In the context of performance testing it is, the number of transaction, that is sequence of one request and one response, per unit time. Our goal always should be to increase the value of throughput, because then the system will be fast enough.
In general the throughput depends on various factors, like application server health, network latency and the work load applied on the system.
Another term very commonly use in the context of performance testing, along with Throughput is the Utilization.
Utilization is measures as the ratio of throughput of any application with respect to its maximum capacity. If the system utilization is very high, then the response time will become very slow. So, it’s always better to keep the utilization between 75%-80%.
In the below mentioned post, written by PerfMatrix, they have clearly explained the throughput-graph. Please take a look how it looks in real time.