQL Azure, Blob Storage, Service Bus) when lowering the MTU from the default.

Answer 1

Why don't you set your MTU config on a start-up task? That would run again on every start/reboot You can also check if the role wasn't moved, and ignore the script...

https://azure.microsoft.com/en-us/documentation/articles/cloud-services-startup-tasks-common/#detect-that-your-task-has-already-run

Jumbo Frames are not enabled by default, so you can run a PS script on start, and manually enable it.

Get-NetAdapterAdvancedProperty -DisplayName "Jumbo Packet" | Set-NetAdapterAdvancedProperty –RegistryValue "9014"

I hope it helps in some way.

Answer 2

Turns out the Azure instances wasn't really sending Jumbo Frames. The packages captured by Wireshark just indicates this because of TCP large Segment Offload . After disabling this, Wireshark indicates properly formatted TCP packets. Our hunt for the lost packages between Azure and AWS continues.