Abstract: Embodiments disclosed herein include a server drawer. The server drawer may include a lower deck heat exchanger with an inlet, an outlet, and a fluid pump. The server drawer may also include an upper deck heat exchanger located in an upper airflow path after first heat producing elements and before a component of concern, and a manifold. The manifold may include distributors connecting i) the inlet to the upper deck heat exchanger and ii) the outlet to the upper deck heat exchanger. The fluid pump may pump cooled fluid between the upper deck heat exchanger and the lower deck heat exchange.

Abstract: A method for parallel additive manufacturing includes receiving a three-dimensional digital model file of an object for construction utilizing multiple additive manufacturing devices. The method also includes determining a decomposition of the three-dimensional digital model file of the object into multiple portions and instructing each additive manufacturing device from the multiple additive manufacturing devices to generate at least one portion from the multiple portions of the object. The multiple additive manufacturing devices operate in parallel to generate the multiple portions of the object. The method also include constructing the object with the multiple portions generated by the multiple additive manufacturing devices.

Abstract: A controlled extraction of an electronic connector from an electronic circuit board includes a press mechanism of a device positioned in spaced relation over a base fixture of the device. The base fixture receiving an electronic circuit board which includes a connector. Shafts of a removal tool of the press mechanism can be inserted into a series of openings in the connector, respectively, in response to applying a downward force to the press mechanism. A capturing element on each of the shafts can be actuated where the capturing element couples to a wall defining the opening of each of the openings. The connector can be removed from the electronic circuit board by applying an upward force to the press mechanism which removes the connector with the capturing element of the shafts coupled to the wall of the opening, and releasing the press mechanism from the base fixture.

Abstract: An embodiment senses an object from an object screening system, and responsive to the sensed object, generates a context information based on the sensed object. The embodiment generates a cultural classification by a Culturally Aware Objects Model based in part on the context information wherein the Culturally Aware Objects Model is trained on a cultural factor of an object. The embodiment generates a moderated context information by a Sensitive Information Moderation Model based in part on the cultural classification and the context information wherein the Sensitive Information Moderation Model is trained on a privacy parameter. The embodiment also computes a recommendation by a Recommendation Component based in part on the moderated context information wherein the object screening system is updated with the recommendation.

Abstract: Managing container updates is provided. A changed file corresponding to a new version of an application is retrieved from a persistent volume based on a mapping between the new version of the application and the changed file. The changed file corresponding to the new version of the application is loaded in a corresponding file of a running container to update the corresponding file of the running container without redeploying the running container using an updated container image.

Abstract: In a first aspect of the invention, there is a computer-implemented method including: generate, by a processor set, an interaction usage graph based on user interaction data on user interactions with an analytical system user interface; generate, by the processor set, an interaction embedding model in a vector space based on the interaction usage graph; determine, by the processor set and based on the interaction embedding model in the vector space, a similarity of a portion of the interaction embedding model that corresponds to a particular analytical task among the user interactions with the analytical system user interface with a particular machine learning model from a set of one or more machine learning models; and output, by the processor set, to the analytical system user interface, an indication of the particular machine learning model.

Abstract: A computer-implemented method for receiving, by a processor set, a workload for processing. The processor set may further collect historical data and real-time data related to processing the received workload and generate a feature map by investigating the received workload, the historical data, and the real-time data. The processor set may further divide the workload into workload steps and assign each workload step to one of a plurality of groups. System resources may also be calculated for each of the workload steps. The processor set trains a machine learning model to determine influence factors for each of the plurality of groups, and then, the processor set reallocates the workload steps based on the calculated system resources and the influence factors.

Abstract: An embodiment monitors a code repository for changes made to source code. The embodiment detects a change made to the source code, and upon detecting the change made to the source code, identifies a potentially impacted workflow node of a workflow based on the change made to the source code. The embodiment identifies an execution tree of the workflow that contains the potentially impacted workflow node. The embodiment identifies a set of test cases corresponding to the execution tree. The embodiment generates a test case recommendation based on the set of test cases identified. The embodiment transmits the test case recommendation to a user device.

Abstract: A 3D compute-in-memory accelerator system and method for efficient inference of Mixture of Expert (MoE) neural network models. The system includes a plurality of compute-in-memory cores, each in-memory core including multiple tiers of in-memory compute cells. One or more tiers of in-memory compute cells correspond to an expert sub-model of the MoE model. One or more expert sub-models are selected for activation propagation based on a function-based routing, the tiers of the corresponding experts being activated based on this function. In one embodiment, this function is a hash-based tier selection function used for dynamic routing of inputs and output activations. In embodiments, the function is applied to select a single expert or multiple experts with input data-based or with layer-activation-based MoEs for single tier activation. Further, the system is configured as a multi-model system with single expert model selection or with a multi-model system with multi-expert selection.

Abstract: Mechanisms are provided to implement an enhanced privacy deep learning system framework (hereafter “framework”). The framework receives, from a client computing device, an encrypted first subnet model of a neural network, where the first subnet model is one partition of multiple partitions of the neural network. The framework loads the encrypted first subnet model into a trusted execution environment (TEE) of the framework, decrypts the first subnet model, within the TEE, and executes the first subnet model within the TEE. The framework receives encrypted input data from the client computing device, loads the encrypted input data into the TEE, decrypts the input data, and processes the input data in the TEE using the first subnet model executing within the TEE.

Abstract: Systems and techniques that facilitate neuronal activity modulation of artificial neural networks are provided. In various embodiments, an artificial neural network can comprise a set of base neuron populations that collectively generate, during an inferencing phase or a training phase of the artificial neural network, an inferencing task result based on a data candidate. In various aspects, the artificial neural network can comprise a control neuron population that is independent of the set of base neuron populations. In various instances, the control neuron population can modulate, during the inferencing phase or the training phase, neuronal activity of at least one base neuron population of the set of base neuron populations. In various cases, the control neuron population can modulate the neuronal activity of the at least one base neuron population by scaling one or more operands internally produced by the at least one base neuron population.

Abstract: Systems, methods are provided for encrypting, decrypting, re-encrypting an existing database with neither downtime nor storage overhead. Create a list of tables to be processed. Identify status of each table by an indicator in a database catalog. Store a second status in each data page. Select a table to process from list of tables. Issue a table lock against the selected table. Select a configurable number of data pages to process during the duration of the table lock. Following the processing, flush the data pages to storage. If there are no more data pages to encrypt for the selected table, remove the selected table from the list of tables to be processed. Release the table lock. If there are more data pages to process in the selected table, and there are waiters for the selected table, release the table lock; select another table from the list of tables to process.

Abstract: Processing within a computing environment is facilitated by using a machine learning implicit-anomaly model to determine a possibility of an implicit anomaly within a system based on system metrics data obtained during running of one or more test cases on the system. The process further includes determining, using artificial intelligence, occurrence of an incident within the system associated with running of the one or more test cases. Determining the occurrence of the incident uses the determined possibility of the implicit anomaly within the system based on the system metrics data, and the process further includes initiating an action based on the occurrence of the incident within the system with running of the one or more test cases.

Abstract: A microelectronic structure that includes a stack nanosheet transistor comprising a lower nanosheet transistor and an upper nanosheet transistor. The lower nanosheet transistor includes a lower source/drain and the upper nanosheet transistor includes an upper source/drain. An airgap located between the upper source/drain and the lower source/drain. The airgap is vertically aligned with the upper source/drain and the lower source/drain. A first layer located between the airgap and the lower source/drain.

Abstract: A packaged device that carries multiple component devices uses a back-mounted structure to reduce the area of the substrates in the package. The package includes a first organic laminate substrate and a second organic laminate substrate. The first organic laminate substrate is the base substrate of the packaged device. The second organic laminate substrate has higher wiring density than the first organic laminate substrate. The second organic laminate substrate is joined to a top surface (or module mounting side) of the first organic laminate substrate. A first component device is mounted on a top surface of the second organic laminate substrate. A second component device is mounted on a bottom surface of the second organic laminate substrate. The second component device recesses into a cavity at the top surface of the first organic laminate substrate.

Abstract: Computer implemented methods, systems, and computer program products include program code executing on a processor(s) identifies in a query from a requestor, a predicate(s) comprising a partial access of large object (LOB) data from a LOB. The processor(s) caches a location of the partial access of the LOB data in a buffer, labeling the location with predicates. The processor(s) determines, based on comparing a pre-set access count to a count of accesses for the location that the location is a frequently accessed location and stabilizes the location in an index. The processor(s) determines that the location meets a length threshold and a frequency threshold and saves the LOB data requested in the partial access of the LOB data inline with a base row.

Abstract: Within a database table being replicated from a source database to a target database, a first logged database operation on a database column holding large object (LOB) data is identified. The source database is queried for a plurality of rows of LOB data, each row in the plurality of rows of LOB data comprising LOB data stored in the database column holding LOB data. Responsive to determining that a first row of LOB data in the plurality of rows of LOB data is smaller than a message size threshold, the first row of LOB data is sent to the target database, the sending performed in a first message.

Abstract: A computer-implemented method for automated policy decision making optimization is disclosed. The computer-implemented method includes creating a dataset from a tabular database, wherein the dataset includes one or more columns selected as state variables, a column selected as action variables, and a column selected as reward variables. The computer-implemented method further includes determining a candidate function approximator Q based on applying at least one state variable, one action variable, and one reward variable to a trained regression model. The computer-implemented method further includes learning a decision policy based on applying the candidate function approximator Q to a reinforcement learning algorithm. The computer-implemented method further includes determining, based on the learned decision policy, an expected reward.

Abstract: A computer-implemented method, according to one approach, includes receiving at least one set of qubits at a first client component, and using the at least one set of qubits to index a configuration dictionary to determine a first configuration. The configuration dictionary defines a plurality of different configurations. The method further includes causing the first client component to be configured according to the first configuration. A computer program product, according to another approach, includes a computer readable storage medium having program instructions embodied therewith. The program instructions are readable and/or executable by a first client component to cause the first client component to perform the foregoing method. A system, according to another approach, includes a processor, and logic integrated with the processor, executable by the processor, or integrated with and executable by the processor. The logic is configured to perform the foregoing method.

Abstract: Examples described herein provide a computer-implemented method for data lakehouse encryption that includes receiving a query for data stored in a data lakehouse. The method further includes determining whether the query is a frequent query by comparing a frequency of the query to a threshold. The method further includes, responsive to determining that the query is a frequent query, storing envelope encrypted data and a query index in a frequent data access database, the envelope encrypted data being data that relates to the query. The method further includes, responsive to receiving the query at a later time, returning, to a user, the envelope encrypted data from the frequent data access database without decrypting other data stored in the data lakehouse.

Abstract: A process is described to allow an early power estimate where switching activity provided by a simulation of a logical hierarchy in a chip design is accomplished by mapping ports in the physical hierarchy to ports in a logical hierarchy after hierarchy manipulation has been done. Power estimation is done using the switching activity from the simulation of the logical hierarchy mapped to the ports of the physical hierarchy. Early power estimation using results of the hierarchy manipulation is better than power estimation on the logical hierarchy because blocks and signals in the logical hierarchy may be replicated or merged during the hierarchy manipulation.

Abstract: In one general embodiment, a method includes positioning a read transducer of a magnetic head at an initial position over a data track on a magnetic recording tape using servo information derived from servo reader signals. A reading characteristic is measured during reading of the data track. The position of the magnetic head relative to the data track is sequentially changed. The reading characteristic is measured during reading of the data track at the changed positions. Information derived from the measured reading characteristic and the positions of the magnetic head is assembled into a data set. The data set is processed to create a read offset corresponding to a final position approximately geometrically centered within the data track. The read offset is stored in a memory.

Abstract: A local clock buffer for improving energy efficiency using a power saving clock buffer for clock grid-based scanning includes a grid node that receives a global clock signal from a global clock grid; a clock enable latch configured to latch a clock enable signal, wherein the clock enable latch is clocked based on the global clock signal; a clock gate configured to output a functional clock signal in dependence upon a latched value of the clock enable signal; a scan clock enable latch configured to latch a scan enable signal, wherein the scan enable latch is a non-clocked latch; and a scan clock gate coupled to the output of the functional clock gate and configured to output a scan clock signal in dependence upon a latched value of the scan enable signal.

Abstract: Mechanisms are provided for dynamically implementing reactive actions in edge nodes of a network in response to user equipment (UE) behaviors. Data of UE events are collected to infer UE movements and UE behavior within the network. A machine learning computer model is executed on the collected data of UE events to predict UE movements and UE behavior and their impact on edge node conditions within the network with regard to quality of service (QoS) metrics. An accuracy of the precited impacts of the predicted UE movements and UE behavior is evaluated and, based on the accuracy, reactive action(s) to execute to reduce the predicted impact of inaccurate predictions on edge node conditions with regard to the QoS metrics are determined and recommended to a control plane of the network for implementation of at least one of the one or more reactive actions on edge node(s) of the network.

Abstract: A semiconductor structure, system, and method of forming a crescent-shaped dielectric isolation layer for stacked field-effect transistors (FETs). The semiconductor structure may include a transistor including an epi. The semiconductor may also include a substrate, where the epi is directly connected to the substrate. The semiconductor may also include an isolation layer directly connected to the epi and the substrate. The system may include a semiconductor structure. The method may include forming an isolation layer directly connected to a substrate. The method may also include forming a first transistor, where forming the first transistor includes growing a first epi, where the first epi is directly connected to the isolation layer and the substrate.

Abstract: Embodiments disclosed herein include a semiconductor structure. The semiconductor structure may include a device layer having a first top source/drain (S/D), a second top S/D, a first bottom S/D, and a second bottom S/D. The semiconductor structure may further include a first direct back side contact connected to the first bottom S/D, a second direct back side contact connected to the second bottom S/D, and a double via. The double via may include a first half connecting a front side of the device layer to the first direct back side contact, and a second half connecting the front side to the second direct back side contact. The double via may also include a dividing cut splitting the first half of the double via from the second half of the double via.

Abstract: A method of fabricating a semiconductor device, includes providing a semiconductor structure having a dielectric stack and a mandrel layer positioned on the dielectric stack. An array of sacrificial mandrel features are patterned into the mandrel layer and on top of an insulating layer of the dielectric stack. A self-aligned non-mandrel cut is formed adjacent a spacer element located adjacent to one of the sacrificial mandrel features. The sacrificial mandrel features are removed. One or more self-aligned mandrel cuts are formed in one or more of the plurality of trenches. Non-mandrel openings are formed on top of the insulating layer. Continuity line openings are etched into the dielectric stack, wherein the non-mandrel cut interrupts a first of the continuity line openings and the mandrel cut is disposed to interrupt a second of the continuity line openings. Metal lines are formed in the continuity line openings, except where the non-mandrel cut and mandrel cut are disposed.

Abstract: A semiconductor device includes a backside power rail; a transistor source/drain structure that has a backside facing the backside power rail and has a frontside facing away from the backside power rail; and a via disposed between and electrically connecting the backside power rail and the source/drain structure. The via includes a buried portion that is disposed between the backside power rail and the backside of the transistor source/drain structure. A part of the buried portion overlaps and contacts at least a part of the backside of the source/drain structure. The via also includes a side portion that is electrically connected with the buried portion and extends along a vertical side of the source/drain structure between the frontside and the backside; and a top portion that is electrically connected with the side portion and covers at least a part of the frontside of the source/drain structure.

Abstract: A computer system, computer readable storage medium, and computer-implemented method for enhancing geospatial mobility management of unmanned aerial vehicles (UAVs). The method includes generating a mapping between a UAV designator and a plurality of geospatial region designators. The method also includes instantiating a network slice. The network slice is associated with the UAV designator and the plurality of geospatial region designators. The method further includes generating a flight path for a UAV through a plurality of geospatial regions that is at partially determined subject to the network slice and the plurality of geospatial region designators. The method also includes executing a transit of the UAV through the flight path.

Abstract: A system includes a processor that executes computer executable components stored in a memory. The computer executable components can include a generation component that generates a knowledge graph of a web-based application, and an intent component that receives a query and predicts an intent based upon the received query. The computer executable components can further include an identification component that identifies a target feature associated with the web-based application based upon the predicted intent, determines a criticality of the feature based upon the predicted intent, and assigns a weight to the feature based upon the determined criticality.

Abstract: An example operation may include at least one of receiving thermal measurements from at least one thermal sensor embedded with a plurality of processing cores of a multicore processor while the plurality of processing cores process workloads, modifying operating characteristics of the plurality of processing cores based on the thermal measurements to generate modified operating characteristics, and dispatching a workload to the plurality of processing cores based on the modified operating characteristics.

Abstract: Mechanisms are provided for intrusion detection based on regular expression matching. The mechanisms partition a regular expression (RegEx) rule set into a plurality of different partitions and distributes the plurality of different partitions to a plurality of different edge computing devices associated with a protected network of computing resources. The mechanisms route data packets of an incoming data stream to the plurality of edge computing devices, each of which processes the data packets to determine whether the data packets match RegEx rules in a partition distributed to the edge computing device. A determination is made as to whether the incoming data stream represents an intrusion based on a combination of results of the processing of the data packets by the plurality of edge computing devices.

Abstract: Embodiments are provided for providing a continuous knowledge graph in a computing system by a processor. One or more weighted values of an edge between a pair of entities in a knowledge graph may be predicted based on one or more candidate statements. A confidence score may be generated for the one or more predicted weighted values.

Abstract: A semiconductor device including a magnetic tunnel junction (MTJ) stack, where a cross section of a bottom electrode of the stack includes a hexagonal profile. A semiconductor device including a lower word line, a magnetic tunnel junction (MTJ) stack, where a cross section of a first electrode of the MTJ stack comprises a hexagonal profile. A method including forming a bottom electrode, the bottom electrode includes a side surface including a width at a middle section of the bottom electrode greater than a width at a lower surface of the bottom electrode, and the width at the middle section of the bottom electrode greater than a width at an upper surface of the bottom electrode.

Abstract: From data of a first interaction environment, user activity data while using the first interaction environment, and user biometric data while using the first interaction environment, a first cognitive load index is generated. A first directed acyclic graph corresponding to the first cognitive load index is generated. Using a graph convolutional network, the first directed acyclic graph, and a second directed acyclic graph generated from a second cognitive load index, a set of cognitive overload causation factors is generated. Using the set of cognitive overload causation factors, the first interaction environment is adjusted.

Abstract: Obtain, using at least one hardware processor, data characterizing a physical system governed by a physical conservation law. Apply, using the at least one hardware processor, contrastive learning to the data to automatically capture system invariants of the physical system. Employ, using the at least one hardware processor, a neural projection layer to guarantee that a corresponding dynamic machine learning model preserves the captured system invariants. Optionally, predict performance of the physical system using the corresponding dynamic machine learning model.

Abstract: A method, system, and a computer program product characterize a semiconductor. The method includes receiving longitudinal and transverse magnetoresistance measurements of a sample with first and second channels in a parallel magnetic field and receiving longitudinal and transverse magnetoresistance measurements of the sample in a perpendicular magnetic field. The method also includes subtracting the magnetoresistances measured in the parallel magnetic field from the magnetoresistances measured in the perpendicular field to obtain longitudinal and transverse magnetoresistance values Rxx and Rxy. The method further includes determining, based on the values Rxx and Rxy, harmonic magnetoresistance coefficients a0, and a2, b1, and b3. A method and system for semiconductor testing and characterization are disclosed as well.

Abstract: A computer-implemented method, according to one embodiment, includes receiving, on a first satellite, a first connection request, and analyzing the first connection request to determine services needed to fulfill the first connection request. In response to a determination that the first satellite does not have at least a predetermined number of the determined services loaded, a predetermined offload process is performed.

Abstract: A computer-implemented method, a computer system and a computer program product orchestrate event-driven workloads in a satellite environment. The method includes receiving a workload from a client for deployment to a satellite in a plurality of satellites, where the workload includes requirements. The method also includes obtaining satellite data for each satellite in the plurality of satellites, where the satellite data includes a capability of the satellite and a current status of the satellite. The method further includes determining that the capability of the satellite satisfies the requirements of the workload. In addition, the method includes identifying a relevant satellite in the plurality of satellites based on the current status of the satellite. Lastly, the method includes deploying the workload to a ground station associated with the relevant satellite.

Abstract: Disclosed are a computer-implemented method, a computer system and a computer program product for processing data in a data format with key-value pairs. All keys can be extracted from a plurality of data objects of the data. Duplicated keys can be removed such that only one of the same keys remains. The data in the data format can be reorganized into a format with a key portion followed by a value portion. The extracted keys can be arranged in the key portion in a predetermined order. The values of each of the plurality of data objects can be arranged in the value portion in an order corresponding to the order of the keys.

Abstract: Dynamically coordinating interactive fence zones and alerts includes receiving, by one or more processors, location data corresponding to a first object. A virtual fence location and a virtual fence type are determined for the first object based on the received location data, and a first virtual fence is then established surrounding the first object. A plurality of parameters associated with the first virtual fence is received by the one or more processors, and a rule table is generated based on the received plurality of parameters associated with the first virtual fence. Data regarding a virtual fence boundary of the first virtual fence is received by the one or more processors, and based on the received data meeting a first rule of the rule table, an alert is sent to a user device.

Abstract: A computing system employs one or more security managers that, in response to an active session user revocation request, such as from a security administrator, to revoke a user during an active session, update an active session access revocation list that includes data representing the user. The security manager revokes active user access to protected system resources for the user based on the updated active session access revocation list, during the active session. The active session user revocation request also causes an update of a user profile in a security manager database that allows the security manager to deny authentication of the user for new sessions. The updated active session access revocation list serves as a type of centralized list with all users that have been fully revoked on the computing system. Related methods and computer products are also described.

Abstract: A computer-implemented method for predicting power outages for use in vegetation management is provided. Aspects include obtaining historical weather data for a geographical location including a utility line, obtaining historical outage data for the utility line, and training an outage prediction model based at least in part on the historical weather data and the historical outage data. Aspects also include generating, based on the historical weather data, a future storm data set, obtaining an outage prediction score for the utility line by inputting the future storm data set into the outage prediction model, and creating a vegetation trimming score for the utility line based at least in part on the outage prediction score, wherein the vegetation trimming score for the utility line indicates an urgency for performing trimming vegetation adjacent to the utility line.

Abstract: Performing a three-dimensional (3D) reconstruction process of a set of selected image frames. In some instances, a plurality of sequential image frames that are extracted from a video file (including video content created through the use of Augmented Reality) are converted into a grayscale format. Once these image files are converted to a grayscale format, a two-dimensional (2D) entropy process is performed that is ultimately used to create a set of user-defined clusters, with each of these user-defined clusters being associated with at least a sub-set set of the grayscale frames. After a cluster is selected, a 3D reconstruction process is performed on the frames associated with the selected cluster.

Abstract: An integrated circuit includes a semiconductor substrate and integrated circuitry on the semiconductor substrate. The integrated circuitry includes a memory array including a plurality of memory macros including at least first and second memory macros. Each of the plurality of memory macros includes multiple partial arrays and a shared macro controller configured to control read and write access to the multiple partial arrays. The memory array also includes spare access control logic configured to direct an access to a first partial array in the first memory macro to a second partial array in the second memory macro.

Abstract: Provided herein are methods and systems for providing a function as a service in a predictably timed manner, for example within a deadline or in a given frequency. A specialized executor function, namely a custom runner, is called from the function service using a wrapper function on the framework controller. The process allocates some, or all of the required computing resources required to execute a function on the initialization stage, based on foresight. The custom runner function may keep the resources for expected function service requests, or terminate in accordance with prediction rules thereof. The disclosure supports different platforms, and custom scheduling algorithms, for real-time use-cases, demand prediction, custom function execution, and the like. The disclosure mitigates overhead due to startup and shutdown delays, which may be incurred by opening and closing a database connection, loading and unloading a neural network to physical memory, and the likes.

Abstract: Management of test case champions includes generating a plurality of fingerprints for a plurality of test cases, wherein each of the plurality of fingerprints is associated with one of the plurality of test cases and identifies a specific code path covered by its associated test case for a System Under Test (SUT). Based on the fingerprints and based on version control information for the SUT, one or more test cases in the plurality of test cases that cover a portion of code of the SUT that has changed are identified. Based on one or more criteria and test execution results, at least one test case of the identified one or more test cases is designated as a champion test case.

Abstract: In some implementations, a computing device may obtain a set of log lines associated with operations performed at a computing device. The computing device may select a first sample of the set of log lines. The computing device may identify a first set of log templates from the first sample, the first set of log templates having a first quantity of log templates. The computing device may select a second sample of the set of log lines. The computing device may identify a second set of log templates from the second sample, the second set of log templates having a second quantity of log templates. The computing device may apply the first set of log templates to the set of log lines for log templatization based at least in part on the first quantity of log templates being equal to the second quantity of log templates.

Abstract: Methods, apparatuses, and systems for interleaving cache line data transfers and handling misses of speculatively fetched data are disclosed, including receiving, by a first cache, one or more data requests from a requesting device; transferring, by the first cache in response to the one or more data requests, data of a first cache line to a second cache; interrupting the transfer of data of the first cache line by transferring a first portion of data of a second cache line to the second cache; and transferring, by the first cache, remaining data of the first cache line to the second cache.

Abstract: Replication links are selected from multiple asymmetric replication links for a specified tenant according to pre-defined service requirements to satisfy a wide variety of replication services. Link selection is based on applying predicted performance of the asymmetric replication links, which is derived from both static and dynamic performance parameters. A link rating model is trained on static parameter values and historic dynamic parameter values. Current dynamic parameter values and the static parameter values for a given replication link and input into the link rating model to generate predicted performance metrics and to establish a performance rating for the asymmetric replication links. The rated replication links are mapped to various pre-defined tenant service requirements. In that way, a tenant may specify a service requirement, and the replication link engine identifies one or more rated replication links that will satisfy the service requirement.