Network stacks today are fundamentally unreliable and opaque. This has two key implications for distributed applications and disaggregated architectures. First they must choose between performance and reliability for network interactions. Second they must embrace all the operational complexity inherent in debugging performance issues stemming from the network.
In Gazelle, we aim to provide distributed applications and disaggregated architectures with an Ethernet network stack that:
- is as reliable as the underlying network hardware
- can be profiled and analyzed by always-on automated tooling