Maximum Potential Concurrency On Azure Function Triggered By Service Bus Queue - azure

Question

We are using an azure function that is triggered by a service bus queue.
Our function's host.json is as follows:
{
"version": "2.0",
"extensions": {
"serviceBus": {
"prefetchCount": 0,
"autoCompleteMessages": true,
"maxAutoLockRenewalDuration": "00:05:00",
"maxConcurrentCalls": 16,
"maxConcurrentSessions": 2000,
"maxMessages": 1000,
"sessionIdleTimeout": "00:01:00",
"enableCrossEntityTransactions": false
}
},
"logging": {
"applicationInsights": {
"samplingExcludedTypes": "Request",
"samplingSettings": {
"isEnabled": true
}
},
"logLevel": {
"Function.Delegation.User": "Information",
"Function": "Error"
}
}
}
We also have an exponential retry in case messages fail, with infinite retry as we cannot afford to lose any data:
[FunctionName("Delegation")]
[ExponentialBackoffRetry(-1, "00:00:01", "00:01:00")]
public async Task Run([ServiceBusTrigger(DelegationConstants.PlotQueueName, Connection = "ServiceBusConnectionString", IsSessionsEnabled = true)] string myQueueItem, MessageReceiver messageReceiver, string lockToken, string messageId, ILogger log)
Our service bus namespace has a single queue that is triggering this. The queue has sessions and partitions enabled and we currently have a 1GB size across 16 partitions (16GB). Our partition key is a physical device identifier (an IMEI on a mobile device if you are familiar), so it has a very broad range of values (around 55k in total in our estate).
Our service bus message is created with a session id of the partition key (IMEI):
var serviceBusMessage = new ServiceBusMessage(JsonConvert.SerializeObject(delegationMessage))
{
SessionId = delegationMessage.PartitionKey
};
Each of our function invocations take ~200-300ms:
Messages 'per device' must be processed FIFO, hence the sessions. At any one time, we could need up to (and in future possibly more than) 1000 messages per second across many devices, each at 200-300ms.
We have about reached the maximum optimisation of our function code itself, so I can't make that any quicker unfortunately.
Given all the above information, our service bus queue is still increasing, indicating that perhaps the service bus + function setup we have right now is not sufficient to cope with our throughput. Is there anything I am missing here or have I reached the limit of concurrency with this configuration? Our function is running on a P1V3 and sits at ~45% cpu and ~25% memory