Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
1999-08-26
2004-06-15
Robinson, Greta (Department: 2177)
Data processing: database and file management or data structures
Database design
Data structure types
C707S793000
Reexamination Certificate
active
06751634
ABSTRACT:
TECHNICAL FIELD
This invention relates generally to replication between networked servers or computers and, more particularly, relates to a method to detect partially replicated objects in a database in order to ensure that such objects are used in a consistent state in distributed database implementations.
BACKGROUND OF THE INVENTION
Use of distributed computer networks to implement databases results in more responsive databases, including local updating and management in multiple master systems. Redundancy built into a distributed implementation results in a more resilient and reliable database. A database may be thought of as constituting two essential properties, viz., a collection of data objects and a set of procedures/tools/methods for locating a particular one of the data objects. The procedures/tools/methods for locating a particular data item are included in a directory or directory service. There may be more than one directory in a database, and possibly a directory service for assistance in locating the proper directory. The directory service is a namespace that aids in resolution of an identifier to a corresponding object, or even attribute, using appropriate mappings that include simple directory tables. Commercial databases typically include storage for the objects and implementations of directory services for navigating the directories describing the organization of the stored data in the database.
Databases are central to meaningful transactions on a computer regardless of whether a computer is networked or not. A file stored in an ordinary personal computer hard drive is, ultimately, an entry in a database accessed through a file access table having file names and a plurality of memory locations. Moreover, changes to this file may be incremental in that the entire file is not rewritten immediately, but instead, a record is made of the desired changes. If two different entities try to access and modify a typical file, traditionally only one is permitted with the other one being locked out. This is a safety feature designed to avoid inconsistent changes that may interfere with the operation of software. Commercial databases are designed for different contexts but often have similar concerns during their use by client applications.
In order to minimize the risk of service disruption and facilitate faster access, commercial databases are preferably maintained on several networked computers, playing the role of servers, which update each other's versions. These updating operations between computers are termed replications. If a particular computer crashes, another computer takes-over. Users of the database may actually be only aware of a database-implementing network as a single entity rather than the individual components that make up the network.
A replication operation may result in the copying the entire state of a database to another database. However, this is often impractical and usually an incremental replication scheme (IRS) is implemented instead. In an IRS, changes since the last replication step, or portions thereof, are replicated to the other machine.
Generally, there are two commonly used strategies for carrying out an IRS, and there are numerous variations within each strategy. The first strategy utilizes the single-master server (SMS) model in which one of the computers is designated as the primary server (PS). One or more backup servers (BS) replicate from the PS. Changes are first made to the PS and are then propagated to the backup servers via replication. During replication between PS and BS, often a lock is placed on both to ensure that the data does not change while replication is taking place. Furthermore, the replicating components are usually unavailable for the duration of the lock. The reasons for placing a lock include the desire to avoid identifying two machines as being synchronized when one or both may change during the process for synchronization itself, with the possibility of the next incremental replication operation treating non-replicated data as already having been replicated.
The second strategy for IRS is based on networking multiple-master servers (MMS). In a multiple-master system, many servers replicate from each other, and changes may be independently made to their respective databases. This permits updates to be made locally with local management while the changes eventually are reflected in the entire network. Thus, in the MMS system there is no need for a primary server. Furthermore, a server may even restrict itself to a defined subset of the data while routing requests requiring additional data to other servers in the network.
A difficulty presented by MMS units is that changes to the database can occur, in any order, on any of the MMSs in a network. Furthermore, because many older networks utilize the single-master server model many hybrid networks may exist where the multiple master servers replicate to other multiple master servers while one of the multiple master servers emulates a primary server for the benefit of single master servers in the network. Further details about managing replication in a hybrid network are provided in the U.S. patent application Ser. No. 09/360,498, “METHOD AND SYSTEM FOR REPLICATION IN A HYBRID NETWORK,” filed on Jul. 26, 1999.
Many implementations of MMS networks are based on the multi-master loose consistency with convergence model in which a change to a replica propagates to other replicas but without any guarantees that at any given time the replicas will be consistent with each other. If no new changes are made then the system is guaranteed to converge on the same set of values. This model is attractive because it is compatible with highly distributed database implementations capable of growing to include millions of objects and thousands of replicas. In addition, the model also allows for inclusion of nodes with intermittent connectivity in the network.
Accessing inconsistent data can be a handicap for many applications. Inconsistent data among various replicas in multi-master loose consistency with convergence model is due to propagation time needed by the updates to reach the entire network. There is no known strategy for predicting the future state of a system based on such a model or even predicting when changes would be applied in the system resulting in a “nondeterministic latency,” which is a fact of life for applications operating in such an environment.
From the perspective of applications, which are the usual clients for database services, in a multi-master loose consistency with convergence model, there are three states that may be encountered. First is the possibility that none of the changes to a source have propagated to a destination replica resulting in a “version skew.” Applications have to expect such a possibility of having consistent but stale data. Second, only some of the changes applied to a source may have propagated to a destination replica resulting in a “partial-update.” An application retrieving data from the replication replica would benefit from knowing that it is a partial-update, specially, if consistency between data objects is important. Third, all of the changes applied at a source may have propagated to a destination replica resulting in a “fully replicated” state.
Directory-enabled applications use directory services to store configuration information, operational instructions or other information. It is possible that multiple copies of the application may be executing on the network at the same time, and as a consequence, reading/writing to replicas of the directory service. It is possible that some directory-enabled applications may use more than one replica. Some of these applications may handle version skew should an inconsistency be detected between replicas, provided partial-updates can be ruled out.
An exemplary application, sensitive to consistency concerns, is a remote access service using the directory to store policy and profile information. The policy information is stored in one set of objects, and the profile in another
Dodds, Jr. Harold E.
Leydig Voit & Mayer Ltd
Microsoft Corporation
Robinson Greta
LandOfFree
Method and system for detecting object inconsistency in a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for detecting object inconsistency in a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for detecting object inconsistency in a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3323580