Network endpoint congestion management
A congestion management protocol that can be used for small messages in which the last-hop switch determines the congestion of the end point. The last-hop switch drops messages when the end point is congested and schedules a retransmission. A second congestion management protocol transmits small messages in a speculative mode to avoid the overhead caused by reservation handshakes.
1. A method for congestion management in a network, the method comprising the steps of:
(a) transmitting a message packet from a source to a destination, said message packet being a speculative packet, said message packet having a message packet size, the message packet size being smaller than or equal to a predetermined packet size for the respective network;
(b) receiving said message packet at a last-hop switch;
(c) determining, at said last-hop switch, a destination congestion level;
(d) upon determining, by said last-hop switch, that the destination congestion level exceeds a predetermined threshold, dropping, at said last-hop switch, said message packet;
(e) responsive to said step of dropping, generating, at said last-hop switch, both a negative acknowledgement and a reservation, the negative acknowledgement and the reservation transmitted simultaneously from said last-hop switch to said source, the negative acknowledgement and the reservation being a non-speculative packet;
(f) scheduling, at said source, a transmittal of the message packet; and
(g) transmitting, at a predetermined time, said message packet from said source to said destination, said message packet being a non-speculative packet.
2. The method of claim 1 , further including the step of:
repeating steps (a)-(g) a desired number of times.
3. The method of claim 1 , wherein the step of determining a congestion level further includes:
a. queuing said message packet at the last-hop switch;
b. determining a queuing level;
c. comparing the queuing level to a predetermined congestion metric.
4. The method of claim 3 , wherein the queuing level is a number of message packets in the respective queue and the predetermined congestion metric is a maximum number of message packets.
5. The method of claim 3 , wherein the queuing level is a first elapsed time measurement and the predetermined congestion metric is a second elapsed time measurement.
6. The method of claim 1 , wherein said message packet is transmitted at a first time, the negative acknowledgement is received at a second time, wherein the step of scheduling further includes
a. calculating a third time by subtracting said first time from said second time;
b. calculating a time differential by dividing said third time by two;
c. calculating the predetermined time by adding the reservation time to said time differential.
7. The method of claim 1 , wherein said message packet is transmitted at a first time, wherein the step of scheduling further includes calculating the predetermined time by adding the reservation time to said first time.
8. The method of claim 1 , wherein said message packet contains data.
9. The method of claim 1 , wherein said source is one of a plurality of sources, said destination is one of a plurality of destinations, said last-hop switch is one of a plurality of last-hop switches, each of the plurality of last-hop switches connected concurrently to more than one of the plurality of sources and more than one of the plurality of destinations.
10. The method of claim 1 , wherein the speculative packets use a first virtual channel and the non-speculative packets use a second virtual channel, said first virtual channel having a lower priority than said second virtual channel.
11. The method of claim 2 , and further comprising the step of:
reassembling, at said destination, a large message from a plurality of message packets.
12. A method for congestion management in a network, the method comprising the steps of:
(a) transmitting a message packet from a source to a destination, said message packet being a speculative packet, said message packet having a message packet size, the message packet size being smaller than or equal to a predetermined packet size for the respective network;
(b) receiving said message packet at one of a plurality of network switches;
(c) determining, at the one of a plurality of network switches, a destination congestion level;
(d) upon determining, at one of a plurality of network switches, that the destination congestion level exceeds a predetermined threshold, dropping, at the one of a plurality of network switches, said message packet;
(e) generating, at the one of a plurality of network switches, a negative acknowledgement, the negative acknowledgement transmitted to the source, the negative acknowledgement being a non-speculative packet;
(f) transmitting, from said source to said destination, a reservation, the reservation being a non-speculative packet;
(g) transmitting, from said destination to said source, a grant, the grant being a non-speculative packet;
(h) scheduling, at said source, a transmittal of said message packet; and
(i) transmitting, at a predetermined time, said message packet from said source to said destination, said message packet being a non-speculative packet.
13. The method of claim 12 , further including the step of:
repeating steps (a)-(i) a desired number of times.
14. The method of claim 12 , wherein the step of determining a congestion level further includes:
a. queuing said message packet at the one of a plurality of network switches;
b. determining a queuing level;
c. comparing the queuing level to a predetermined congestion metric.
15. The method of claim 14 , wherein the queuing level is a number of message packets in the respective queue and the predetermined congestion metric is a maximum number of message packets.
16. The method of claim 14 , wherein the queuing level is a first elapsed time measurement and the predetermined congestion metric is a second elapsed time measurement.
17. The method of claim 12 , wherein said message packet contains data.
18. The method of claim 12 , wherein said source is one of a plurality of sources, said destination is one of a plurality of destinations.
19. The method of claim 12 , wherein the speculative packets use a first virtual channel and the non-speculative packets use a second virtual channel, said first virtual channel having a lower priority than said second virtual channel.