Congestion-aware load balancing in data center networks

An aggregate switch configured to perform congestion-aware load balancing in a data center network that includes a source switch and a destination switch, the aggregate switch, and multiple intermediate switches including a first intermediate switch and a second intermediate switch, wherein the aggregate switch comprises:

• a first port to receive a first probe packet from the first intermediate switch that connects the aggregate switch with the destination switch via a first path, wherein the first probe packet identifies the destination switch and includes first congestion state information having a first maximum link utilization level along a path from the first intermediate switch to the destination switch;
• a second port to receive a second probe packet from the second intermediate switch that connects the aggregate switch with the destination switch via a second path, wherein the second probe packet identifies the destination switch and includes second congestion state information having a second maximum link utilization level along a path from the second intermediate switch to the destination switch;
• and
• a third port to receive data packets from a source endpoint connected via the source switch that are destined for a destination endpoint connected via the destination switch, wherein the aggregate switch is configured to:
• in response to receiving the first probe packet via the first port, select the first intermediate switch from the multiple intermediate switches based on the first congestion state information, wherein the multiple intermediate switches connect the aggregate switch with the destination switch via multiple paths;
• after having selected the first intermediate switch and in response to receiving the second probe packet via the second port, compare the second maximum link utilization level and local congestion state information associated with an ingress port via which the second probe packet is received from the second intermediate switch to generate a first comparison result and compare the first comparison result and the first maximum link utilization level to generate a second comparison result;
• determine that a least congested path from the aggregate switch to the destination switch includes the second intermediate switch based on the second comparison result;
• update switch state information with the second intermediate switch and path utilization information associated with the destination switch;
• determine whether to send a third probe packet to the source switch, wherein the third probe packet identifies the destination switch and the path utilization information in the switch state information;
• and
• in response to receiving the data packets via the third port, send the data packets to the second intermediate switch such that the data packets travel to the destination switch along the least congested path.