Impact of processing models on throughput and latency