Handling 5000 websocket connections with axum server

[sumit-kumar-iitm]

I am using axum server for handling websocket connections. I am reading messages from kafka topic and if the websocket connection is established to receive messages from a particular session then the message will be relayed on to that connection.
I am facing issues when I am load testing the application for 5000 connections (4CPU and 8GB RAM). I am getting i/o timeout error on client side.
I am benchmarking this service against Java spring boot but java service is working fine with very low latency and the configuration is same.

I am sharing the code snippet, though I am using the standard code from example

lazy_static! {
    pub static ref ACTIVE_SESSIONS: DashMap<String, SplitSink<WebSocket, axum::extract::ws::Message>> =
        DashMap::new();
}

match TcpListener::bind(&ADDR).await {
        Ok(listener) => {
            let local_addr = listener.local_addr().expect("Failed to get local address");
            tracing::info!("server is running on port {}", local_addr.port());
            axum::serve(listener, app.into_make_service())
                .tcp_nodelay(true)
                .await
                .expect("Server failed to start");
        }
        Err(e) => {
            tracing::error!("Failed to initiate the listener: {}", e);
            std::process::exit(1);
        }
    }

the handler function of websocket is adding entry to ACTIVE_SESSIONS. On the other hand, kafka consumer consumes and check if there is a message for an already established connection and send it to that connection.
Can I get some help how to make it better??