Abstract: An embodiment of an electronic apparatus may comprise one or more substrates and a controller coupled to the one or more substrates, the controller including circuitry to apply a reliability, availability, and serviceability (RAS) policy for access to a memory in accordance with a first RAS scheme, change the applied RAS policy in accordance with a second RAS scheme at runtime, where the second RAS scheme is different from the first RAS scheme, and access the memory in accordance with the applied RAS policy. Other embodiments are disclosed and claimed.

Abstract: An embodiment of an electronic apparatus may include one or more substrates, and logic coupled to the one or more substrates, the logic to process memory operation requests from a memory controller, and provide a front end interface to remote pooled memory hosted at a near edge device. An embodiment of another electronic apparatus may include local memory and logic communicatively coupled the local memory, the logic to allocate a range of the local memory as remote pooled memory, and provide a back end interface to the remote pooled memory for memory requests from a far edge device. Other embodiments are disclosed and claimed.

Abstract: A memory request manager in a memory system registers a tenant for access to a plurality of memory devices, registers one or more service level agreement (SLA) requirements for the tenant for access to the plurality of memory devices, monitors usage of the plurality of memory devices by tenants, receives a memory request from the tenant to access a selected one of the plurality of memory devices, and allows the access when usage of the plurality of memory devices meets the one or more SLA requirements for the tenant.

Abstract: Examples described herein relate to a network interface device that includes: a host interface; a direct memory access (DMA) circuitry; a network interface to receive, in at least one packet, time data associated with at least one of multiple layers, wherein the multiple layers provide inputs to a collective operation associated with a large language model (LLM); and circuitry. The circuitry is to based, at least in part, on the time data associated with the multiple layers, identify a first operation of a first layer of the multiple layers as a late completing process relative to times to completion of multiple first operations of other layers and based on the first operation being identified as a late completing process, perform a remedial action to adjust at least one configuration of a first device to execute a second operation of the first layer.

Abstract: Examples described herein relate to configuring a switch in an accelerator fabric to: monitor accesses to a memory region by one or more accelerators coupled to the accelerator fabric and report the accesses to the memory region to one or more specified accelerators coupled to the accelerator fabric. In some examples, the configuration includes a call to an application programing interface (API), a configuration file, a remote procedure call (RPC), or execution of a binary.

Abstract: Methods, apparatus and systems for adaptive fabric allocation for local and remote emerging memories-based prediction schemes. In conjunction with performing memory transfers between a compute host and memory device connected via one or more interconnect segments, memory read and write traffic is monitored for at least one interconnect segment having reconfigurable upstream lanes and downstream lanes. Predictions of expected read and write bandwidths for the at least one interconnect segment are then made. Based on the expected read and write bandwidths, the upstream lanes and downstream lanes are dynamically reconfigured. The interconnect segments include interconnect links such as Compute Exchange Link (CXL) flex buses and memory channels for local memory implementations, and fabric links for remote memory implementations. For local memory, management messages may be used to provide telemetry information containing the expected read and write bandwidths.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed that reduce latency and bandwidth consumption when sharing memory across a distributed coherent Edge computing system. The distributed coherent Edge computing system disclosed herein configures a compute express link (CXL) endpoint to share data between memories across an Edge platform. The CXL endpoint configures coherent memory domain(s) of memory addresses, which are initialized from an Edge device connected to the Edge platform. The CXL endpoint also configures coherency rule(s) for the coherent memory domain(s). The CXL endpoint is implemented to snoop the Edge platform in response to read and write requests from the Edge device. The CXL endpoint selectively snoops memory addresses within the coherent memory domain(s) that are defined as coherent based on the coherency rule(s).

Abstract: Systems, apparatuses, and methods provide for memory management where an infrastructure processing unit bypasses a central processing unit. Such an infrastructure processing unit determines if incoming packets of memory traffic trigger memory rules stored by the infrastructure processing unit. The incoming packets are routed to the central processing unit in a default mode when the incoming packets do not trigger the memory rules. Conversely, the incoming packets are routed to the infrastructure processing unit and bypass the central processing unit in an inline mode when the incoming packets trigger the memory rules. A memory architecture communicatively coupled to the central processing unit receives a set of atomic transactions from the infrastructure processing unit that bypasses the central processing unit and performs the set of atomic transactions from the infrastructure processing unit.

Abstract: An embodiment of an electronic apparatus may comprise one or more substrates, and a controller coupled to the one or more substrates, the controller to allocate a first secure portion of a pooled memory to a first instantiation of an application on a first node, and circuitry coupled to the one or more substrates and the controller, the circuitry to provide a failover interface for a second instantiation of the application on a second node to access the first secure portion of the pooled memory in the event of a failure of the first node. Other embodiments are disclosed and claimed.

Abstract: Methods and apparatus for application aware memory patrol scrubbing techniques. The method may be performed on a computing system including one or more memory devices and running multiple applications with associated processes. The computer system may be implemented in a multi-tenant environment, where virtual instances of physical resources provided by the system are allocated to separate tenants, such as through virtualization schemes employing virtual machines or containers. Quality of Service (QoS) scrubbing logic and novel interfaces are provided to enable memory scrubbing QoS policies to be applied at the tenant, application, and/or process level. This QoS policies may include memory ranges for which specific policies are applied, as well as bandwidth allocations for performing scrubbing operations. A pattern generator is also provided for generating scrubbing patterns based on observed or predicted memory access patterns and/or predefined patterns.

Abstract: An apparatus includes a host interface, a network interface, and programmable circuitry communicably coupled to the host interface and the network interface, the programmable circuitry comprising one or more processors are to implement network interface functionality and are to receive a prompt directed to an artificial intelligence (AI) model hosted by a host device communicably coupled to the host interface, apply a prompt tuning model to the prompt to generate an initial augmented prompt, compare the initial augmented prompt for a match with stored data of a prompt augmentation tracking table comprising real-time datacenter trend data and cross-network historical augmentation data from programmable network interface devices in a datacenter hosting the apparatus, generate, in response to identification of the match with the stored data, a final augmented prompt based on the match, and transmit the final augmented prompt to the AI model.

Abstract: Methods and apparatus for software-defined coherent caching of pooled memory. The pooled memory is implemented in an environment having a disaggregated architecture where compute resources such as compute platforms are connected to disaggregated memory via a network or fabric. Software-defined caching policies are implemented in hardware in a processor SoC or discrete device such as a Network Interface Controller (NIC) by programming logic in an FPGA or accelerator on the SoC or discrete device. The programmed logic is configured to implement software-defined caching policies in hardware for effecting disaggregated memory (DM) caching in an associated DM cache of at least a portion of an address space allocated for the software application in the disaggregated memory.

Abstract: There is disclosed in one example an application-specific integrated circuit (ASIC), including: an artificial intelligence (AI) circuit; and circuitry to: identify a flow, the flow including traffic diverted from a core cloud service of a network to be serviced by an edge node closer to an edge of the network than to the core of the network; receive telemetry related to the flow, the telemetry including fine-grained and flow-level network monitoring data for the flow; operate the AI circuit to predict, from the telemetry, a future service-level demand for the edge node; and cause a service parameter of the edge node to be tuned according to the prediction.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed to estimate workload complexity. An example apparatus includes processor circuitry to perform at least one of first, second, or third operations to instantiate payload interface circuitry to extract workload objective information and service level agreement (SLA) criteria corresponding to a workload, and acceleration circuitry to select a pre-processing model based on (a) the workload objective information and (b) feedback corresponding to workload performance metrics of at least one prior workload execution iteration, execute the pre-processing model to calculate a complexity metric corresponding to the workload, and select candidate resources based on the complexity metric.

Abstract: Technologies for accelerated key caching in an edge hierarchy include multiple edge appliance devices organized in tiers. An edge appliance device receives a request for a key, such as a private key. The edge appliance device determines whether the key is included in a local key cache and, if not, requests the key from an edge appliance device included in an inner tier of the edge hierarchy. The edge appliance device may request the key from an edge appliance device included in a peer tier of the edge hierarchy. The edge appliance device may activate per-tenant accelerated logic to identify one or more keys in the key cache for eviction. The edge appliance device may activate per-tenant accelerated logic to identify one or more keys for pre-fetching. Those functions of the edge appliance device may be performed by an accelerator such as an FPGA. Other embodiments are described and claimed.

Abstract: Technologies for fulfilling service requests in an edge architecture include an edge gateway device to receive a request from an edge device or an intermediate tier device of an edge network to perform a function of a service by an entity hosting the service. The edge gateway device is to identify one or more input data to fulfill the request by the service and request the one or more input data from an edge resource identified to provide the input data. The edge gateway device is to provide the input data to the entity associated with the request.

Abstract: Technologies for providing selective offload of execution of an application to the edge include a device that includes circuitry to determine whether a section of an application to be executed by the device is available to be offloaded. Additionally, the circuitry is to determine one or more characteristics of an edge resource available to execute the section. Further, the circuitry is to determine, as a function of the one or more characteristics and a target performance objective associated with the section, whether to offload the section to the edge resource and offload, in response to a determination to offload the section, the section to the edge resource.

Abstract: There is disclosed in one example an application-specific integrated circuit (ASIC), including: an artificial intelligence (AI) circuit; and circuitry to: identify a flow, the flow including traffic diverted from a core cloud service of a network to be serviced by an edge node closer to an edge of the network than to the core of the network; receive telemetry related to the flow, the telemetry including fine-grained and flow-level network monitoring data for the flow; operate the AI circuit to predict, from the telemetry, a future service-level demand for the edge node; and cause a service parameter of the edge node to be tuned according to the prediction.

Abstract: Technologies for quality of service based throttling in a fabric architecture include a network node of a plurality of network nodes interconnected across the fabric architecture via an interconnect fabric. The network node includes a host fabric interface (HFI) configured to facilitate the transmission of data to/from the network node, monitor quality of service levels of resources of the network node used to process and transmit the data, and detect a throttling condition based on a result of the monitored quality of service levels. The HFI is further configured to generate and transmit a throttling message to one or more of the interconnected network nodes in response to having detected a throttling condition. The HFI is additionally configured to receive a throttling message from another of the network nodes and perform a throttling action on one or more of the resources based on the received throttling message. Other embodiments are described herein.

Abstract: An embodiment of an electronic apparatus may include one or more substrates, and logic coupled to the one or more substrates, the logic to process memory operation requests from a memory controller, and provide a front end interface to remote pooled memory hosted at a near edge device. An embodiment of another electronic apparatus may include local memory and logic communicatively coupled the local memory, the logic to allocate a range of the local memory as remote pooled memory, and provide a back end interface to the remote pooled memory for memory requests from a far edge device. Other embodiments are disclosed and claimed.