@kshegunov said:

I advise benchmarking the code. For one a datagram will always arrive whole (not like with TCP where you can have the message chopped in pieces), but there's a maximum size to it. Also there's no reception notification (like with TCP) as UDP is a "connectionless" protocol. These two things in mind you either send the datagram or not, you can't have anything else returned for the written bytes.

I suspected as much. Good to know.

QByteArray datagram; // ... Insert size checks here as well! PendingCompressedFrame & frame = *reinterpret_cast<PendingCompressedFrame *>(datagram.data()); // No copying, valid while the byte array is still there.

As of yet, the receiver's processing time hasn't been a bottleneck so I haven't looked into optimizing it (it completes in around 0-1ms). Thank you for the suggestion though.

Additionally, why the mutex? What are you protecting with it exactly?

The mutex is not needed there. There's some other data that is being accessed from a DLL from a separate thread, but I should have removed the mutex from the simplified receiver.

One good way to deal with such throughput is to have a dedicated thread for the reception that'll put the received datagrams (no processing) into a thread safe queue (or you can also use a signal-slot connection). From there you can have another thread that orders them (they may not come in the order of sending) and ultimately preparing them for display.

That's a good idea. Keep readyRead() as minimal as possible to get data out of the socket's buffer, allowing more data in.

The biggest question I'm still wondering is the performance of writedatagram. I've been looking into the assorted socket options, but it will take some reading to really understand what's going on with each flag.