This paper presents three complementary components of the flow control solution adopted for the Fed-X fabric: high-speed scalable interconnect for a multi-computer system. Each of the three addresses performance problems caused by a particular characteristic of realistic network workloads. The three flow-control techniques introduced in this paper are backpressure flow control, alpha scheduling, and balanced injection. Our variation on backpressure flow control ensures that all of the buffers in a node are not filled with one particular category of packet, such as those destined for the same node, so that buffers are available for other packets. Alpha scheduling exploits the distinction between message latency and packet latency to maximize the effective performance of the interconnect from the perspective of applications. Balanced injection throttles the injection of traffic into a node from outside the network based on local congestion as measured by local buffer usage. These three flow-control techniques, together, provide a significant increase in fabric performance, both in sustainable throughput and in message latency, under a wide variety of realistic traffic patterns.