Routing policies for graphical processing units
Discussed herein is a routing mechanism for graphical processing units (GPUs) that are hosted on several host machines in a cloud environment. For a packet transmitted by a GPU of a host machine and received by a network device, the network device determines an incoming port-link of the network device on which the packet was received. The network device obtains a flow information associated with the packet, and computes, based on the flow information, an outgoing port-link of the network device in accordance with a hashing algorithm. The hashing algorithm is configured to hash packets received on a particular incoming port-link of the network device to be transmitted on a same outgoing port-link of the network device. The network device forwards the packet on the outgoing port-link of the network device.
1. A method comprising:
for a packet transmitted by a graphical processing unit (GPU) of a host machine and received by a network device,
obtaining, by the network device, a flow information associated with the packet;
computing, by the network device, based on the flow information, an outgoing port-link of the network device in accordance with a hashing algorithm, wherein the network device is configured to direct all packets received on a particular incoming port-link of the network device to be transmitted on a same outgoing port-link of a plurality of outgoing port-links of the network device;
verifying, by the network device, a condition associated with the outgoing port-link of the network device;
responsive to the condition being satisfied, forwarding the packet on the outgoing port-link of the network device; and
responsive to the condition being unsatisfied
executing, by the network device, an equal cost multi-path algorithm to obtain a new outgoing port-link of the network device based on the flow information; and
forwarding, by the network device, the packet on the new outgoing port-link of the network device.
2. The method of claim 1 , wherein the condition corresponds to determining whether the outgoing port-link of the network device is active.
3. The method of claim 1 , wherein the network device is a top-of-rack (TOR) switch.
4. The method of claim 1 , wherein information related to a hash operation performed on packets received by the network device is stored in a forwarding table database.
5. The method of claim 1 , wherein for a first packet and a second packet that are received on a first incoming port-link of the network device, and a third packet and a fourth packet that are received on a second incoming port-link of the network device, the network device is configured for:
transmitting, by the network device, the first packet and the second packet on a first outgoing port-link of the network device; and
sending, by the network device, the third packet and the fourth packet on a second outgoing port-link of the network device, wherein the first incoming port-link of the network device is different than the second incoming port-link of the network device, and the first outgoing port-link of the network device is different than the second outgoing port-link of the network device.
6. The method of claim 5 , wherein the first packet, the second packet, the third packet and the fourth packet belong to a GPU workload.
7. The method of claim 5 , wherein the first incoming port-link and the second incoming port-link of the network device are a first set of links connecting the host machine to a top-of-rack (TOR) switch, and the first outgoing port-link and the second outgoing port-link of the network device are a second set of links connecting the TOR to a spine switch.
8. A network device comprising:
a processor; and
a memory including instructions that, when executed with the processor, cause the network device to, at least:
for a packet transmitted by a graphical processing unit (GPU) of a host machine and received by a network device,
obtain a flow information associated with the packet;
compute based on the flow information, an outgoing port-link of the network device in accordance with a hashing algorithm, wherein the network device is configured to direct all packets received on a particular incoming port-link of the network device to be transmitted on a same outgoing port-link of a plurality of outgoing port-links of the network device;
verify a condition associated with the outgoing port-link of the network device;
responsive to the condition being satisfied, forward the packet on the outgoing port-link of the network device; and
responsive to the condition being unsatisfied
execute an equal cost multi-path algorithm to obtain a new outgoing port- link of the network device based on the flow information; and
forward the packet on the new outgoing port-link of the network device.
9. The network device of claim 8 , wherein the condition corresponds to determining whether the outgoing port-link of the network device is active.
10. The network device of claim 8 , wherein the network device is a top-of-rack (TOR) switch.
11. The network device of claim 8 , wherein information related to a hash operation performed on packets received by the network device is stored in a forwarding table database.
12. The network device of claim 8 , wherein for a first packet and a second packet that are received on a first incoming port-link of the network device, and a third packet and a fourth packet that are received on a second incoming port-link of the network device, the network device is configured to:
transmit the first packet and the second packet on a first outgoing port-link of the network device; and
send the third packet and the fourth packet on a second outgoing port-link of the network device, wherein the first incoming port-link of the network device is different than the second incoming port-link of the network device, and the first outgoing port-link of the network device is different than the second outgoing port-link of the network device.
13. The network device of claim 12 , wherein the first packet, the second packet, the third packet and the fourth packet belong to a GPU workload.
14. The network device of claim 12 , wherein the first incoming port-link and the second incoming port-link of the network device are a first set of links connecting the host machine to a top-of-rack (TOR) switch, and the first outgoing port-link and the second outgoing port-link of the network device are a second set of links connecting the TOR to a spine switch.
15. A non-transitory computer readable medium storing specific computer-executable instructions that, when executed by a processor, cause a computer system to perform operations comprising:
for a packet transmitted by a graphical processing unit (GPU) of a host machine and received by a network device,
obtaining, by the network device, a flow information associated with the packet;
computing, by the network device, based on the flow information, an outgoing port-link of the network device in accordance with a hashing algorithm, wherein the network device is configured to direct all packets received on a particular incoming port-link of the network device to be transmitted on a same outgoing port-link of a plurality of outgoing port-links of the network device;
verifying, by the network device, a condition associated with the outgoing port-link of the network device;
responsive to the condition being satisfied, forwarding the packet on the outgoing port-link of the network device; and
responsive to the condition being unsatisfied
executing, by the network device, an equal cost multi-path algorithm to obtain a new outgoing port-link of the network device based on the flow information; and
forwarding, by the network device, the packet on the new outgoing port-link of the network device.