Question on supporting a socket-based transport implementation

[shenweihai1]

I’ve been using erpc in my project, and it has been working very well.

My workflow is: coding everything locally and then running it on Azure. What I'm doing is: using rdma link add type rxe command on Ubuntu to create a soft RoCE device; and then using Infiniband transport and RoCE enabled in eRPC. While this approach works, I find the soft RoCE setup somewhat inconvenient.

I was wondering if it would be feasible to implement a simple socket-based transport under transport_impl, which is widely supported by all new or old servers. So that

• I don't need to depend on anything else (e.g., rdma-core);
• I don't need to change any code in my project.

Also, it is just for development purposes; there are no performance considerations.

If this sounds reasonable, I’d greatly appreciate any insights on how to approach implementing this transport in eRPC. I’m happy to explore adding it myself.

---

The exposed interfaces would look like below (copy it from fake_transport.h)

{
  inline void init_hugepage_structures(HugeAlloc *, uint8_t **) {}
  inline void fill_local_routing_info(routing_info_t *) const {}

  inline bool resolve_remote_routing_info(routing_info_t *) { return false; }
  inline size_t get_bandwidth() const { return 0; }

  static std::string routing_info_str(routing_info_t *) { return ""; }

  void tx_burst(const tx_burst_item_t *, size_t) {}
  void tx_flush() {}
  size_t rx_burst() { return 0; }
  void post_recvs(size_t) {}
}