Using a data processing unit (DPU) as a pre-processor for graphics processing unit (GPU) based machine learning
A system includes a network adapter and a processor. The network adapter is configured to receive packets from a network, to process the received packets in accordance with a policy, and to derive a packet digest from the received packets. The processor is configured to adapt the policy based on the packet digest derived by the network adapter, and to provide the adapted policy to the network adapter for subsequent packet processing.
1. A system, comprising:
a network adapter, to receive a flow of packets from a network, to process the received flow in accordance with a security policy, and to extract, from the received packets of the flow, a traffic pattern of at least some of the received packets, the traffic pattern being representative of the flow and independent of a size of the flow; and
a processor, to detect an anomaly in the traffic pattern extracted by the network adapter, to adapt the security policy in response to the anomaly, and to provide the adapted security policy to the network adapter for subsequent packet processing.
2. The system according to claim 1 , wherein the network adapter is to extract the traffic pattern by setting one or more counters in response to the received packets, and wherein the processor is to adapt the security policy in response to the one or more counters.
3. The system according to claim 1 , wherein the processor is to adapt the security policy by applying a machine learning (ML) algorithm to the traffic pattern extracted by the network adapter.
4. The system according to claim 1 , wherein the network adapter is a data processing unit (DPU).
5. The system according to claim 1 , wherein the processor is a graphical processing unit (GPU).
6. A method, comprising:
in a network adapter, receiving a flow of packets from a network, processing the received flow in accordance with a security policy, and extracting, from the received packets of the flow, a traffic pattern of at least some of the received packets, the traffic pattern being representative of the flow and independent of a size of the flow; and
in a processor that is separate from the network adapter, detect an anomaly in the traffic pattern extracted by the network adapter, adapting the security policy in response to the anomaly, and providing the adapted security policy to the network adapter for subsequent packet processing.
7. The method according to claim 6 , wherein extracting the traffic pattern comprises setting one or more counters in response to the received packets, and wherein adapting the security policy is performed in response to the one or more counters.
8. The method according to claim 6 , wherein adapting the security policy comprises applying a machine learning (ML) algorithm to the traffic pattern extracted by the network adapter.
9. The method according to claim 6 , wherein the network adapter is a data processing unit (DPU).
10. The method according to claim 6 , wherein the processor is a graphical processing unit (GPU).