Infrastructure processing unit
Examples described herein relate to an Infrastructure Processing Unit (IPU) that comprises: interface circuitry to provide a communicative coupling with a platform; network interface circuitry to provide a communicative coupling with a network medium; and circuitry to expose infrastructure services to be accessed by microservices for function composition and to selectively provide a barrier to halt operation of at least one microservice based on event data from a composite node that performs the at least one microservice.
1. A system comprising:
a network interface device that comprises:
a direct memory access (DMA) circuitry;
interface circuitry to provide a communicative coupling with a first host platform;
network interface circuitry to provide a communicative coupling with a network medium to transmit and receive Ethernet packets;
circuitry to:
provide access to distributed nodes to microservices executed by the first host platform, wherein the distributed nodes comprises circuitry of a second host platform and a second network interface device and wherein the second host platform and the second network interface devices are communicatively coupled to the network interface device via the interface circuitry or the network interface circuitry; and
selectively provide, from the network interface device, a barrier to halt operation of at least one microservice of the microservices based on event data from the distributed nodes and based on the event data from the distributed nodes, release the barrier to permit operation of the at least one microservice of the microservices, wherein the release the barrier comprises permit devices of the distributed nodes to execute awaiting microservices.
2. The system of claim 1 , wherein the circuitry is to selectively provide a barrier to halt operation of at least one microservice based on the event data from the distributed nodes to determine a malfunction that could negatively impact performance of one or more currently executed microservices or one or more microservices to be executed.
3. The system of claim 1 , wherein the event data comprises one or more physical events at a device in the distributed nodes.
4. The system of claim 1 , wherein the event data comprises one or more of: voltage droop at a device in the distributed nodes, too high a memory error correction rate at a device in the distributed nodes, excess temperature of a device in the distributed nodes, or intrusion detection at a device in the distributed nodes.
5. The system of claim 1 , wherein the event data comprises one or more performance extrema of an operation of the distributed nodes.
6. The system of claim 1 , wherein the event data comprises one or more of: heartbeat signal timeout from an XPU of the distributed nodes, timeouts reported from hardware that is part of the distributed nodes, or excess memory errors from the distributed nodes.
7. The system of claim 1 , wherein the barrier to halt operation of at least one microservice comprises an acknowledgement that devices in the distributed nodes have entered active standby mode.
8. The system of claim 1 , wherein during application of the barrier, the circuitry is to place devices of the distributed nodes into active-standby mode to execute commands from the network interface device.
9. The system of claim 1 , wherein in response to the barrier, the circuitry is to cause migration of microservices and data from the distributed nodes to another distributed nodes.
10. The system of claim 9 , comprising:
a managed node comprising one or more distributed nodes, wherein the distributed nodes comprises computing resources that are logically coupled, wherein the computing resources include memory devices, data storage devices, accelerator devices, or general purpose processors, wherein resources from the one or more distributed nodes are collectively utilized in execution of a workload and wherein the workload execution is implemented by a group of microservices.
11. A method comprising:
at a network interface device comprising a direct memory access (DMA) circuitry, an interface to a first platform, and a network interface to transmit and receive Ethernet packets:
providing access to distributed nodes to microservices executed by the first platform, wherein the distributed nodes comprises circuitry of a second host platform and a second network interface device and wherein the second host platform and the second network interface device are communicatively coupled to the network interface device via the interface or the network interface;
selectively providing, by the network interface device, a barrier to halt operation of at least one microservice of the microservices based on event data from the distributed nodes; and
based on the event data from the distributed nodes, releasing the barrier and permitting devices of the distributed nodes to execute operation of the at least one microservice of the microservices.
12. The method of claim 11 , wherein the selectively providing a barrier to halt operation of at least one microservice based on the event data from the distributed nodes to determine a malfunction that could negatively impact performance of one or more currently executed microservices or one or more microservices to be executed.
13. The method of claim 11 , wherein the event data comprises one or more physical events at a device in the distributed nodes.
14. The method of claim 11 , wherein the event data comprises one or more of: voltage droop at a device in the distributed nodes, too high a memory error correction rate at a device in the distributed nodes, excess temperature of a device in the distributed nodes, or intrusion detection at a device in the distributed nodes.
15. The method of claim 11 , wherein the event data comprises one or more performance extrema of an operation of the distributed nodes.
16. The method of claim 11 , wherein the event data comprises one or more of: heartbeat signal timeout from an XPU of the distributed nodes, timeouts reported from hardware that is part of the distributed nodes, or excess memory errors from the distributed nodes.
17. The method of claim 11 , wherein during application of the barrier, the network interface device placing devices of the distributed nodes into active-standby mode to execute commands from the network interface device.
18. The method of claim 11 , comprising: the network interface device releasing the barrier and permitting devices of the distributed nodes to execute awaiting microservices.
19. The method of claim 11 , wherein in response to the barrier, the network interface device causing migration of microservices and data from the distributed nodes to another distributed nodes.
20. An apparatus comprising:
a network interface device, comprising a direct memory access (DMA) circuitry, host interface, network interface, and circuitry, wherein the circuitry is to:
manage, at the network interface device, forward progress of microservices executed by distributed nodes by halting forward progress of microservices executed by the distributed nodes and based on event data from the distributed nodes, release a barrier and permit resuming progress of the microservices executed by the distributed nodes based on event data from the distributed nodes.
21. The apparatus of claim 20 , wherein to manage forward progress of microservices executed by distributed nodes, the circuitry is to issue a barrier to halt progress of a first process.
22. The apparatus of claim 20 , wherein the manage forward progress of microservices executed by distributed nodes is based on telemetry data comprising one or more physical events in the distributed nodes.