International Business Machines Corporation

Tatsuhiro Chiba

Sunyanan Choochotkaew

Hiroshi Horii

Hitomi Takahashi

Takeshi Yoshimura

Tamiya Onodera

Mikio Takeuchi

Qi Zhang

Kazunori Ogata

Michihiro Horie

Tatsushi Inagaki

Yohei Ueda

Marcelo Carneiro Do Amaral

A computer-implemented method for building socket transferring between containers in cloud-native environments by using kernel tracing techniques is provided including probing a connection-relevant system call event by using an eBPF to collect and filter data at a router, creating a mirror call at a host namespace with a dummy server and dummy client by creating the dummy server with mirror listening parameters, sending a server host address mapping to overlay the server host address to the client coordinator in an overlay process, and creating and connecting the dummy client to return a client host address to the server coordinator. The method further includes transferring mirror connections to the overlay process via a forwarder by temporary namespaces entering and injecting socket system calls and probing a transfer call event to map an overlay socket with a transferred dummy socket to activate duplication when the overlay socket is not locked.

Socket transferring for HPC networks using kernel tracing

A computer-implemented method for scheduling and scaling a cloud system for microservice applications is provided including identifying a plurality of nodes within one or more clusters associated with a plurality of containers, generating a model for predicting resource usage among the plurality of nodes, automatically deciding on a number of replicated containers, node bindings, and weight for each replicated container according to application requests and current usage status of a cluster of the one or more clusters that reduce resource usages and microservice cohesion, and determining at least node redistribution of the plurality of nodes within the plurality of containers and workload partitioning to reconfigure scaling, scheduling, and balance deployment requirements of the microservice applications.

Full-dimensional scheduling and scaling for microservice applications

Networking methods and systems include determining a first state of a connection on a first network based on connection buffers at a host. A first system call relating to the connection is identified. A next state of the connection that would result from the first system call is determined. The first system call is executed responsive to a determination that the next state does not move the connection farther from a safe transition state.

Live socket redirection

A computer-implemented method for detecting bottlenecks in microservice cloud systems is provided including identifying a plurality of nodes within one or more clusters associated with a plurality of containers, collecting thread profiles and network connectivity data by periodically dumping stacks of threads and identifying network connectivity status of one or more containers of the plurality of containers, classifying the stacks of threads based on a plurality of thread states, constructing a microservice dependency graph from the network connectivity data, aligning the plurality of nodes to bar graphs to depict an average number of working threads in a corresponding microservice, and generating, on a display, an illustration outlining the plurality of thread states, each of the plurality of thread states having a different representation.

Detecting layered bottlenecks in microservices

A new host is detected being added to a network cluster, wherein each of a plurality of hosts are on the network cluster. Available interfaces on each of the plurality of hosts on the network cluster are detected responsive to detecting the new host being added. A classless inter-domain routing (CIDR) range is calculated for hosts and interfaces on the network cluster using the available interfaces. Pod routes with interface range and L3 host routes are set for each host.

Network plugin for multiple network interfaces

A method is provided for generating a classification model configured to select an optimal execution combination for query processing. The method provides, to a processor, training queries and different execution combinations for executing the training queries. Each different execution combination involves a respective different query engine and a respective different runtime. The method extracts, from a set of Directed Acyclic Graphs (DAGs) using a set of Cost-Based Optimizers (CBOs), a set of feature vectors for each of the plurality of training queries. The method adds, by the processor to each of merged feature vectors a respective label indicative of the optimal execution combination based on actual respective execution times of the plurality of different execution combinations, to obtain a set of labels. The method trains, by the processor, the classification model by learning the set of merged feature vectors with the set of labels.

Selecting an optimal combination of systems for query processing

A computer-implemented method for increasing record availability on a distributed database system implementing multiple partitioning strategies includes identifying first processing devices of a distributed database system for storing one or more first records in accordance with multiple partitioning strategies, determining that at least two of the first processing devices are identified for storing a given one of the one or more first records, and selecting one or more new processing devices of the distributed database system for storing the given first record to achieve sufficiently high availability of the given record within the distributed database system.

Record replication for multi-column partitioning on distributed database systems

A computer-implemented method is provided for compilation rescheduling from among four compilation levels comprising level 1, level 2, level 3, and level 4 on a parallel distributed computing framework running processes for a plurality of jobs of a virtual machine. The method bypasses a program analysis overhead that includes measuring a compiled method execution time by identifying completed compilation levels of a Just In Time compilation. The method finds a repetition of a same process in the processes for the plurality of jobs of the virtual machine from profiles by comparing main class names, virtual machine parameters, and Jar file types therein. The method applies a compilation scheduling for the same process a next time the same process runs based on a result of the checking the transition, by (i) compiling at the level 1 at least some methods for the same process responsive to the virtual machine finishing without compiling the at least some methods for the same process at the level 4 after compiling the at least some of the methods at a level in between the level 1 and the level 4, and (ii) compiling at the level 4 at least a subset of the methods earlier than an original scheduled time responsive to at least the subset of the methods compiled at the level 4 being infrequently invoked below a threshold amount.

Rescheduling JIT compilation based on jobs of parallel distributed computing framework

A computer-implemented method includes identifying a set of anomalies during a runtime of a first set of replica application containers. The computer-implemented method further includes determining a root cause of the set of anomalies based, at least in part, on comparing uniform resource identifier (URI) requests generated by each replica application container in the set of replica application containers. The computer-implemented method further includes determining a first set of URI requests can be dropped, wherein each URI request in the first set of URI requests includes the root cause of the set of anomalies. The computer-implemented method further includes dropping the first set of URI requests during the runtime of the first set of replica application containers. A corresponding computer system and computer program product are also disclosed.

Prevention of application container failure between replicated containers

A method is provided for generating a classification model configured to select an optimal execution combination for query processing. The method includes providing training queries and different execution combinations for executing the training queries. Each different execution combination involves a respective different query engine and a respective different runtime. The method includes extracting, using Cost-Based Optimizers (CBOs), a set of feature vectors for each training query. The method includes merging the set of feature vectors for the each of the training queries into a respective merged feature vector to obtain a set of merged feature vectors. The method includes adding, to each of the merged feature vectors, a respective label indicative of the optimal execution combination based on actual respective execution times of the different execution combinations, to obtain a set of labels. The method includes training the classification model by learning the merged feature vectors with the labels.

A method is provided for improving a hit ratio of a buffer cache in a system in which vertices of a DAG have tasks that generate intermediate data stored in the buffer cache. The method tracks (i) a buffer cache usage by vertices that have finished running and (ii) a current available buffer cache space. Responsive to a new task being runnable and having dependent parent vertices, the method estimates a total buffer cache usage of current running vertices based on a partial result of the current running vertices. Responsive to the estimate exceeding current available buffer cache space, the method (i) selects a vertex having a most amount of intermediate data stored in the buffer cache for its dependent parent vertices, and (ii) increases a priority of the tasks in the selected vertex to obtain prioritized tasks. The method executes the prioritized tasks earlier than other remaining runnable tasks.

Lazy data loading for improving memory cache hit ratio in DAG-based computational system

A computer-implemented method is provided for container sharing and scheduling in an environment having a plurality of nodes. The method includes deploying a container management service system that manages a list of layered images. The list specifies one or more container attributes for each layered image stored at the nodes. The method includes selecting, by a processor-based scheduler of the container management system based on the list of layered images, a given one of the nodes which (i) includes image layers residing in a container and (ii) communicates with one or more candidate nodes that include one or more missing image layers. The candidate nodes are determined from among the nodes based on one or more availability criterion. The method includes pulling the one, or more missing image layers horn the candidate nodes and copying the one or more missing image layers to the given one of the nodes.

Sharing container images between mulitple hosts through container orchestration

A computer-implemented method is provided for container sharing and scheduling in an environment having a plurality of nodes. The method includes deploying a container management service system that manages a list of layered images. The list specifies one or more container attributes for each layered image stored at the nodes. The method includes selecting, by a processor-based scheduler of the container management system based on the list of layered images, a given one of the nodes which (i) includes image layers residing in a container and (ii) communicates with one or more candidate nodes that include one or more missing image layers. The candidate nodes are determined from among the nodes based on one or more availability criterion. The method includes pulling the one or more missing image layers from the candidate nodes and copying the one or more missing image layers to the given one of the nodes.

Growing community of inventors

Tokyo, Japan

Tatsuhiro Chiba