Fabric independent PCIe cluster manager
A cluster manager of a computer cluster determines an allocation of resources from the endpoints for running applications on the nodes of the computer cluster and configures the computer cluster to provide resources for the applications in accordance with the allocation. The cluster may include a Peripheral Component Interconnect express (PCIe) fabric. The cluster manager may configure PCIe multi-root input/output (I/O) virtualization topologies of the computer cluster. The allocations may satisfy Quality of Service requirements, including priority class and maximum latency requirements. The allocations may involve splitting I/O traffic.
1. A cluster manager of a node of a computer cluster, the computer cluster comprising a plurality of computers and a Peripheral Component Interconnect express (PCIe) fabric, the computer cluster having endpoints connected to the computer cluster by a plurality of interconnects, the cluster manager comprising:
a communication module to receive specifications of resource requirements of applications to run on the computer cluster, the resource requirements including storage and network connectivity requirements; and
a configuration module to determine resources available from the endpoints of the computer cluster, to determine allocations of resources from the endpoints to satisfy the resource requirements of the applications, and to configure the computer cluster to provide resources for the applications in accordance with the allocations, the configuring including configuring the interconnects to provide the allocated resources to the applications, including allocating PCIe lanes from the PCIe fabric for interprocess communications and cluster service traffic.
2. The cluster manager of claim 1 , wherein the plurality of computers comprises at least two servers.
3. The cluster manager of claim 1 , wherein the plurality of computers comprises at least a server and a client computer.
4. The cluster manager of claim 1 , wherein:
the specification of resource requirements includes a specification of Quality of Service (QoS) requirements, including priority class and a maximum latency requirements; and
the configuration module is to determine an allocation of resources to satisfy the QoS requirements, including the priority class and maximum latency requirements, the determining comprising:
selecting classes of traffic of the applications; and
mapping the classes of traffic to Data Center Bridging (DCB) priority classes based upon priorities of the classes of traffic.
5. The cluster manager of claim 4 , wherein the configuration module is to allocate resources of virtual functions from the endpoints to the applications.
6. A method comprising:
receiving at a cluster manager of a node of a computer cluster a specification of resource requirements of an application to run on the computer cluster, the resource requirements including storage and network connectivity bandwidth requirements, wherein the computer cluster includes a Peripheral Component Interconnect express (PCIe) fabric;
determining resources available from endpoints of the computer cluster, wherein the endpoints are connected to the computer cluster by a plurality of interconnects;
determining an allocation of resources from the endpoints to satisfy the resource requirements of the application; and
configuring the computer cluster to provide resources for the application in accordance with the allocation, including configuring the interconnects to provide the allocated resources to the application, wherein the configuring comprises allocating PCIe lanes from the PCIe fabric for interprocess communications and cluster service traffic.
7. The method of claim 6 , wherein:
the specification of resource requirements includes a specification of Quality of Service (QoS) requirements; and
the determining the allocation of resources comprises determining an allocation of resources to satisfy the QoS requirements.
8. The method of claim 7 , wherein:
the QoS requirements comprise priority class and maximum latency requirements, and
the determining the allocation of resources comprises determining the allocation of resources to satisfy the priority class and maximum latency requirements.
9. The method of claim 8 , wherein the determining the allocation of resources comprises:
selecting classes of traffic of the application; and
mapping the classes of traffic to Data Center Bridging (DCB) priority classes based upon priorities of the classes of traffic.
10. The method of claim 7 , wherein the QoS requirements include Ethernet DCB QoS guarantees and InfiniBand latency limits.
11. The method of claim 6 , wherein:
the computer cluster comprises a multi-root input/output virtualization (MR-IOV) PCIe fabric;
the endpoints comprise PCIe devices connected to multi-root aware (MRA) switches of the PCIe fabric; and
the determining resources available comprises determining the characteristics of the PCIe devices based upon their connections to the MRA switches.
12. The method of claim 11 , wherein the configuring comprises configuring PCIe MR-IOV topologies of the computer cluster.
13. The method of claim 6 , wherein the endpoints communicate with the computer cluster via at least one of the following communications protocols:
InfiniBand;
Fibre Channel;
Internet Small Computer System Interface (i SCSI); and
Ethernet.
14. The method of claim 6 , wherein:
the method further comprises determining by the cluster manager an allocation of resources from the endpoints for another application, the application to run on another node of the computer cluster; and
the configuring comprises setting by the cluster manager input/output configuration parameters for the other node of the computer cluster.
15. The method of claim 6 , wherein the configuring comprises configuring multi-queue input/output adapters with PCIe bandwidth allocated for the application.
16. A method comprising:
receiving at a cluster manager of a node of a computer cluster a specification of resource requirements of an application to run on the computer cluster, the resource requirements including storage and network connectivity bandwidth requirements, wherein the computer cluster includes a Peripheral Component Interconnect express (PCIe) fabric;
determining resources available from endpoints of the computer cluster, wherein the endpoints are connected to the computer cluster by a plurality of interconnects;
determining an allocation of resources from the endpoints to satisfy the resource requirements of the application; and
configuring the computer cluster to provide resources for the application in accordance with the allocation, including configuring the interconnects to provide the allocated resources to the application, wherein the configuring comprises allocating resources from virtual functions of the endpoints to the application, and allocating PCIe lanes from the PCIe fabric for interprocess communications and cluster service traffic.
17. The method of claim 16 , wherein the configuring comprises allocating a portion of a capacity of an endpoint to the application and allocating another portion of the capacity of the endpoint to another application running on another node of the computer cluster.
18. The method of claim 16 , wherein the configuring comprises allocating a portion of a capacity of an endpoint to the application and allocating another portion of the capacity of the endpoint to another application running on another node of the computer cluster.