Intel Corporation

Karthik Kumar

Francesc Guim Bernat

Thomas Willhalm

Mark A Schmisseur

Timothy Verrall

Ned M Smith

Raj K Ramanujan

Suraj Prabhakaran

Marcos Emanuel Carranza

Susanne M Balle

Kshitij Arun Doshi

Rajesh Poornachandran

Brinda Ganesh

Brian J Slechta

Andrew J Herdrich

Daniel Rivas Barragan

Patrick G Kutch

Alexander Bachmutsky

Mustafa Hajeer

Rahul Khanna

Narayan Ranganathan

Patrick Lu

Rashmin N Patel

Benjamin Graniello

Monica Kenguva

Alexander Vul

Martin P Dimitrov

Cesar Martinez-Spessot

Felipe Pastor Beneyto

Ignacio Astilleros Diez

Vincent J Zimmer

Mohan J Kumar

Robert G Blankenship

Sujoy Sen

Ramanathan Sethuraman

Subrata Banik

Petar Torre

Evan Custodio

Edwin Verplanke

Mateo Guzman

Anil Rao

Nicolas A Salhuana

Ginger Gilsdorf

Rita H Wouhaybi

Kapil Sood

John Joseph Browne

Tarun Viswanathan

Dimitrios Ziakas

Trevor Cooper

John Chun Kwok Leung

Tushar Gohad

Rita D Gupta

Patrick Bohan

Nicolae Octavian Popovici

Ravi L Sahita

Murugasamy K Nachimuthu

Pete D Vogt

Philip R Lantz

Thomas E Willis

Lidia Warnes

Chandan Egbert

Robert G Southworth

Slawomir Putyrski

Durgesh Srivastava

Nilesh K Jain

Amin Firoozshahian

Shrikant M Shah

Zhongyan Lu

Donald L Faw

Sunil K Cheruvu

Ravi Iyer

Uzair Qureshi

Lakshmi N Talluru

Francois Dugast

Ben McCahill

Kevin Bohan

Kirill Instrumentov

Mark D Tetreault

Steven Briscoe

Benjamin A Graniello

Francesc G Bernat

System and techniques for infrastructure managed workload distribution are described herein. An infrastructure processing unit (IPU) receives a workload that includes a workload definition. The workload definition includes stages of the workload and a performance expectation. The IPU provides the workload, for execution, to a processing unit of a compute node to which the IPU belongs. The IPU monitors execution of the workload to determine that a stage of the workload is performing outside of the performance expectation from the workload definition. In response, the IPU modifies the execution of the workload.

Infrastructure managed workload distribution

Methods, apparatus, systems, and articles of manufacture are disclosed to partition neural network models for executing at distributed Edge nodes. An example apparatus includes processor circuitry to perform at least one of first, second, or third operations to instantiate power consumption estimation circuitry to estimate a computation energy consumption for executing the neural network model on a first edge node, network bandwidth determination circuitry to determine a first transmission time for sending an intermediate result from the first edge node to a second or third edge node, power consumption estimation circuitry to estimate a transmission energy consumption for sending the intermediate result to the second or the third edge node, and neural network partitioning circuitry to partition the neural network model into a first portion to be executed at the first edge node and a second portion to be executed at the second or third edge node.

Apparatus, articles of manufacture, and methods to partition neural networks for execution at distributed edge nodes

Examples described herein relate to a network interface device that comprises circuitry, when operational, to select a platform to execute a function and based on load of the platform, selectively cause the function to execute on one or more other platforms to attempt to achieve or finish before the time-to-completion. In some examples, the circuitry is to detect progress of function execution to determine whether completion of execution of the function is predicted to not finish within the time-to-completion and cause the function to execute on one or more other platforms based on completion of execution of the function predicted to not finish within the time-to-completion. In some examples, the circuitry is to select the one or more other platforms to execute the function based on one or more of: processor computing utilization, available memory capacity, available cache capacity, network availability, or malfunction of a processor, memory, and/or cache.

Network interface device management of service execution failover

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. For remote memory, telemetry information is provided to a fabric management component that is used to dynamically reconfigure one or more fabric links.

Adaptive fabric allocation for local and remote emerging memories based prediction schemes

Systems, methods, articles of manufacture, and apparatus for end-to-end hardware tracing in an Edge network are disclosed. An example compute device includes at least one memory, instructions in the compute device, and processing circuitry to execute the instructions to, in response to receiving detecting an object having a global group identifier, generate monitoring data corresponding to a respective process executing on the compute device, the monitoring data including a process identifier, index the monitoring data having the process identifier to the corresponding global group identifier, synchronize a time stamp of the monitoring data to a network time protocol corresponding to the global group identifier, and transmit the indexed and synchronized monitoring data as tracing data in to the a tracing datastore.

Systems, methods, articles of manufacture, and apparatus for end-to-end hardware tracing in an edge network

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).

Methods and apparatus to share memory across distributed coherent edge computing system

A network switch includes a memory device to store a stream information of a plurality of data streams being handled by the network switch, the stream information including a stream identifier, a stream service level agreement (SLA), and a stream traffic type; accelerator circuitry to apply stream transformation functions to data streams; telemetry circuitry to monitor egress ports of the network switch; and scheduler circuitry to: receive telemetry data from the telemetry circuitry to determine that a utilization of egress ports of the network switch is over a threshold utilization; determine a selected data stream of the plurality of data streams to transform; use the accelerator circuitry to transform the selected data stream to produce a transformed data stream, wherein the transformed data stream complies with a corresponding stream SLA; and transmit the transformed data stream on an egress port.

Switch-based adaptive transformation for edge appliances

Implementation of object versioning and consistency at scale is described. An example of a computer-readable storage medium includes instructions to receive a storage request from an object management authority to store a version of a data object at a storage appliance; generate a hash value of the data object; store the data object in a scratchpad storage; transmit a verification request for the hash value of the data object to an object versioning authority; and, upon receiving affirmative verification for the hash value, move the data object from the scratchpad storage to a media storage and transmit an acknowledgement to the object management authority.

Implementation of object versioning and consistency at scale

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.

Memory schemes for infrastructure processing unit architectures

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.

Failover for pooled memory

A system comprising a traffic handler comprising circuitry to determine that data of a memory request is stored remotely in a memory pool; generate a packet based on the memory request; and direct the packet to a path providing a guaranteed latency for completion of the memory request.

Memory pooled time sensitive networking based architectures

Methods, apparatus, systems and articles of manufacture are disclosed for edge data prioritization. An example apparatus includes at least one memory, instructions, and processor circuitry to at least one of execute or instantiate the instructions to identify an association of a data packet with a data stream based on one or more data stream parameters included in the data packet corresponding to the data stream, the data packet associated with a first priority, execute a model based on the one or more data stream parameters to generate a model output, determine a second priority of at least one of the data packet or the data stream based on the model output, the model output indicative of an adjustment of the first priority to the second priority, and cause transmission of at least one of the data packet or the data stream based on the second priority.

Systems, apparatus, and methods for edge data prioritization

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.

Application aware memory patrol scrubbing techniques

The platform data aging for adaptive memory scaling described herein provides technical solutions for technical problems facing power management for electronic device processors. Technical solutions described herein include improved processor power management based on a memory region life-cycle (e.g., short-lived, long-lived, static). In an example, a short-term memory request is allocated to a short-term memory region, and that short-term memory region is powered down upon expiration of the lifetime of all short-term memory requests on the short-term memory region. Multiple memory regions may be scaled down (e.g., shut down) or scaled up based on demands for memory capacity and bandwidth.

Platform data aging for adaptive memory scaling

Examples described herein relate to a switch device. The switch device can perform replication of content stored in a source memory region to two or more memory regions available from two or more nodes, wherein the two or more memory regions available from two or more nodes are identified to the circuitry for use to store replicated content. The two or more nodes can be on different racks than that of a memory device that stores the source memory region. The switch device can select the two or more memory regions available from two or more nodes based, at least, in part on resiliency criteria associated with the two or more nodes.

High-availability memory replication in one or more network devices

Growing community of inventors

Chandler, AZ, United States of America

Karthik Kumar