International Business Machines Corporation

Symantec Operating Corporation

Veritas Operating Corporation

Hewlett-packard Company

Robert Baird

Anand A Kekre

William H Tetzlaff

Gerald P Bozman

James J Myers

Alexander S Lett

Ronald S Karr

Oleg Kiselev

John Colgrove

Raghu Krishnamurthy

George Eisenberger

Douglas L Voigt

Santosh Ananth Rao

Marvin D Nelson

Michael B Jacobson

Gopal Sharma

Paul Hodges

Deepak Tawri

Jay H Unger

Richard Roland Guyette

Chio Fai Aglaia Kong

Pong-sheng Wang

Deepak K Tawri

Robert R Berbec

Nancy Y Young

Jay G Unger

Hamid Vazire

John W Fordemwalt

John J McAssey

Albert Kamerman

Thomas Beretvas

Applications executing on various nodes in a distributed storage environment may write data to primary storage and may also replicate the data to secondary storage via a replication target. An interval coordinator may coordinate the periodic saving of checkpoints or snapshots of the replicated data. The interval coordinator may determine the length of consistency intervals between the saving of each of the checkpoints. Writes to the replication target from each of the nodes may be associated with the current consistency interval and, in some embodiments, with a unique per-node sequence number. When transitioning between consistency intervals, each node may be configured to temporarily suspend completion of the writes and to send the replication target a consistency interval marker indicating that the node has completed all writes for the current consistency interval.

System and method for replication using consistency interval markers in a distributed storage environment

A system for ensuring availability using volume server sets in a storage environment employing distributed block virtualization includes a plurality of volume servers, one or more volume clients, one or more physical block devices and a volume server manager. The volume server manager may be configured to designate the plurality of volume servers as a volume server set with an associated volume server set management policy, and to verify each volume server in the volume server set has access to storage within each block device. In addition, the volume server manager may be configured to designate a first volume server of the volume server set to aggregate storage within the block devices into a logical volume, to make the logical volume accessible to the volume clients, and to share configuration information about the volume with the other volume servers of the volume server set.

System and method for providing availability using volume server sets in a storage environment employing distributed block virtualization

A system and to prevent data corruption due to split brain in shared data clusters includes two or more nodes of a cluster, a shared storage device, and an update manager. The update manager may be configured to maintain a local persistent store corresponding to each node of the cluster. On receiving an update request directed to the shared storage device from a first node, the update manager may be configured to redirect the update to the local persistent store corresponding to the first node. The update manager may be further configured to verify a cluster membership status of the first node, and to transfer the contents of the update from the local persistent store to the shared storage device if the cluster membership verification succeeds.

System and method to prevent data corruption due to split brain in shared data clusters

Various systems and methods for maintaining write order fidelity in a distributed environment are disclosed. One method, which can be performed by each node in a cluster, involves associating a current sequence number with each of several write operations included in a set of independent write operations. In response to detecting that one of the write operations in the set is ready to complete, a new sequence number is selected, and that new sequence number is thereafter used as the current sequence number. None of write operations in the set is allowed to return to the application that initiated the write operations until the new sequence number has been advertised to each other node in the cluster. The method also involves receiving a message advertising a first sequence number from another node in the cluster, and subsequently using the first sequence number as the current sequence number.

System and method for maintaining write order fidelity in a distributed environment

One goal of consistency interval replication is to achieve a consistent copy of data generated by independent streams of writes from nodes in a clustered/distributed environment. Two writes to the same block from different nodes may arrive at a replication target in a different order from the order in which they were written to primary storage. A consistency interval coordinator may analyze a list of blocks modified during a consistency interval to determine conflict blocks written to by two different nodes during the same consistency interval. Conflict resolution may involve a node reading data for a conflict block from primary storage and forwarding it to the replication target or a node completing a suspended in-progress write for the conflict block. Once the conflicts have been resolved, the replication target may checkpoint the data modified during the interval and nodes may resume writes to the conflict blocks for the new interval.

System and method for block conflict resolution within consistency interval marker based replication

A system for data transfer using a recoverable data pipe includes a data producer, one or more data consumers, a storage device and a data pipe manager. The data producer may be configured to append updated data blocks to the storage device via a producer storage input/output (I/O) channel. The data consumers may be configured to read data blocks from the storage device via consumer storage I/O channels. The data pipe manager may be configured to maintain metadata identifying data blocks of the storage device that have been read by each data consumer, and to release backing storage corresponding to the data blocks that have been read by all data consumers.

System and method for data transfer using a recoverable data pipe

A log-structured temporal shadow store may comprise a logical storage aggregation including a plurality of blocks, a log-structured storage device, and shadow management software. The log-structured storage device may include a plurality of log entries, where each log entry includes one or more modified blocks of the logical storage aggregation and an index to the modified blocks. In response to a new batch of changes to the logical storage aggregation, the shadow management software may be configured to append a new log entry to the log-structured storage device, including newly modified blocks and an index to the newly modified blocks. The index may be organized as a modified B+ tree, and the log-structured storage device may be a logical volume, such as a mirrored logical volume.

Log-structured temporal shadow store

A system for loosely coupled temporal storage management includes a logical storage aggregation including a plurality of data blocks, a data producer, one or more data consumers, and a temporal storage manager. The temporal storage manager may be configured to maintain a producer shadow store including entries stored in a log-structured logical volume, where each entry is indicative of one or more data blocks of the logical storage aggregation that have been modified by the data producer. The temporal storage manager may also be configured to maintain a repository containing a baseline version of the logical storage aggregation, and to provide the data consumers with read-only access to the producer shadow store and the repository.

System and method for loosely coupled temporal storage management

A method for dynamically detecting loss of map integrity in a form of system-managed storage (SMS). In SMS, maps are used to define access paths to data and to allocate and reallocate storage resources among applications running thereon. The method steps include incorporating as an indivisible part of an overwriting commmand the duplication of map information by appending a portion of it to each data block in store, and detecting loss of map integrity as a function of a comparison mismatch between the portion stored with a counterpart data block and the map upon each read/write access.

Method for verification and restoration of directories in CPU system

A computer implemented method for robustly copying pages to system managed storage in order to maintain data in a consistent state and in order to provide continuous access availability of the pages to executing applications. The method achieves data consistency by atomically shadow copying application referenced pages and amending directories in a failure independent medium on (1) an access path interrupt as well as on (2) a page update basis. Availability is enhanced by duplexing the pages and directories as part of the atomic shadow copying step.

Highly available fault tolerant relocation of storage with atomicity

A method for managing concurrency using a serializing token as a supplement to locks for accessing the same page by different processes and ensuring coherence between data caches and a shared access backing store supporting the processes defined onto multiple processors. A shared lock and a new token are issued by a local lock manager (LLM) and local cache manager (LCM) in response to a page read request from a process. An exclusive lock, an invalidation of all cache resident copies of the page, a cache write through to backing store of the changed page, a copyback of a new token, and lock release are responsive to a page write/update request from a process where the token issued to the process during a prior request matches the token stored with a cache resident copy of the page. Otherwise, a write request fails and the process must first issue a read request to the page of interest.

Concurrency management using version identification of shared data as a

A disk array has a plurality of disks, a disk array controller for coordinating data transfer to and from the disks, and a RAID management system for mapping two different RAID areas onto the disks. The RAID management system stores data in one of the RAID areas according to mirror redundancy, and stores data in the other RAID area according to parity redundancy. The RAID management system then shifts or migrates data between the mirror and parity RAID areas on the disks in accordance with a predefined performance protocol, such as data access recency or access frequency.

Memory systems with data storage redundancy management

A method for managing space re-use with respect to the indices (nodes) of shadow written tree organized dynamic random accessed files/records/pages located in the external store of a CPU. The method reserves space in all non-leaf nodes and maintains a list of available node addresses. When a new node is required then space, if available, is obtained from the parent node list. Only when the parent list becomes exhausted is space (node) obtained from a node inventory manager. Deletion of a node causes its address to be placed on the free or available list maintained by that node's parent. If there is no space, then space on the parent node list is obtained by returning to the inventory manager that node on the list having the least locality with the existing subordinate (children) nodes of the parent.

Managing locality in space reuse in a shadow written B-tree via interior

A conditional broadcast or notification facility of a global lock manager is utilized to both serialize access to pages stored in local caches of counterpart processors in a distributed system and to ensure consistency among pages common to the caches. Exclusive use locks are obtained in advance of all write operations. When a page is to be updated, which page is cached in a processor other than that of the requester, then a delay is posed to the grant of the exclusive lock, all shared use lock holders to the same page notified, local copies are invalidated, exclusive lock granted, page is updated and written through cache, after which the lock is demoted to shared use.

Multiprocessor system includes operating system for notifying only those

A computer-implemented method for the name-oriented accessing of files having at least zero records, any access path to files and records through an external store coupling the computer being defined by a pair of related directories. A first directory of record entries is sorted on a two-part token. The token consists of a unique sequence number assigned to the record and the sequence number of any parent record entry. Each record entry includes the token, file or record name, and external store address or pointer A traverse through the tokens constitutes a leaf-searchable B-tree. Rapid access to target records is by way of a name-sorted, inverted directory of names and tokens as a subset and which is reconstitutable from the first directory in the event of unavailability.

Growing community of inventors

San Jose, CA, United States of America

Robert Baird