Zerotier slow between directly connected hosts


I have two VMs attached to the same vswitch on the same hypervisor. Iperf3 performance directly between the virtual NICs on these VMs is about 20Gbps in both directions. Running the same iperf3 test between the same two VMs over ZeroTier (instead of the directly connected NICs) yields throughput of about 500Mbps. During the test over ZeroTier, one of the CPU cores on the VMs goes up to about 90% and then goes back down to nothing once the test is done, leading me to believe that some part of ZeroTier is single threaded, which is causing the bottleneck.

I’m guessing it’s the encryption piece of ZeroTier that is single-threaded, as most types of encryption are. Are there any plans on multi-threading whatever part of ZeroTier is single-threaded, or similar plans to remove the bottleneck? ZeroTier is a great product. It would be even better if multi-gigabit speeds could be achieved like some other competing products are already doing.


Yes, the main communication loop in ZeroTier is currently single threaded. We are working on making it multithreaded in the future, but don’t currently have an ETA for when that will be shipped

This topic was automatically closed after 30 days. New replies are no longer allowed.