High Available

Auto connect/reconnect to proxy and any brother node only when need. Select the fastest way to communicate with each other automately. Less memory cost for even many server using shared memory.

High Performence

5000K+ QPS or 1.5+GB/s when using shared memory channel and about 2500K+ QPS or 600+MB/s when using tcp channel. See benchmark of libatbus for detail.

Decentralized

There is no any center or master server. So you can horizontal expansion any of your service group if you want.

Customizable

Open source under MIT License, so you can add any yours protocol or component.