Patent Application
20060020613
1. Field of the Invention
The present invention relates to the field of directory services. In particular, the present invention is directed to application of X.500, LDAP and similar services to a relational database, a database design and use of the database to perform X.500 services.
One aspect of the invention relates to the design of a directory service(s) system, including table organisation and clustering, and method of operation which facilitates queries in and out of the directory.
Other aspects of the present disclosure are directed to an implementation using a RDBMS (Relational Database Management System) and also a table structure and methods of operation of a database application.
2. Description of the Related Art
X.500 is the International Standard for Electronic Directories [CCITT89 or ITU93]. These standards define the services, protocols and information model of a very flexible and general purpose directory. X.500 is applicable to information systems where the data is fairly static (e.g. telephone directory) but may need to be distributed (e.g. across organisations or countries), extensible (e.g. store names, addresses, job titles, devices etc.), object oriented (i.e. to enforce rules on the data) and/or accessed remotely.
Relational Database Management System
(RDBMS) provide facilities for applications to store and manipulate data. Amongst the many features that they offer are data integrity, consistency, concurrency, indexing mechanisms, query optimisation, recovery, roll-back, security. They also provide many tools for performance tuning, import/export, backup, auditing and application development.
RDBMS are the preferred choice of most large scale managers of data. They are readily available and known to be reliable and contain many useful management tools. There is a large base of RDBMS installations and therefore a large amount of existing expertise and investment in people and procedures to run these systems, and so data managers are looking to use this when acquiring new systems. Most relational database products support the industry standard SQL (Structured Query Language).
There has also been a move towards Object Oriented systems which provide data extensibility and the ability to handle arbitrarily complex data items. In addition, many corporations and government departments have large numbers of database applications which are not interconnected. Data managers are looking for solutions which enable them to integrate their data, and to simplify the management of that data. X.500 and it's associated standards provide a framework and a degree of functionality that enables this to be achieved. The fact that X.500 is an international standard means that data connectivity can be achieved across corporations and between different countries.
The problem, therefore, is to address the need of data managers and implement X.500 with all the flexibility of object-oriented systems but using an SQL product so that it can achieve the scalability and performance inherent in relational systems coupled with the stability, robustness, portability and cost-effectiveness of current SQL products.
There have been a number of attempts of solving the above problem and over a considerable period of time. None of the attempts have resulted in a product which has proven to be commercially accepted by the market, and thus in the market place there is a long felt need yet to be addressed.
The Proceedings of IFIP WG6.6 International Symposium (ISBN: 0444 889 167) have published a paper presented by Francois Perruchond, Cuno Lanz, and Bernard Plattner and entitled “A Relational Data Base Design for an X.500 Directory System Agent”. The Directory System disclosed, as with many prior art systems, is relatively slow in operation, particularly where the database is relatively extensive and is incomplete in its implementation of X.500, such as aliases, subsearch and entry information.
Another attempt is disclosed in the proceedings of IREE, ISBN 0909 394 253, proceedings Apr. 22-24, 1991 by C. M. R. Leung. In that disclosure, there is described a database scheme in which a single entry table holds detailed information about each directory object, and is also incomplete in its implementation of X.500.
This approach has been discredited by a number of text books and knowledge in the art, such as “Object-Oriented Modeling and Design” by J. Rumbaugh, et al, 1991, ISBN 0-13-630054-5, in which at paragraph 17.3.8 it is clearly stated that “putting all entities in the one table is not a good approach to relational database design”.
As noted above, there have been a number of attempts made to address prior art problems, but none of the attempts have resulted in a product that has proven to be commercially accepted by the market. In particular, the prior art has been unable to address problems that are associated with queries in and out of the directory, table organisation and layout, and clustering.
An object of the present inventions is to address problems associated with the prior art relating to design of a directory service(s) system and method of operation which facilitates queries in and out of the directory, table organisation/layout and clustering, or at least one of the prior art problems.
The present invention provides a directory service arrangement in which tables are organised corresponding to their function. Preferably, a directory service is provided in which arbitrary data is processed using a fixed set of queries/services.
Furthermore, in the present invention each service is modeled, instead of each data type as seen in the prior art, and relationships are defined between each service, rather than between each data type as in the prior art. Implementation of service modeling using relational queries to satisfy directory services, such as X.500 services, enables the benefits of RDBMS to be optimally exploited.
Moreover, a directory service arrangement that is based on service modeling and functional organisation leads to clustering and a preferred layout design. In clustering, data of the same type and similar values are clustered in the same area of a storage medium, such as a disk. The number of disk pages that have to be accessed are then significantly reduced. The preferred, but not only, layout design that incorporates principal, conceptual, logical and/or physical designs, is subsequently described and illustrated in the accompanying tables and figures.
A further aspect of invention resides in providing a directory services database in which tables are arranged according to a conceptual design. The conceptual design includes at least attribute, object and hierarchy tables. The attribute table allows new attribute types to be defined by adding respective rows to this table. The object table defines the attributes within each object. The hierarchy table defines the relationship between objects.
A further aspect of invention is a method of arranging tables of a directory service, such as a X.500 or LDAP directory service, the method including:
Preferably, in applying service decomposition, at least one of the following consideration are made:
Preferably, the hierarchy table can be decomposed into at least 4 tables, comprising a DIT (Directory Information Tree including information required for navigation), NAME (including information required for returning names (raw RDN)), TREE (including information about an entry's path) and ALIAS (includes entries that are aliases), or their equivalent.
Preferably, the object table can be decomposed into at least 2 tables, namely SEARCH (used to resolve filters in a Search service and includes one row for each attribute value of each entry) and ENTRY (used to return values in Read and Search services and includes one row for each attribute value of each entry), or their equivalent.
Preferably, the invention includes a number of tables and designs having characteristics and relationships, as illustrated in
A detailed description of the present invention can be found at least in the summary of invention, and in sections 1, 2, 4, (particularly 4.1 and 4.6) and 6 of the description of the preferred embodiments, and the related Figures.
With regard to the remainder of the specification as a whole, in general, it seeks to disclose a number of other inventions related to the implementation of X.500 services in a RDBMS which supports SQL or any other relational language. X.500 services can be invoked via a number of protocols, such as X.500 and LDAP.
The scope of the present invention is outlined in this specification, including the claims.
In this document, at the time of filing, SQL is the most popular relational language and although it is only one form of relational language, the intent of the present invention is to have application to any other form of relational language, not just SQL.
These inventions can be related to the following headings:
The X.500 standard in no way dictates how the directory is to be implemented, only its capabilities and behaviour. One key to solving the implementation problem is the realisation that X.500 defines a fixed set of services (e.g. Add, Modify, Search etc.) that can operate on arbitrary data.
It has been discovered that problems associated with the prior art may be alleviated by a unique approach, by what may be described as inverting relational theory modeling from a data modeling approach to a service modeling approach. That is, from the problem of:
Each service is modelled (instead of each data type) and the relationships between each service defined (instead of the relationships between each data type).
Implementation of service modeling using relational queries to satisfy X.500 services enables benefits of RDBMS to be exploited.
The benefits of this approach are many. A summary is illustrated in
Based on this unique approach, the following disclosure will detail a number of inventions in an order with reference to
1. Principal Design
The X.500 prior art attempts at implementation have been unable to overcome the relatively basic structural and operational differences between the X.500 requirements and functionality and SQL. The X.500 standard has a particular structure by nature, whereas SQL is designed to operate on relational structured tables.
For a typical relational database application, the nature of data is well known, i.e. tables will consist of a number of columns and each column contains data relating to a particular data type (see Table B1). The different data types that can be stored is limited to the columns of the table. The data types are also limited to the types supported by the database (e.g. string, numeric, money, date). The database may also store data of a form not understood by the database per se, but understood by the application e.g. binary data.
If a new data type needs to be added (e.g. mobile) then a new column will have to be added to the table. This can cause problems if data table changes are not easy to implement. Also if the new data type is not well used (e.g. less than 1% of the organisation) then significant redundant data storage may result. See Table B2.
In essence, one invention in the application of X.500 resides in overcoming the extensibility by representing the X.500 attributes of the prior art:
A further development based on the above principal design is the adaption of the ‘principal design’ to X.500. This adaption has been realised by the provision of a ‘property table’, in which object name and parent name is added to the ‘principal design’.
Further benefits accrue from the implementation disclosed above; including:
2. Conceptual Design
The prior art has had difficulty in implementing X.500 as it has not been structured for extensibility, object oriented and hierarchy which are requirements of X.500.
This is addressed, in one form, by functionally decomposing the ‘property table’ and thus resulting in what is called the Conceptual Design.
The conceptual design resides in providing at least one of:
In another invention, this problem is addressed by providing table structures in accordance with those disclosed in
Yet further inventions reside in addressing problems of data tolerance by providing in the present X.500 system for the replacement of the ‘value’ column of the object table with value ‘norm’ and value ‘raw’ columns and/or replacing the RDN column in the hierarchy table with ‘name norm’ and ‘name raw’ columns.
Further, the difficulty in prior art of accommodating aliases is addressed in the present X.500 system by providing an ‘alias’ column in the hierarchy table. The ‘alias’ column is flagged to indicate that, that entry is an alias.
Further refinement may be provided by replacing the ‘alias’ column with alias and A-EID columns. The A-EID provides information about where the alias points.
Still further refinement may be provided by replacing the ‘parent’ column in the hierarchy table with ‘parent’ and ‘path’ columns.
The ‘path’ addresses the problem of implementing X.500 search, with aliases and subtrees. The ‘path’ has at least two unique properties: a) to determine the absolute position in the hierarchy; and b) it is used to determine if an entry is in a given subtree by its prefix.
3. Conceptual Method
A number of unique methods of interrogating the conceptual design are disclosed in the detailed description following, including:
A further invention is realised by using the attribute table for incoming data to find the AID from the X.500 object ID and outgoing data read from the database, vice versa.
Furthermore, for any incoming distinguished name, it is navigated to its appropriate EID, then each search is performed as required by X.500.
Still furthermore, for a search, filter and subtree searches can be provided by a single pass resolution and using the path column. One invention is to utilise a ‘path’ field to simultaneously apply an arbitrary filter over an arbitrary subtree. The complications of aliases is handled by applying the above method to a uniquely resolved subtree.
Yet another unique method is to store the “path” of each entry as a string. Each path will then be prefixed by the path of its parent entry. This is useful for the filter in the search service.
4. Logical Design
The logical design is based on a service decomposition of the conceptual design, though the realisation that X.500 service components are independent.
The advantages accruing from this include:
1. Reduces the number of indexes per table, as more tables are provided. It has been found that primary indexes are most efficient (speed, size) and secondary indexes may have large overheads (speed, size).
2. Enable data in tables to be clustered. Clustering occurs as a result of its primary key (storage structure) and thus data may be organised on disk around its key. E.g. for the ‘search’ table, surnames may be clustered together.
3. Management—smaller tables are easier to manage, e.g. faster to update indexes, collect statistics, audit, backup, etc.
4. Reduced I/O—speed improvements due to smaller rows, means more rows per page and thus operations perform less I/O's.
5. Logical Methods
A number of unique methods of interrogating the logical design tables are disclosed in the detailed description following.
In addition, one method resides in caching the attribute table. Thus, (with the exception of initial loading) no SQL statements are issued to the database. In the present X.500 system, conversions are performed in memory. This provides a substantial speed advantage.
Further, validation is performed in memory which avoids database roll-back. Roll-backs are time and system consuming.
Still further, for the arbitrary filter, a dynamic SQL equivalent is built. This enables arbitrary complexity in X.500 searches.
Also for search results, the present system utilises set orientation queries of SQL to avoid ‘row at a time’ processing. Thus search results may be assembled in parallel in memory.
6. Physical Design
New tables and new columns are introduced to overcome column width and key size restrictions and to achieve space optimisations.
The following text is a disclosure of embodiments of the inventions outlined:
1. Principal Design
With reference to
Throughout this and the following sections all column names and their positions in each table are arbitrary. The intent is to define what they contain and how they are used.
1.1 Extensibility
For a typical relational database application, the nature of data is well known, i.e., tables will consist of a number of columns and each column contains data relating to a particular data type (see Table 1.1a). The table is self descriptive, i.e. the relations between data items is implied by being on the same row (this is the basis of relational theory).
However, the above approach is not extensible because the number of different data types is limited to the number of columns of the table. If a new data type needs to be added (e.g. mobile phone number) then a new column will have to be added to the table (see Table 1.1b). Any application accessing this table will need to be updated to explicitly query it.
Other problems also exist in practice. If the new data type is not well used (e.g. less than 1% of the organisation has a mobile phone) then the table will be sparse (e.g. if a given person does not have a mobile then that row/column entry will be NULL). Also, the data types are limited to the types supported by the database (e.g. string, numeric, money, date, etc.).
The solution is to treat the data types as generic. The present invention adopts the method of representing arbitrary attributes (e.g. XOM [X/OPEN Object Management]. API [Application Programming Interface]) as a type, syntax, value combination (see Table 1.1c)
1.2 Object Oriented
X.500 defines objects (e.g. people, organisations, etc.) which may contain an arbitrary number of “attributes”. Since many objects must appear in the table a mechanism is required to distinguish each object. An “object name” column is added to the table for this purpose (see Table 1.2a).
The above method allows any number of attributes to be assigned (related) to an entry. These attributes could be of arbitrary complexity (e.g. a multi-line postal address could be handled). As the number of columns is fixed new attributes can be added to any object without having to redefine the application. If a new attribute is added then an application that reads the entry will get back an extra row.
1.3 Hierarchical
A method of representing hierarchical systems (e.g. parts explosion) is to use a parent/child combination (see Table 1.3a)
X.500 defines its objects to be hierarchical. The relationships between objects follow a tree structure where each object has a parent object and each parent can have zero or more children. This relationship can be represented in a general PROPERTY table by the addition of a “parent name” column, which is used to store the name of the parent object (see Table 1.3b).
Note that the root of the tree has no parent. Thus, if both Chris and Alana work for Datacraft and Datacraft is a child of the root then we can say that Chris and Alana are children of Datacraft and that Datacraft is the parent of Chris and Alana.
2. Conceptual Design
In Section 1 it was shown that a single Property Table could represent the extensible, object oriented and hierarchical nature of X.500 (see Table 2a).
With reference to
The conceptual design addresses major problems with implementing full X.500 functionality in relational tables. As each major design issue is presented, examples are provided to illustrate the solution.
2.1 Functional Decomposition
The Property Table (
These tables result from a process called functional decomposition.
To address the problem of correlating the relationships between tables, arbitrary numeric identifiers are introduced. The EID or “entry identifier” correlates each object with its hierarchy information. The AID or “attribute identifier” correlates each value in the object table with its attribute information.
The design is considered very efficient because the repeating groups in the Property table (type-syntax and object name-parent name) have been removed. Also, for SQL, the joining columns are simple integers.
2.2 X.500 Attributes
X.500 attributes have a protocol identifier which is transferred when any data is communicated between end systems. These identifiers are internationally defined and are called OBJECT IDENTIFIERS (e.g. 2.5.4.4 means a surname string). Thus an “ObjectId” column can be added to the Attribute table so that conversions between X.500 object identifiers and the internal attribute identifiers can be performed.
In addition, X.500 allows an attribute to have an arbitrary number of values (e.g. the mobile phone could be treated just as a second telephone number). Thus a “value identifier” or VID is introduced to identify values within an attribute in the Object Table.
2.3 X.500 Names
In X.500, each entry uses one or more of its attribute values (Distinguished Values) for naming the entry. A “Disting” column is added to the Object Table to flag the distinguished values.
The Distinguished Values combine to form a Relative Distinguished Name (RDN) which names the entry. The “Name” column in the Hierarchy table stores the RDN. This is an optimisation that negates the need for the RDN to be constructed from the distinguished values in the Object table.
An entry is uniquely named by a Distinguished Name (DN) which consists of all the RDN's of the of its ancestors down from the root and the RDN of the object itself. An innovation is to add a “path” column to the Hierarchy table which defines the absolute position of the entry in the tree as a list of EID's. The path has three important properties;
3) it is independent of its DN (any of the RDN's in the DN can be renamed without affecting the path).
2.4 X.500 Aliases
X.500 also has the concept of ‘aliases’. An alias object effectively points to another entry and thus provides an alternate name for that entry. Thus an “alias” flag is added to the Hierarchy Table. When an alias is discovered during Navigation (i.e. the supplied DN contains an alias), then the alias value must be read from the Object Table. This alias DN must be resolved to where the alias points before Navigation of the original entry can continue.
An innovation is to use an “aliased EID” column or A_EID to store “where” the alias “points to”. This removes the need to repeatedly navigate through an alias.
2.5 X.500 Data Tolerance
Every X.500 attribute has a (internationally defined) syntax. X.500 attribute syntaxes define how each attribute should be treated. In all string syntaxes (e.g. Printable, Numeric etc.) superfluous spaces should be ignored. In some syntaxes the case is not important (e.g. Case Ignore String and Case Ignore List) and so the names “Chris Masters”, “Chris MASTERS” and “ChRis MaSTeRS” are considered identical.
In order to do comparisons (e.g. search for a particular value), the syntax rules can be applied to create a normalised form (e.g. “CHRIS MASTERS”). If this normalised form is stored in the database, then any variations in input form are effectively removed, and exact matching can be used (which is necessary when using SQL).
Both the normalised data and “raw” data are stored in the database. The “raw” data is necessary so that users can retrieve the data in exactly the same format as it was originally input. As per the X.500 and LDAP standard, data received from a user, raw data, accords with ASN.1 (Abstract Syntax Notation No.1). Thus the “Name” column in the Hierarchy Table becomes the “NameRaw” and a “NameNorm” column is added. Similarly, the “Value” column in the Object Table becomes the “ValueRaw” and a “ValueNorm” column is added.
3. Conceptual Methods
This section introduces the basic X.500 services and shows how the conceptual table design, shown in Table 3a or
The example hierarchy shown in Table 3b, as seen in
In the example, entry “1” has an RDN with a value of “Datacraft”, entry “11” has an RDN with a value of “Sales”, entry “20” has an RDN with a value of “Network Products” and entry “31” has an RDN with a value of “Alana Morgan”. The DN of entry “31” is made up of a sequence of RDN's, namely, “Datacraft”, “Sales”, “Network Products”, “Alana Morgan”.
The alias entry “Datacraft/Networks” points to the entry “Datacraft”, “Sales”, “Network Products”. When navigating to this entry the navigate process would find the alias entry, then find the DN of the object pointed to by the alias and then navigate from the root to the object entry returning an EID of “20” and a path of “1.11.20.”.
Listed below are sample tables which show how data is stored. The Hierarchy table (Table 3c) shows how the entries for the example hierarchy are stored. The Attribute table (Table 3e) shows attributes which are contained in the entry “Datacraft/Sales/Network Products/Chris Masters”. The Object table (Table 3d) shows how the values of these attributes are stored.
Distinguished Names
For the entry shown in the sample Object Table (Table 3d) two of the attributes, commonName and surname, are distinguished values (or naming values) which combine to form the RDN for the entry. This RDN is stored in the Hierarchy Table.
Multi-Valued Attributes
In X.500, it is permissible for an attribute to be multi-valued. The VID column is used to distinguish between values for an attribute. In the sample Object Table, the telephoneNumber attribute is multi-valued.
3.1 Mapping Services to SQL
3.1.1 Attribute Types and Values
Any data supplied by an X.500 service is supplied as a list of ObjectId's and their associated values. These must be converted into AID's (using the Attribute table) and normalised values (using the Object table) for use by the X.500 application. The database returns data as AID's and Raw Values, which must then be converted into ObjectId's and their associated values in the X.500 result.
3.1.2 Navigation
Each X.500 service supplies a Distinguished Name which is converted into an EID for use by the X.500 application. When the application processes a service it returns one or more EID's. These EID's can then be translated back into Distinguished Names in the X.500 result.
All X.500 services rely on navigating the directory tree. To navigate to a particular entry, the following procedure is performed:
Example
Navigate to the entry “Datacraft/Sales/Network Products/Peter Evans”. This will result in a number of select statements, with each returned EID being used as the value of the PARENT in the next statement.
3.1.3 Read
Selected attributes to be read can be supplied. Only the values of these attributes (if they are present in the entry) will be returned.
‘Types only’ can be selected as a read option, in which case no values will be returned. All types present in the entry, or those selected, will be returned.
Navigate to the entry to be read. Store the EID. In the Object Table, read the values of all rows which match the stored EID.
Example
Navigate to the entry (as in 3.1.2) and then;
3.1.4 Compare
Compare returns a ‘matched’ or ‘not matched’ result. A raw value is input but the compare is performed using the normalised value.
Navigate to the required entry. Store the EID. In the Object Table, test for a matching value in all rows which match the stored EID and the specified AID.
Example
Navigate to the entry and then;
If a value is selected then return “matched” else return “not matched”.
3.1.5 List
Navigate to the required entry. Store the EID. In the Hierarchy Table, return the RDN's for all rows with a parent matching the stored EID.
Example
Navigate to the entry and then;
3.1.6 Add Entry
Navigate to the required parent entry. Store the EID of the parent. Add a new EID to the Hierarchy table and add rows to the Object table for each value in the new entry.
Example
(EID, PARENT, PATH, ALIAS, A-EID, NAMENORM, NAMERAW)
values (33, 20, 1.11.20.33, 0, 0, EDWIN MAHER, Edwin Maher)
3.1.7 Remove Entry
Navigate to the required entry. Check that the entry is a leaf on the tree, (i.e. check that it has no subordinate entries on the tree). Store the EID. Remove the entry from the Hierarchy table. In the Object Table, remove all rows which match the stored EID.
Example
Navigate to the entry and then;
3.1.8 Modify Entry
Navigate to the required entry. Store the EID. In the Object Table, Add, Remove or Modify rows matching the stored EID.
Example
Add value—title=“Branch Manager”.
Navigate to the entry and then;
Test the returned rows for an attribute of title. If none exist, the attribute can be added, otherwise the attribute must be checked to see if it can be multi-valued and whether it already exists.
Insert into OBJECT
(EID, AID, VID, DISTING, VALUENORM, VALUERAW)
values (31,12,1,0, BRANCH MANAGER, Branch Manager).
3.1.9 Modify RDN
Navigate to the required entry. Check that the new name (RDN) does not exist in the current level of the subtree (i.e. that the new DN is distinct). Store the EID. Modify the entry in the Hierarchy and Object tables.
Example
Navigate to the entry and then;
If no entries are returned then the new RDN may be inserted. First set the old RDN to be a non-distinguished value.
3.2 Search Strategy
The most powerful and useful X.500 service is the search service. The search service allows an arbitrary complex filter to be applied over a portion of the Directory Information Tree (the search area).
One technique for resolving searches is to apply the filter and then to see if any matching entries are in the search area. In this case a filter is applied to the entire tree and EID's for all rows matching the filter are returned. Then, for each EID found, step search up through the hierarchy to see if the entry is a subordinate of the base object (i.e. the entry has a parent/grandparent/ . . . that is the base object). If the number of matches is large and the subtree small this is very inefficient. This technique doesn't cope with aliases as an alias is not a parent of the object that it points to and many aliases may point to a single object.
A second strategy is to obtain a list of all EID's in the search area and then apply the filter to these EID's. If an alias is resolved that points outside of the original search area then the subtree pointed to by the alias is expanded and the EID's in that subtree are added to the list. The filter is then applied to the set of expanded EID's. This is very poor if the search area is large.
An innovation is to simultaneously apply the filter over the search area (instead of sequentially as in the two methods described above). This is called single pass resolution. This method is considered to provide considerable performance improvement over the above methods because the rows that are retrieved are those that satisfy both the filter and scope requirements of the search.
When performing a one level search the filter is applied to all entries that have a parent equal to the EID of the base object (for example; search where parent=20 will apply the filter to entries 30, 31 and 32).
When performing a subtree search the path is used to expand the search area. The “path” of each entry is a string of numbers (e.g. “1.10.50.222.” which indicates that entry 222 has a parent of 50, a grandparent of 10 and a great grandparent of 1). The path has the unique property that the path of an entry is a prefix of the path of all entries that are subordinate to the entry. That is the path of an entry forms the prefix of the paths of all entries in the subtree below the entry. Therefore when performing a subtree search we obtain the base object of the subtree and then apply the filter to all entries that have a path which is prefixed by the path of the base object (for example; to search for all entries under “Sales” we perform a search where PATH LIKE 1.11.%).
Base Object Search:
Navigate to the base object. Store the EID. In the Object Table, read nominated values from rows which match the stored EID where a filter criteria is satisfied, eg, telephone prefix=“727”.
Example
Search from the base object “Datacraft/Sales/Network Products” for an entry with surname=“MORGAN”, using a “base-object-only” search. Navigate to the base object and then;
One Level Search:
Navigate to the base object. Store the EID. Return the list of EID's which have a parent EID matching the stored EID (in Hierarchy table) and have values which satisfy the filter criteria (OBJECT table). In the Object Table, read nominated values for the returned EID's.
Example
Subtree Search:
Navigate to the base object. Store the EID. Return the list of all EID's with a path like that of the base object (Hierarchy table) and have values which satisfy the filter criteria (OBJECT table). In the Object Table, read nominated values for the returned EID's.
Example
3.3 Aliases and Navigate
Aliases are resolved during navigation if the “don't-dereference-alias” flag is not set and the service is not an update service (add, delete, modify, modifyRDN).
When an alias is discovered during navigation the alias must be resolved. That is, the object that the alias points to must be obtained. First we check the A_EID column of the Hierarchy table. If the A_EID is 0 then the object that the alias points to must be obtained from the Object table and this object must then be navigated to and the resultant EID stored in the A_EID column. If this is done successfully then the remainder of the path can be navigated. By storing the EID of the aliased object in the A_EID column of the Hierarchy table it is possible to avoid navigating to aliased objects. This can save time, especially if the aliased object is at a low level of the hierarchy.
3.4 Aliases and Search
Aliases are dereferenced during a search if the “search-aliases” flag in the search argument is set. The performance of the search service while dereferencing aliases becomes a two step process. Firstly, define the search area and then apply the filter to the entries within the search area. Aliases dereferenced as part of the search service can expand the search area to which the filter is applied. They also restrict the search area in that any dereferenced aliases are excluded from the search area.
Aliases and OneLevel Search
If aliases are being dereferenced as part of a one level search and an alias entry is found then the alias must be resolved (using the Object table or the A_EID). The aliased object is then added to the search area to which the filter is applied. In a oneLevel search where aliases are found the search area will consist of non-alias entries directly subordinate to the base object and all dereferenced aliases.
Aliases and Subtree Search
If aliases are being dereferenced as part of a whole subtree search and an alias entry is found then the alias must be resolved (using the Object table or the A_EID) and this EID must then be treated as another base object, unless it is part of an already processed sub tree.
When dereferencing aliases during a search the “Path” column can be used to find alias entries within a subtree join. If an alias entry is found that points outside of the current subtree then the subtree pointed to by the alias can also be searched for aliases. One property of the hierarchical tree structure is that each subtree is uniquely represented by a unique base object (i.e. subtrees do not overlap). When performing a subtree search we build up a list of base objects which define unique subtrees. If no aliases are found then the list will contain only one base object. If an alias is found that points outside of the subtree being processed then we add the aliased object to the list of base objects (unless one or more of the base objects are subordinate to the aliased object in which case the subordinate base object(s) are replaced by the aliased object). The search area will therefore consist of non-alias entries that have a path prefixed by the path of one of the base objects.
4. Logical Design
Whilst the Conceptual Design (see Table 4a) is sufficient to implement the X.500 functionality, further performance improvements can be made.
Performance improvements in conventional relational design can be achieved because assumptions can be made about the data—the data is essentially fixed at the time an application is designed. In X.500, none of the data types are known. However performance improvements can still be made because assumptions can be made about the services—these are known at the time the X.500 application is designed.
With reference to
4.1 Service Decomposition
The practical reality for most RDBMS's is that big tables with many columns do not perform as well as smaller tables with fewer columns. The major reasons are to do with indexing options, I/O performance and table management (see Sections 4.5 and 4.6). This is why prior art relational design techniques aim to focus primary information into separate tables and derive secondary information via table joins (i.e. normalisation and fragmentation techniques).
One innovation in achieving X.500 performance is to decompose the tables around primary service relationships and derive secondary services via joins. This process is called service decomposition. The following considerations are made:
A first level analysis of column usage is shown in Table 4.1.
Key to symbols in the above table:
H—Hierarchy table
O—Object table
S—Supplied value (used in the SQL for Searching the table)
R—Returned value (value retrieved from the tables)
( )—item may or may not be present depending on the options of the service.
From the above information and further analysis, the Conceptual Design tables can be decomposed into a number of smaller tables as described in the following sections.
4.2 Hierarchy Table Decomposition
The Hierarchy table contains the following columns:
The Hierarchy Table contains information about objects and their parents, their names, their absolute positions in the hierarchy and if they are aliases. This table can therefore be split into four tables: DIT, NAME, TREE and ALIAS.
The parent information is used for finding a given child or acting on entries that have a given parent. Finding a given child (e.g. Parent=0, NameNorm=“DATACRAFT”) is the basis for Navigation and update checking (checking for the existence of an object before an Add or ModifyRdn). Acting on entries that have a given parent is used during List or OneLevel Search. Thus the DIT (Directory Information Tree) table has information required for Navigation, but allows its PARENT column to be used by other services.
An object is differentiated from its siblings via its Relative Distinguished Name (RDN). RDNs are returned for a List (in conjunction with a given Parent) or as part of a full Distinguished Name (Read, Search). Thus the NAME table has information required for returning names (the raw RDN).
An object's absolute position in the hierarchy is necessary for building DN's (from which the raw RDN's are retrieved) and for expanding subtrees during Search. Thus the TREE table has information about an entry's Path (the sequence of EID's down from the root).
Alias information is cached so that every time an alias is encountered during Navigate it does not have to be repeatedly resolved. Thus the ALIAS table only contains entries that are aliases. It is also used during OneLevel Search (in conjunction with the DIT Parent column) and Subtree Search (in conjunction with the Path column) to determine if there are any aliases in the search area.
4.3 Object Table Decomposition
The Object table contains the following columns:
The Object Table essentially contains information for finding a particular value (e.g. AID=surname, ValueNorm=“HARVEY”) and for retrieving values (e.g. AID=surname, ValueRaw=“Harvey”). This table can therefore be split into two tables: SEARCH and ENTRY.
The Search Table is used to resolve filters in the Search service. It is also used to find values during Compare, Modify and ModifyRDN. The Search table contains one row for each attribute value of each entry. Only the normalised values are stored in this table.
The Entry table is used to return values in Reads and Searches. The Entry table contains one row for each attribute value for each entry. The RAW value is the value exactly as initially supplied when the entry was added or modified.
4.4 Attribute Table
The Attribute table is essentially the same as the Conceptual Design. In practice the “type” field is only descriptive, since any incoming/outgoing X.500 Object Identifier gets converted to/from the internal attribute identifier, AID. Thus this column has been renamed DESC to signify that it is a description field.
4.5 Index Selection
Performance when using SQL is achieved because the RDBMS is able to satisfy the query using a relevant index. This means that every query that has a condition (the “where” clause in SQL) is preferred to have an associated index (otherwise the RDBMS has to resort to a table level scan). However in practical RDMS's:
One innovation of the table decomposition in the previous sections is to maximise the use of primary indexes across tables. This reduces the number of secondary indexes (i.e. they become primary indexes on their own table). Following is a list of the indexes for each of the six tables used in the logical design.
The table design means that many queries can be handled without joins, giving substantial performance improvement.
The joins that are considered necessary are listed below:
Note that the above joins are first level joins (i.e. between only two tables). It is preferable not to use higher order joins.
4.6 Input/Output Performance
An innovation of decomposing tables around services, which increases the number of tables, is that the new tables are much smaller than the unfragmented tables. This can significantly reduce the amount of I/O for the following reasons:
Row Size
By reducing the number of columns in any row, the row width will be shortened. This means that more rows will fit onto a page (where it is assumed that one disk I/O returns one “page” of information). In combination with clustering below, whenever a set of rows need to be retrieved, only one (or a few) page(s) may actually have to be read off the disk (e.g. when reading the attributes of an object, if the ENTRY table is keyed on EID, AID, VID then all the rows relating to that object will be together and will probably be on the same page).
Clustering
Each of the fragmented tables is preferred to have their own (independent) primary key which enables them to cluster data according to how it is used. The primary key may dictate the “storage structure”. Thus in the SEARCH table, if the primary key is on AID, NORM (i.e. type, value) then all the data of the same type (e.g. surname) and similar values (e.g. Harvey, Harrison) will be clustered in the same area of the disk. This means that during a Search (e.g. surnames beginning with “HAR”) similar data will collected together on the one (or just a few) disk page(s). If the rows are small then the number of disk pages that have to be accessed is significantly reduced.
Caching
Most commercial RDBMS's have the ability to cache pages frequently accessed. Since tables are effectively input (e.g. Navigating using the DIT table), or output (e.g. retrieving information from the ENTRY table) then similar requests (e.g. Searches over the same portion of the Tree) will tend to result in frequently used pages being cached, meaning frequently invoked queries will gain significant benefits. Also the caching is more efficient since pages are “information intensive” as a result of small row size and clustering.
Management
Smaller tables are generally easier to manage: e.g. viewing, creating indexes, collecting statistics, auditing, backups, etc.
5. Logical Methods
This section describes methods of interrogating the Logical Design tables, with reference to
Throughout this section, each X.500 method is defined and illustrated with an example. Referring again to
NOTE:
[ . . . ] indicates a binary encoding of the exact data entry value.
NOTE:
[ . . . ] indicates a binary encoding of the exact data entry value.
5.1 Common Services
Tree Navigation
All X.500 services rely on navigating the directory tree, illustrated in
The DIT Table is the primary table used for tree navigation. Referring to the example hierarchy tree, illustrated as table 5a in
The DN has now been resolved and any values relating to the object can be obtained from the Entry Table using the key EID=32.
Aliases
Sometimes a DN can contain an alias, which is effectively another DN. Aliases complicate the tree walk process because the tree walk cannot continue until the alias is resolved. This requires a separate tree walk for the alias.
As an example, consider the DN “Datacraft/Networks/Peter Evans”. The first two steps in resolving this DN would be:
At this stage we discover that this entry is an alias. The Alias Table is checked to see if the EID of the alias has been cached. If this is the first time an attempt has been made to resolve this alias then the A_EID column in the Alias Table will be zero. For the purpose of discussion it will be assumed that this is the first time.
To resolve the alias, the DN of the aliased object must be determined. This is stored in the “aliasedObjectName” attribute of the alias entry. The aliasedObjectName has an AID=1 (from the ATTR table) and so the DN is obtained from the Entry Table (RAW value) where EID=10 and AID=1.
In this example, the DN of the alias is “Datacraft/Sales/Network Products”. This DN is resolved completely using the normal tree walking technique. The value of EID is 20.
At this stage, navigation continues for the unresolved RDN's in the original DN, namely “PETER EVANS”. The last step required is then:
Once an alias has been resolved it can be added (cached) in the Alias Table. This table contains a reference, A_EID, to the aliased object. In the above example, an entry in the Alias Table with an EID of 10 would have an A_EID of 20. Once an alias has been cached a tree walk is no longer necessary to resolve the alias.
Directory Paths
When objects are added to the DIT table, a corresponding row is added to another table called the Tree Table. This table stores the list of the EID's which identify a “Path” to the object.
Distinguished Names
Most services require the distinguished name to be returned in the Service Result. Using the directory path from the Tree Table, a DN can be constructed from the RAW RDN values stored in the Name Table.
Entry Information Selection
Many of the X.500 Services are requested with an argument called “EntryInformationSelection” or EIS. The EIS argument is used to indicate what information in the Entry should be returned. Basically, EIS can be optionally;
Entry Information
Entry Information is a return parameter for Read and Search. It always contains the Distinguished Names of selected entries and, optionally, attributes and/or values as specified in the EIS argument of the request.
Common Arguments
All of the X.500 Services pass a set of common arguments in the Service Request. Common Arguments contain information such as service controls (time limit and size limit), the DN of the requestor of the service and security information.
Common Results
Some X.500 Services pass a set of common results in the Service Response. Common Results contain information such as security parameters, the DN of the performer of the service and an alias dereferenced flag.
5.2 Read Service
A Read operation is used to extract information from an explicitly identified entry.
Method
Example
Read the entry “Datacraft/Sales/Network Products/Peter Evans”.
EIS is set to: attribute Types=allAttributes, InfoTypes=attributeTypesAndValues.
Using the DIT table perform a Tree Walk traversing EID's 1, 11, 20 and 32 for the normalised RDN's DATACRAFT, SALES, NETWORK PRODUCTS, PETER EVANS. The EID of the selected object is 32.
Extract the PATH from the Tree Table for EID=32. The PATH is 1.11.20.32.
Build aDN from the RAW values in the Name Table for EID's 1, 11, 20, 32.
Using the Entry Table and the Attribute Table, for each matching EID;
return the OBJECTID's from the Attribute Table and the ASN.1 encoded RAW values from the Entry Table
5.3 Compare Service
A Compare operation is used to compare a value (which is supplied as an argument of the request) with the value(s) of/particular attribute type in a particular object entry.
Method
Example
Compare the DN “Datacraft/Sales/Network Products/Peter Evans” with a purported AttributeValueAssertion of “title=[Salesperson]”.
Obtain the EID for the given DN using a TreeWalk. The EID of the selected object is 32.
Using the Attribute table, obtain the AID for “title”, ie AID=12.
Using the Search Table locate rows with EID=32 and AID=12 and test for “NORM=SALESPERSON”.
Return TRUE or FALSE depending on the outcome of this test. In this instance the result would be TRUE.
Since no aliases were dereferenced, the DN of the entry is not returned.
5.4 List Service
A list operation is used to obtain a list of immediate subordinates of an explicitly identified entry.
Method
Example
Perform a list for the DN “Datacraft”.
Obtain the EID for the DN using a TreeWalk. The EID of the selected object is “1”.
For each EID with a PARENT=1
As no alias was dereferenced in the tree walk, the DN of the selected object is not returned. Note also that the alias entry [Networks] is not dereferenced.
5.5 Search Service
The Search Service is the most complex of all X.500 services. Search arguments indicate where to start the search (baseObject), the scope of the search (subset), the conditions to apply (filter) and what information should be returned (selection). In addition, a flag is passed to indicate whether aliases should be dereferenced (searchAliases).
The possible values for subset are baseObject, oneLevel and wholeSubtree. Base object indicates that the search filter will only be applied to attributes and values within the base object. OneLevel indicates the Search filter will be applied to the immediate subordinates of the base object. Whole subtree indicates the Search filter will be applied to the base object and all of its subordinates.
A simple example of a filter condition would be: surname=“EVANS” or telephoneNumber PRESENT.
The search procedures for each search scope are outlined as follows:
Base Object
One Level
For each matching EID:
Whole Subtree
For each matching EID:
Example
Perform a search on the baseObject “Datacraft/Sales” with:
Method
Obtain the EID for the base object DN using a TreeWalk. The EID of the base object is “11”.
From the Tree Table, obtain the PATH for EID=11, ie, “1.11”.
Check for any aliases among entries that have a path beginning with “1.11.”. There are no aliases in this case.
Obtain the AID for the attribute “surname” in the Attribute Table, ie, 4.
Apply the filter and scope simultaneously. i.e. Using the Search Table, obtain a list of EID's from the target list where AID=4 and the value begins with “M” joined with the Tree Table who's PATH is LIKE ‘1.11. %’. The matching EID's are 30 and 31.
Using the Entry Table and the Attribute Table, for each matching EID:
i.e.,
5.6 Add Entry Service
An AddEntry operation is used to add a leaf entry either an object entry or an alias entry) to the Directory Information Tree.
Method
Example
Under the object with a DN of “Datacraft/Marketing” add an object with the following attributes and values.
Obtain the EID for the base object DN using a TreeWalk. The EID of the base object is “12”.
Using the DIT Table, look for a duplicate entry, ie, PARENT=12 and RDN=“MARY DELAHUNTY”. No duplicates exist.
Add the following rows to the Tables shown.
5.7 Remove Entry Service
A RemoveEntry operation is used to remove a leaf entry (either an object entry or an alias entry) from the Directory Information Tree.
Method
Perform a tree walk using the DIT table. Obtain the EID of the base object.
If the entry exists, and it is a leaf entry, then for the condition EID=EID of the selected object, delete from the DIT Table, the Name Table, the Tree Table, the Search Table, the Entry Table and, if it is an alias entry, the Alias Table.
Example
Delete the object with a DN of “Datacraft/Marketing/Mary Delahunty”
Method
Obtain the EID for the base object DN using a TreeWalk. The EID of the base object is “21”. Check that no entries have PARENT=21.
Delete all rows added to the DIT Table, the Name Table, the Tree Table, the Search Table and the Entry Table (refer to Add Entry example) where EID=21.
5.8 Modify Entry Service
The ModifyEntry operation is used to perform a series of one or more of the following modifications to a single entry:
modify an alias
Method
For the selected object, perform one or more of the following actions: Add Value, Delete Value, Add Attribute, Delete Attribute
The operations required for each action are as follows:
Add Value
Checks are: If the attribute is single valued test for an existing value; if the attribute is multi-valued check for a duplicate value.
Delete Value
Add Attribute
Delete Attribute
Example
Modify the Entry “Datacraft/Sales/Network Products/Chris Masters” with the following changes:
The Search and Entry Tables reflect the changes.
5.9 Modify RDN Service
The ModifyRDN operation is used to change the Relative Distinguished Name of a leaf entry (either an object entry or an alias entry) from the Directory Information Tree.
Method
If deleteOldRDN is set to TRUE the procedures following the Tree Walk are as follows:
Example
Modify the RDN of “Datacraft/Sales/Network Products/Chris Masters”. The N is “Christine Masters”.
deleteOldRDN is set to FALSE.
The changes to the Tables will be as follows:
5.10 Complications
If error, limit or abandon occurs during processing of any of the services, then the processing is discontinued and an appropriate error message returned.
Errors
Each X.500 service consists of 3 parts; ARGUMENT, RESULT and ERRORS. In the above descriptions of the services, ARGUMENT and RESULT have been included in the X.500 definitions. Error conditions, however, are many and varied and no attempt is made to describe them in this document. The National Institute of Standards and Technology (NIST) document “Stable Implementation Agreements for Open Systems Interconnection Protocols: Version 3” provides a full coverage of errors for the X.500 standard.
Time Limit & Size Limit
Time Limit and Size Limit form part of Service Controls. They can be optionally set to some finite limit and included in the Common Arguments.
Time Limit indicates the maximum elapsed time, in seconds, within which the service shall be provided. Size Limit (only applicable to List and Search) indicates the maximum number of objects to be returned. If either limit is reached an error is reported. For a limit reached on a List or a Search, the result is an arbitrary selection of the accumulated results.
Abandon
Operations that interrogate the Directory, ie Read, Compare, List and Search, may be abandoned using the Abandon operation if the user is no longer interested in the results.
Aliases & Search
If an alias is encountered in a search and that alias points to a separate branch of the directory tree, then dereferencing of the alias requires:
In the example shown in
5.11 Implementation Optimisations
The Logical methods include a number of optimisations that enhance performance. These methods are outlined below.
Caching
The Attribute table can be cached. This means that (apart from initial loading of the attributes) no SQL statements need to be issued to the database when decoding or encoding the attributes. In the present X.500 system attribute conversions are performed in memory. This provides a substantial speed advantage.
Validation
Query validation is performed in memory where possible. This avoids database rollbacks which are time and system consuming. For example when adding an entry each attribute is validated before any attempt is made to add the entry. If an error is found then no SQL calls need to be issued.
Optimise Query Handling
As the format of most services is known, many instances of these services can be resolved using static SQL statements. More complex services, such as searches with complex filters, can be resolved using dynamic SQL. This enables arbitrarily complex searches to be performed.
Parallel Queries
Also when processing search results the present system utilises set orientation queries of SQL to avoid ‘row at a time’ processing. Thus search results may be assembled in parallel in memory.
Data Storage
The tables that store raw data store the data in ASN.1 format. This provides an efficient means of transferring data into or out of the database.
Database Techniques
Complex services can be further improved by using the query optimiser, which provides a mechanism for reducing the time spent in resolving the query. The use of a relational database also provides an efficient use of memory and enables large databases to be constructed without the need for large amounts of memory being available. Many other X.500 applications cache the entire database in memory to achieve performance. This method consumes large amounts of memory and is not scalable.
6. Physical Design
The physical design results from a process called physical transformation of the logical design. The physical design represents a preferred realisation or embodiment of the logical design.
The reasons for the above changes are described below.
6.1 Efficiency
INFO Table
This table holds the highest EID value that has been used in the database. The inclusion of the INFO table enables the next EID to be obtained without any calculation of the maximum EID being performed by the database. This provides improved efficiency in adding entries to the database. More importantly the inclusion of the INFO table removes contention problems which may occur when multiple DSA's are adding entries at the same time.
Shadow Keys
Three tables have had shadow keys added. These are:
Each of these shadow key columns is a shortened version of a larger column. They have been added to shorten the indexes on each table. This gives improved performance for any queries that use the indexes and it also improves disk space usage as small indexes take up less space than large indexes.
The shadow keys in the PATH table utilise the structured nature of the PATH. By being a composite key then exact matching can be used in the SQL instead of the “LIKE” operator.
If each of the LEV columns has their own index, then a sub-tree search needs to only use the base object. e.g. LEV2=10, since all objects under entry 10 will have LEV2=10.
SENTRY Table
Some types of attribute values do not need to be normalised e.g. integer, boolean, date. Instead of storing them twice (SEARCH.NORM and ENTRY.RAW) they can be stored just once in a hybrid table called the SENTRY table. This reduces table sizes and increases storage efficiency at the cost of having to search two tables and retrieve from two tables.
OCLASS Table
Most attributes have a wide variation in their values e.g. surnames could range from AALDERS to ZYLA with a great many different values in between. However, Object Classes (whose values are ObjectIdentifiers or OIDs) have very few values e.g. in an organisation of 10,000 people, the only object classes in the directory may be for organisation, organisationalUnit and organisationalPerson (of which many may be the latter). The OCLASS table gives a numeric descriptor to an object class called an OCID. The OCID can then be stored in the SENTRY table and a mapping done whenever an Object Class is searched or retrieved. The other LIST columns store standard object class configuration information—namely the must and may contain attributes and the inherited superclasses.
6.2 Portability
BLOB Table
This table has been included to hold “Binary Large Objects”. The maximum size of a one row entry in the ENTRY table is limited by the length of the RAW field. This means that entries must be fragmented. Fragmented entries will occupy more than one row and so a VFRAG field must be used to denote the fragment of the entry that is being stored in a particular row.
There are two options for storing very large values:
The second option has a number of advantages. Firstly, the inclusion of a BLOB table prevents the ENTRY table from becoming excessively large. Generally most entries will be less than a few hundred characters in length, so the length of the RAW field in the ENTRY table can accordingly be reduced to cater for those entries and the RAW field in the BLOB table can be increased to a value approaching the maximum record size. This will make storage more efficient, i.e. reduce the amount of unused bytes in each column of each table and reduce the number of fragments needed for each entry in the BLOB table. It also means that each value will have only one entry in the ENTRY table and that the ENTRY and SEARCH tables maintain their one-to-one correlation. Secondly the use of a BLOB table enables the application to make use of any database support for Binary Large Objects. (e.g. 64K Binary Columns).
6.3 Functional Extensibility
FLAGS Columns
FLAGS column(s) are preferred to be added. These column(s) have been added to provide extensibility to the design. Specific values can be added to the flags as new functionality is required, without changing the table structure.
Note:
The FLAGS column(s) may also provide a “summary” function for each of the tables. This means that the nature of an entry can be determined to some extent by checking the value of the FLAGS field. For example, a flag can be set, in the DIT table, when an entry is a leaf. Checking this flag is much simpler than checking for children of the entry.
The FLAGS column can also be used to store security information, whether an alias points inside its parents sub-tree, whether a value is a BLOB, etc.
7. Example Implementation
The following provides an example of system performance and capabilities. It is to be understood that the present inventions should not be limited to the following disclosure.
7.1 Overall System Benefits
The present invention is considered to provide enhanced performance over prior art implementations. Performance can be appraised in many ways, including:
The present invention is considered unique in its ability to claim performance improvement in all areas noted above.
7.2 Test Results
Performance testing of the present invention has been carried out, with the objectives of:
Test results reveal the following Table 7A
Notes:
1. All searches and reads return all info
2. All tests were performed under the following environment;
Sun SparcStation 5 with 32 Mb of memory (entry level UNIX machine)
All numbers are in units of seconds and “K” means 1,000's.
7.3 Test Conclusions
A set of directories was constructed ranging from 1K to 200K entries with varying depth and width of the hierarchy, and a corresponding test plan was produced. The tests were performed a number of times to ensure consistency.
The following conclusions can be drawn from these results;
1. The effects of navigation, in test, were negligible.
2. Reading an object via an alias, in test, showed no appreciable decrease in performance and in some cases reading an object via an alias was in fact faster than reading the object directly. This is due to the reduced navigation required when an alias points “down” to an object that is deeper in the tree structure than the alias entry.
3. Search results were “flat” over different sized subtrees in different sized directories for both exact and initial string searches.
4. Initial and exact full tree searches, in test, were slightly quicker than their respective subtree searches, even though the number of entries searched was greater. This is due to the fact that the full tree searches are able to use more efficient SQL (no table joins are required).
5. All services were, in test, performed in under one second, except for searches returning large amounts of data. However the average time of retrieval perentry drops as the number of entries retrieved increases (e.g for 10 entries retrieval time is approximately 50 milliseconds per entry, for 100 entries this drops to approximately 20 milliseconds per entry).
6. All complex searches, in test, were performed in under one second. However, there may be some obscure searches (e.g containing combinations of NOT) which may not perform as well.
Because this is a disk based system (rather than a memory based system) performance is essentially only dependent on the number of entries actually returned. It is relatively independent of the search complexity, the depth of the hierarchy, the number of attributes per entry or the types of attributes used in the query. In a “live” application of the system it may be possible to improve on the achieved test results by tuning the caching parameters, and by having a greater diversity of attributes.
| Number | Date | Country | Kind |
|---|---|---|---|
| PM 7842 | Sep 1994 | AU | national |
| PM 9586 | Nov 1994 | AU | national |
This is a divisional of U.S. Ser. No. 08/793,575, which is currently pending and which is incorporated herein by reference in its entirety.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 08793575 | May 1997 | US |
| Child | 09427267 | Oct 1999 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10308018 | Dec 2002 | US |
| Child | 11234928 | Sep 2005 | US |
| Parent | 09427267 | Oct 1999 | US |
| Child | 10308018 | Dec 2002 | US |
