International Business Machines Corporation

Honeywell International Inc.

Deepak R Kenchammana-Hosekote

Divyesh Jadav

Jaishankar Moothedath Menon

Omer Ahmed Zaki

Richard Andrew Golding

Claudio Fleiner

Richard Francis Freitas

Hovey Raymond Strong

Allalaghatta Pavan

Nemmara R Vaidyanthan

James Lee Hafner

Dulce Beatriz Ponceleon

Marc Thadeus Roskow

Dingshan He

Garth R Goodson

The embodiments of the invention provide a method, apparatus, etc. for a cache arrangement for improving RAID I/O operations. More specifically, a method begins by partitioning a data object into a plurality of data blocks and creating one or more parity data blocks from the data object. Next, the data blocks and the parity data blocks are stored within storage nodes. Following this, the method caches data blocks within a partitioned cache, wherein the partitioned cache includes a plurality of cache partitions. The cache partitions are located within the storage nodes, wherein each cache partition is smaller than the data object. Moreover, the caching within the partitioned cache only caches data blocks in parity storage nodes, wherein the parity storage nodes comprise a parity storage field. Thus, caching within the partitioned cache avoids caching data blocks within storage nodes lacking the parity storage field.

Cache arrangement for improving raid I/O operations

A fault-tolerant and efficient way of deducing a set of inconsistent stripes for a network RAID protocol, wherein clients forward input/output (I/O) to a particular controller device called the coordinator, which executes RAID logic and which sends out device IOs to the relevant storage devices. If the coordinator fails then a new coordinator reconstructs its state from the storage devices.

System and method for fault tolerant controller for network raid

System and method for fault tolerant controller for network RAID

A client node of a distributed storage system adaptive determines on an operation-by-operation basis whether to process a network-RAID IO operation (IO request) locally at the client node or to forward the IO request to a storage server of the system for processing. The client node determines to process the requested network-RAID operation at the client node when a local estimator for a type of the requested network-RAID operation is less than a central estimator for the type of the requested network-RAID operation. Otherwise, the client node determines to process the requested network-RAID operation at a storage server when the local estimator for the type of the requested network-RAID operation is greater than the central estimator for the type of the requested network-RAID operation.

Decision mechanisms for adapting RAID operation placement

A collective storage system and method for restoring data in the system after a failure in the system. The system includes multiple storage nodes that are interconnected by a network and store data as extents. There are also a set of Data Service (DS) agents for managing the extents, a set of Metadata Service (MDS) agents for managing metadata relating to the nodes and the extents, and a Cluster Manager (CM) agent in each node. After a node failure is detected by one of the CM agents, the agents responsible for coordinating the data restoring are notified of the failure. The agents generate a plan to restore the data extents affected by the failure, and then collectively restoring the affected extents based on the generated plan. The coordinating agents might be the MDS agents or DS agents. The failure might be a node failure or a disk failure.

Distributed storage system capable of restoring data in case of a storage failure

A network scheduler for real time applications is described. The network scheduler supports scheduling of network packets in a multi-threaded operating system to insure correct scheduling order and to preclude kernel preemption of active threads. The network scheduler resides between a kernel (and device drivers) and multiple sessions at the user-level. The network scheduler uses input/output control primitives such as HOLD and RELEASE to control servicing of the queues of packets at the kernel or network device driver. The control primitives serve as a middleware interface to the underlying kernel or network device driver. The network scheduler functions as a proxy scheduler to manage the queues of packets at the kernel or network device driver.

Network scheduler for real time applications

A distributed block-based programming model for control applications such as multimedia based real-time applications is described. According to one aspect of the invention, user program specifies interconnections between a plurality of program blocks distributed across a plurality of processing nodes. The user program is automatically translated into a system-level program having a program fragment for each one of the processing nodes. Each one of the program fragments comprise the program blocks located on the processing node and one or more system-level blocks for establishing connections between the processing node and a different processing node across a network.

System for constructing and distributing block-based fragments

Multiple competing processors cooperatively manage access to a shared resource. Each processor separately stores a lock table, listing shared resource subparts, such as memory addresses of a data storage device, for example. The lock tables are stored in nonvolatile storage. In each lock table, each subpart is associated with a &ldquo;state,&rdquo; such as; LOCAL or REMOTE. In response to access requests from the hosts, the processors exchange various messages to cooperatively elect a single processor to have exclusive access to the subparts involved in the access requests. After one processor is elected, the lock-holding processor configures its lock table to show the identified subpart in the LOCAL state, and all non-lock-holding processors configure their lock tables to show the identified subpart in the REMOTE state. Thus, rather than replicating one lock table for all processors, the processors separately maintain lock tables that are coordinated with each other. Importantly, each processor honors its lock table by refraining from accessing a subpart of the shared resource unless the processor's lock table indicates a LOCAL state for that subpart. In one embodiment, optimized for the two processor environment, the messages exchanged by the processors include lock request, lock release, and lock grant messages.

Method for managing concurrent processes using dual locking

Disclosed is a system for updating and destaging data in a storage system. To overcome the limitations in the prior art described above, preferred embodiments of the present invention disclose a system for updating a data block in a storage device. A first processing unit receives an update to a data block in a storage device. The first processing unit sends a first message including the update and information indicating data blocks that were recently updated to a second processing unit. The second processing unit stores the update in a second storage area and sends a second message to the first processing unit after receiving the first message. The first processing unit stores the update in a first storage area and transfers the update from the first storage area to the data block in the storage device. The first processing unit transfers the update from the first storage area to the data block in the storage device. Data may be destaged from the first storage area by sending two messages before destage begins and two messages after destaging is completed so the second processing unit is aware of the state of the destaging operations.

System for updating data in a multi-adaptor environment

Disclosed is a system for altering the structure of parity groups, e.g., altering the RAID level or number of storage devices included in the RAID array. A parity group consists of a set of data blocks and parity data for the data blocks. The data and parity data are stored in at least two storage devices. The first processing unit alters the parity structure of an Nth set of parity groups, wherein N is an integer value greater than or equal to zero. A set of parity groups includes at least one parity group. The second processing unit determines an Nth range of parity groups including the Nth set of parity groups. After detecting the failure of the first processing unit while altering the parity structure of the parity groups in the Nth set, the second processing unit processes the parity groups in the determined Nth range to determine a parity group whose parity structure was being altered when the first processing unit failed.

System for changing the parity structure of a raid array

To satisfy host requests, two competing processors self-manage access to a shared resource. Each processor maintains a lock table listing that processor's access state regarding the shared resource. Each processor repeatedly sends the other processor a state announcement message representing the processor's state. These include birth cry, heartbeat, or death knell messages. Whenever certain join-entry conditions apply to a processor, the processor performs a JOIN operation to evaluate and attempt to synchronize its lock table with the other processor's lock table. If data is needed from the other processor's lock table, the processor changes its state announcement message to birth cry. If the processor successfully synchronizes its lock table with the other processor, the JOIN operation dictates DUAL mode for that processor. If the synchronization attempt fails, SOLO results. When JOIN completes, the processor operates in its chosen mode. In DUAL, the processor's state announcement message is heartbeat, signifying healthy operation. In DUAL, the processors cooperatively establish temporary single-processor locks on the shared resource to satisfy host requests. In SOLO, a processor sets its state announcement message to death knell, advising the other processor not to access the shared resource. Also in SOLO, the processor attempts to 'acquire' the shared resource by storing a predetermined identifier thereon. In SOLO, a processor refrains from accessing the shared resource unless it has acquired it.

Growing community of inventors

Sunnyvale, CA, United States of America

Deepak R Kenchammana-Hosekote