Best practices for implementing distributed tracing with minimal overhead?

Hey everyone, I'm trying to set up distributed tracing for a new microservices project. I've looked at Jaeger and Zipkin, but I'm concerned about the potential performance impact, especially in high-volume services. What are some of your best practices for implementing tracing with minimal overhead? Are there specific sampling strategies or instrumentation techniques that you've found to be particularly effective without bogging down your services? I'm curious about both open-source and commercial solutions.