Flexible resource sharing in a network
A network processing device connects to one or more devices in a computing node and connects to one or more other network processing devices of other computing nodes. The network processing device identifies a policy for allowing devices in other computing nodes to access a particular resource of one of the devices in its computing node. The network processing device receives an access request to access the particular resource from another network processing device and sends a request to the device hosting the particular resource based on the access request and the policy.
1 . An apparatus comprising:
a network processing device comprising:
a first interface to couple to a first one of a set of devices in a cluster via a first interconnect, wherein the first device includes a set of hardware resources, the first interface supports a plurality of interconnect protocols, and the plurality of interconnect protocols comprise a memory interconnect protocol and a cache coherent interconnect protocol;
a second interface to couple to another network processing device associated with another cluster;
circuitry to:
identify attributes of the set of hardware resources of the first device;
determine, based on the attributes, that a portion of the set of hardware resources can be shared with the other cluster using one or more of the plurality of interconnect protocols;
receive an access request, from a second device in the other cluster, to access a particular hardware resource in the portion of the set of hardware resources; and
send a request to the first device, wherein the request corresponds to the access request and is based on one of the plurality of interconnect protocols.
2 . The apparatus of claim 1 , wherein the circuitry is further to send a message to the other network processing device to advertise that the portion of the set of hardware resources comprise shareable resources.
3 . The apparatus of claim 2 , wherein the message is to identify conditions for access to the portion of the set of hardware resources by requesting devices outside the cluster.
4 . The apparatus of claim 3 , wherein the conditions comprise at least one of a type of the requesting device, an amount of the portion of the set of hardware resources that may be shared, or a particular one of the plurality of interconnect protocols to be used during sharing of the hardware resource.
5 . The apparatus of claim 1 , wherein the plurality of interconnect protocols comprises a plurality of Compute Express Link (CXL) protocols, the memory interconnect protocol comprises CXL.mem, and the cache coherent interconnect protocol comprises CXL.cache.
6 . The apparatus of claim 1 , wherein the circuitry is further to select one of the memory interconnect protocol or the cache coherent interconnect protocol for the request.
7 . The apparatus of claim 1 , wherein the circuitry is further to:
send an advertisement message to a system management entity, wherein the advertisement message identifies the first device and the attributes; and
receive a policy message from the system management entity based on the advertisement message, wherein the policy message identifies that the portion of the set of hardware resources can be shared.
8 . The apparatus of claim 1 , wherein the first cluster comprises a first data center cluster and the second cluster comprises a second data center cluster.
9 . The apparatus of claim 1 , wherein the particular hardware resource comprises a memory of the first device.
10 . The apparatus of claim 1 , wherein the particular hardware resource comprises a compute resource of the first device.
11 . A method comprising:
receiving an indication, at a network processing device, that a hardware resource of a first device is shareable, wherein the first device is coupled to the network processing device by a first interconnect in a first platform, the network processing device couples to another network processing device of a second platform via a second interconnect, the first interconnect supports a plurality of interconnect protocols, and the plurality of interconnect protocols comprise a memory interconnect protocol and a cache coherent interconnect protocol;
receiving an access request, from a second device in the second platform, to access the hardware resource of the first device;
determining a policy for access to the hardware resource by the second device based on the indication; and
sending a request to the first device over the first interconnect, wherein the request corresponds to the access request and is based on the policy.
12 . The method of claim 11 , wherein the request is according to one of the memory interconnect protocol or the cache coherent interconnect protocol based on the policy.
13 . The method of claim 12 , wherein the request comprises a first request, the cache coherent interconnect protocol is used for the first request, and the method further comprises:
receiving a second access request from the second device to access the hardware resource; and
sending a second request to the first device over the first interconnect, wherein the memory interconnect protocol is used for the second request based on the policy.
14 . The method of claim 11 , wherein the indication is received from a management entity of a system and the system comprises the first platform and the second platform.
15 . A system comprising:
a first node comprising:
a first network processing device; and
a first set of devices coupled to the first network device by a first set of links, wherein the first set of links use a first interconnect protocol, the first interconnect protocol defines a memory sub-protocol and a cache coherent sub-protocol;
a second node comprising;
a second network processing device coupled to the first network processing device by an interconnect; and
a second set of devices coupled to the second network device by a second set of links, wherein the second set of links use the first interconnect protocol,
wherein the first network processing device is to:
determine a policy for sharing a hardware resource of a particular one of the first set of devices;
receive an access request from the second network processing device for a second device in the second set of devices to access the hardware resource;
generate a request according to the first interconnect protocol to send to the particular device, wherein the request corresponds to the access request and is based on the policy.
16 . The system of claim 15 , wherein one of the first network processing device or the second network processing device comprises an infrastructure processing unit (IPU).
17 . The system of claim 15 , wherein the first set of devices comprises at least one of a host processor, an accelerator, or memory expander.
18 . The system of claim 15 , wherein the interconnect uses a different interconnect protocol.
19 . The system of claim 15 , further comprising a system manager executable to:
configure a first network to include the first device and the second device in association with a first program, wherein the hardware resource is accessible to the second device in the first network; and
configure a different second network to include the first device and the second device in association with a different second program, wherein the hardware resource is also accessible to the second device in the second network.
20 . The system of claim 19 , wherein a first plurality of devices are interconnected in the first network by links according to the first interconnect protocol, and a different second plurality of devices are interconnected in the second network by links according to the first interconnect protocol.
21 . The system of claim 19 , wherein access to the hardware resource by the second device is provided by a link using the memory subprotocol in the first network and access to the hardware resource by the second device is provided the cache coherent subprotocol in the second network.
22 . The system of claim 15 , wherein the policy permits alternating between the memory sub-protocol and the cache coherent sub-protocol for requests to access the hardware resource.