User Profile

I have an system that is designed around a collection of devices (iPhones or iPads) discovered via bonjour and connected with a NWConnection over TCP. Commands issued from one of the devices are sent to each of the peers and should be executed as soon as they are received. The problem I am encountering is a high variability in transit time device to device that I am having a hard time accounting for. By 'high' I am noting anywhere from 20-80ms of latency device to device. CPU utilization on each iPhone is essentially 0. Keepalives are enabled and firing off every 2 seconds. Additionally, the physical devices all have bluetooth off (as recommended in other posts) The interesting part is when I add into the connection mix an iPhone simulator (running on either a M1 MacBook Air, or my M1U Studio). When a command is issued from the simulator instance, all connected devices report back anywhere within ~0-3ms of deviation from the initiator, which is more what I expect from the network. Thinking that it's perhaps the M1 series of chips being far and away more competent than the A15's in the iPhone 13's and 14 that are in my testbed, I added my M1 iPad to the mix. Invoking a command from the iPad has similar variability as invoking it from one of the iPhones. The code is stupid simple and I'm posting here prior to opening up a DTS case in the hope that there's a magic "shouldUseSpeedholes=true" flag I can set. I have gone through several variations: using UDP instead of TCP (worse variations), changing from a listener/browser on each device to a single browser, multiple listener (no difference), changing from keeping things on the main queue (as in the docs) to a separate concurrent high priority dispatch queue (no difference). There is no TLS in the mix. I have tried both allowing peer-to-peer as well as not (no difference). I'm even using a single purpose project instead of my main codebase to isolate everything else that could be messing with scheduling with communications. I've tried each band of my WIFI (2.4 and 2x5ghz SSIDs) - no change. Sending func send(option: AppFramingOptions, withData data: Data?) { let message = NWProtocolFramer.Message(appFramingOption: option) let context = NWConnection.ContentContext(identifier: "\(option.rawValue)", metadata: [message]) for (uniqueKey, connection) in connections { if uniqueKey.contains(serviceName) { connection.send(content: data, contentContext: context, isComplete: true, completion: .idempotent) } } } In the above function, I am looking for the serviceName because I want to use connections connected via the browser as opposed to the listener (which isn't tagged with service name info in the endpoint). The check avoids a device receiving the command twice. Receiving connection.receiveMessage { content, contentContext, isComplete, error in guard error == nil else { connection.cancel() return } if let msg = contentContext?.protocolMetadata(definition: AppFraming.definition) as? NWProtocolFramer.Message { switch msg.appFramingOption { default: self.messageReceivedHandler?(content, msg.appFramingOption, Date().timeIntervalSince1970, withUniqueKey) } } receiveMessage() } } It's very much patterned off the TicTacToe example code (with a mechanism for multiple connections). My next step is embedding a web server in each device and making REST calls rather than commands over a TCP stream (which is CRAZY INSANE I KNOW). I also do not want to have to have a Macintosh dependency for this system because I cannot get predictable(ish) transit times. Any help is appreciated!