International Business Machines Corporation

Globalfoundries Inc.

Nir Naaman

Avraham Harpaz

Yoav Tock

Gidon Gershinsky

Idan Zach

Eliezer Dekel

Alexander Krits

Lohitashwa Thyagaraj

Cornell G Wright, Jr

Foluso Olaiya Okunseinde

Chen Chen

Hilary A Pike

Benjamin Mandler

Ira Rosen

John J Duigenan

Ian Robinson

Kenneth Wayne Borgendale

Konstantin Shagin

Samuel Kallner

Nili Guy

Lior Turgeman

Venkata Kiran Kumar Darbha

Savitha Joshi

Adar Amir

Cornell Wright

Hilary Alison Pike

John Justin Duigenan

Liane Lewin-Eytan

Michael Masin

Adnan M Agbaria

Shahar Chen

Lian Lewin-Eytan

Harel Paz

Shmuel Kalner

Cornell Wright Jr

Techniques for generating a set of Deep Learning (DL) models are described. An example method includes training an initial set of DL models using the training data, wherein a topology of each of the DL models is determined based on the parameters vector. The method also includes generating a set of estimate performance functions for each of the DL models in the initial set based on the set of edge-related metrics, and generating a plurality of objective functions based on the set of estimated performance functions. The method also includes generating a final DL model set based on the objective functions, receiving a user selection of a selected DL model from the final DL model set, and deploying the selected DL model to an edge device.

Adaptation of deep learning models to resource constrained edge devices

An example system includes a processor to monitor system resources and performance preferences. The processor is to select model fragments based on the system resources and the performance preferences. The processor is to also construct a running artificial intelligence (AI) model from the selected model fragments. The processor is to further automatically modify the running AI model using the model fragments in response to detecting a change in the system resources or a change in the performance preferences.

Modifying artificial intelligence models using model fragments

An example system includes a processor to receive updates for a database of a publish and subscribe system. The processor is to generate an optimistic update based on the received updates. The processor is also to adaptively update the database using the optimistic update. In some examples, the processor generates a bulk optimistic update based on the optimistic update and adaptively updates the database using the bulk optimistic update or a read-modify-write sequence.

Adaptively updating databases of publish and subscribe systems using optimistic updates

Examples of techniques for tracking client sessions in a publish and subscribe system using a shared repository are described herein. An example system includes a processor of a server to receive a connection request with a client identifier from a client a predefined policy for duplicate client sessions. The processor is to query a shared repository of the system using the client identifier. The processor is to also execute a set of actions based on the predefined policy, a query result including a detected duplicate client session from the shared repository, and a set of parameters.

Tracking client sessions in publish and subscribe systems using a shared repository

Methods, systems and computer products for ensuring session correctness for subscribing clients connected to a publish-subscribe broker via a gateway having intermittent connectivity with the broker. The gateway is configured to: (1) subscribe to one or more topics of a publish-subscribe service controlled by the broker in response to one or more subscription sessions initiated by one or more subscribing clients connected to the gateway for subscribing to the topic(s), (2) identify session parameters indicative of activity state of each subscription session, (3) reconnecting to the broker following a disconnection period, (4) receiving from the broker one or more messages published on the topic(s) during the disconnection period, the broker correlates each message with message properties, and (5) publish each message to one or more of the subscribing clients based on compliance of the message properties of the respective message with the session parameters of the respective subscription session.

Enforcing session properties compliance for gateway connected publish-subscribe clients

Migration of a durable client in a publish/subscribe system from a first server of a cluster to a second server of the cluster. The method comprises the second server receiving subscription information of the durable client; the second server establishing a message queue for the durable client; the second server sending a switch command to all active servers of the cluster; in response to the second server obtaining, from an active server, an acknowledgement of the switch command, the second server inserting a marker to the message queue indicating switching point of the active server; and in response to the second server obtaining, from the first server, a message indicating the first server flushed all messages prior to respective switching points for all active servers, connecting the durable client to the second server and transmitting messages thereto by the second server. As a result, the durable client is migrated from the first server to the second server without disrupting data traffic to uninvolved clients.

Migration of durable clients in a clustered publish/subscribe system

A method comprising obtaining a Bloom filter array, wherein a size of a first Bloom filter is smaller than a size of a second Bloom filter and modifying at least a portion of the plurality of Bloom filters to create a modified Bloom filter array comprising a modified plurality of Bloom filters each of which having a same size, wherein the same size is greater than the size of the first Bloom filter, wherein said modifying comprises: duplicating the first Bloom filter to create a modified first Bloom filter comprising two or more concatenated copies of the first Bloom filter. The method may also include performing lookup in the Bloom filter array by computing a hash value of a key and accessing a bit in the hash value location in each Bloom filter of the modified Bloom filter array to determine lookup outcome in each Bloom filter.

Efficient lookup in multiple bloom filters

Method, system and product for efficient size reduction of a Bloom filter. The method includes: creating an interleaved Bloom filter based on a Bloom filter, wherein the interleaved Bloom filter is defined so as to have non-sequential cells of the Bloom filter be positioned sequentially, wherein the interleaved Bloom filter interleaves segments of the Bloom filter, wherein a size of the segments is based on an interleaving factor. A reduced size Bloom filter having a number cells below the number of cells of the Bloom filter is computed. The computation includes computing a value of a cell in the reduced size Bloom filter based on two or more values of cells in the interleaved Bloom filter.

Efficient size reduction of a bloom filter

A computer implemented method of improving server coverage of a publish-subscribe cluster of servers, comprising using a cluster comprising a plurality of servers each having hardware processor(s), the cluster is adapted to execute messaging engines of a publish-subscribe service for forwarding a plurality of messages to a plurality of subscribing clients, each message is associated with one of a plurality of topics, wherein distributed subscriptions client(s) applies shared subscriptions for one or more topics through subscription group(s) each comprising a subset of the subscribing clients, monitor shared subscriptions made by the subscription group(s) to each of the messaging engines for each topic, generate a shared subscriptions information for each topic by analyzing the shared subscriptions and publish the shared subscriptions information to the clients to allow client(s) of the subscription group(s) to subscribe for the topic(s) on a preferred messaging engine to improve a server coverage of their subscription group.

Cluster assisted MQTT client coverage for fat-pipe cloud applications

A computer-implemented method includes sending one or more network packets. Each of the one or more network packets includes a network packet header. The computer implemented method further includes receiving a negative acknowledgement list comprising the network packets not received. The computer-implemented method further includes, responsive to the receiving of a negative acknowledgment list: For each network packet of the negative acknowledgment list, transforming the network packet header into a modified packet header to yield a modified packet. The computer-implemented method further includes combining each modified packet into a modified packet list. The computer-implemented method further includes generating one or more repair packets. Each of the one or more repair packets further include a repair packet header and a portion of the modified packet list. The computer-implemented method further includes sending the one or more repair packets. A corresponding computer system and computer program product are also disclosed.

Adaptive network communication protocols

A computer implemented method of improving server coverage of a publish-subscribe cluster comprising using one or more hardware processors to execute one or more shared subscriptions hubs each adapted to retrieve shared subscriptions information from a cluster comprising multiple servers executing multiple messaging engines of a publish-subscribe service for forwarding messages to a plurality of subscribing clients, each message is associated with one or more of a plurality of topics, receive shared subscriptions requests for subscribing for one or more of the topics from distributed subscriptions client(s) applying shared subscriptions for the topic(s) through subscription group(s) comprising a subset of the subscribing clients, connect to preferred messaging engine(s) selected to improve server coverage of the subscription group(s) according to the shared subscriptions information to serve each of the plurality of shared subscriptions requests and forward messages associated with the topic(s) received from the selected messaging engine(s) to the subscription group(s).

MQTT cluster shared subscription hub for fat-pipe cloud applications

A method, product and apparatus for distributing retained messages information in a clustered publish/subscribe system. The method comprising obtaining a set of Retained Message Filters (RMFs) comprising a RMF for each server in the cluster, wherein the RMF representing a set of topics for which retained messages are retained by the server, the RMF being configured to approximate whether the server retains a retained message that matches a subscription; responsive to receiving a subscription request from a client, applying the set of RMFs to the subscription to obtain an indication of a set of servers that potentially retain retained messages that match the subscription; triggering each server in the set of servers to perform retrieval of retained messages that match the subscription; and sending to the client the retained messages retrieved from the servers in the set of servers.

Distributing retained messages information in a clustered publish/subscribe system

A publish-subscribe data messaging system in which wildcard subscriptions are represented in an approximate membership set or filter, such as a Bloom filter, for distribution to brokers together with formatting rules associated with the wildcard subscriptions, where a broker receiving a publication topic of a data publication for distribution to other brokers uses the formatting rules provided by any given broker to create wildcard versions of the publication topic for matching with the wildcard subscriptions that are represented in the approximate membership sets or filters provided by the given brokers, where, if a match exists, the data publication is forwarded to the given broker.

Publish-subscribe system with reduced data storage and transmission requirements

Enhancing performance in a multicasting system. Flow rates in a multicasting system are monitored as system characteristics change. The multicasting system is operated based on a first mapping between data flows and multicasting groups. Data is disseminated by way of the mapping to subscribers to one or more of the data flows. The flow rate thresholds of satisfied flows are decreased to minimize the difference between the flow rate thresholds and respective target transmission rates of the satisfied flows, as the system's feasible data transmission workload is maintained with respect to subscriber reception rate capacities. The flow rate thresholds of unsatisfied flows are increased to minimize the difference between the flow rate threshold and respective target transmission rates for the unsatisfied flows and to limit occurrences of data transmission delays or losses, as the system's feasible data transmission workload is maintained with respect to subscriber reception rate capacities.

Growing community of inventors

Haifa, Israel

Nir Naaman