In a batch processing system such as Hadoop, we usually care about throughput—the number of records we can process per second, or the total time it takes to run a job on a dataset of a certain size.iii In online systems, what’s usually more important is the service’s response time—that is, the time between a client sending a request and receiving a response.
