International Business Machines Corporation

Martin Petermann

Slavisa Sarafijanovic

Tohru Hasegawa

Hiroshi Itagaki

Atsushi Abe

Robert Beverley Basham

Alessandro Sorniotti

Elli Androulaki

Thomas D Weigold

Takashi Ashida

Takeshi Ishimoto

Harold Joseph Roberson, Ii

Mathias Bjoerkqvist

Mark Alfred Lantz

Charalampos Pozidis

Nikolaos Papandreou

Simeon Furrer

Jan Van Lunteren

Wayne A Sawdon

Ilias Iliadis

Milos Stanisavljevic

Ning Wang

Bo Zou

Feng Shao

Nikola Knezevic

Jin Yin

Jiong Lu

A method of dynamically optimizing decision tree inference is provided. The method, which is performed at the computerized system, repeatedly executes one or more decision trees for inference purposes and repeatedly performs an optimization procedure according to two-phase cycles. Each cycle includes two alternating phases, i.e., a first phase followed by a second phase. The decision trees are executed based on a reference data structure, whereby attributes of nodes of the decision trees are repeatedly accessed from the reference data structure during the first phase of each of the cycles. First, the accessed attributes are monitored during the first phase of each cycle, which leads to update statistical characteristics of the nodes. Second, a substitute data structure is configured during the second phase of each cycle based on the updated statistical characteristics. Third, the reference data structure is updated in accordance with the substitute data structure.

Dynamically optimizing decision tree inferences

Provided are a computer program product, device, system, and method for write data management. In one embodiment, a write management controller calculates an order for writing data and parity blocks of write data code words on tape media in a variable interleaved fashion which ensures a minimum physical longitudinal separation along the length of tape between data and parity blocks of the same code word. The calculated write order information may be provided back to a host or other data source so that the write data is eventually transmitted to the tape drive for writing on tape media in accordance with the calculated write order. In this manner, a minimum physical longitudinal separation along the length of tape between data and parity blocks of the same code word may be achieved to increase the reliability of the data storage.

Write data management

A computer-implemented method, according to one embodiment, is for integrating magnetic tape storage with a distributed disk file system. The computer-implemented method includes: registering for a subset of data operations that are recorded at a central location, where the central location is coupled to a plurality of distributed accessor nodes. A subset of available resources in one or more magnetic tape libraries that are coupled to the plurality of distributed accessor nodes are registered for and further managed. Moreover, a performance of the subset of data operations using the registered subset of available resources is scheduled. Supplemental data operations which correspond to the subset of data operations are also automatically received from the central location.

Magnetic tape integration with distributed disk file systems

A key-value store for storing and retrieving user files based on key-value pairs, hereafter referred to as KVPs. For each user of the key-value store, each file of a set of files of said each user is stored as one or more pairs of KVPs. Each of said one or more pairs comprises a KVP of a first type and a KVP of a second type. The KVP of the first type comprises at least a part of contents of said each file, whereas the KVP of the second type comprises metadata (and possibly attributes) of said each file. Each KVP of the second type links to one or more KVPs of the first type. Further provided are related method and computer program products.

Key-value store for managing user files based on pairs of key-value pairs

A multi-tiered file locking service provides file locking at the thread and process level, and can optionally include locking at the file system level. A local locking mechanism maintains a list of local locks for threads within a process. When a thread requests a lock for a file, and a local lock is obtained, a process lock for the file may be requested. When no file system locking is used, when the process lock is obtained, the thread receives the lock for the file. When file system locking is used, when the process lock is obtained, a file system lock for the file may be requested. When the file system lock for the file is obtained, the thread receives the lock for the file. The result is a file locking service that functions across threads, processes and nodes in a distributed computing environment.

Multi-tiered file locking service in a distributed environment

A system according to one embodiment includes a processor and logic integrated with and/or executable by the processor. The logic is configured to cause the processor to store hierarchically-organized global configuration information for each node and each tape library resource in a storage cluster to at least one memory accessible by each node of the storage cluster. The storage cluster includes at least one tape library. Also, the logic is configured to cause the processor to migrate data to and/or recall data from a tape cartridge pool within a tape library, using the hierarchically-organized global configuration information and via a node which has access to the tape cartridge pool, in response to receiving a migration and/or recall request at any node of the storage cluster. Other systems, methods, and computer program products for management of data and resources in a tiered data storage system are described in more embodiments.

Scalable and efficient access to and management of data and resources in a tiered data storage system

In a general embodiment, a method includes storing hierarchically-organized global configuration information for each node and each tape library resource in a storage cluster to at least one memory accessible by each node of the storage cluster, the storage cluster comprising at least one tape library. The method further includes migrating and/or recalling, using the hierarchically-organized global configuration information, data to and/or from a tape cartridge pool within a tape library. The migration or recall is performed by an appropriate node of the storage cluster and in response to receiving a migration or recall request at an arbitrary node of the storage cluster.

Growing community of inventors

Zurich, Switzerland

Martin Petermann