User categorization system and method

[0001] None.

[0002] Resource allocation is a key aspect of efficient management of highly used resources such as server-based software packages. As user demand increases, server resources become scarce and may not be adequate to meet user demand. If server resources are inadequate to meet user demand, a business or organization may experience financial or other losses as users are unable to complete important tasks or activities.

[0003] When server resources are insufficient to meet the demands of all users, they are typically allocated according to established priorities. Establishing priorities, however, can be difficult for a business or organization. While any business or organization would like to meet the needs of all users, such a goal is impractical or even impossible. At best, the business or organization is likely to be able to meet all the needs of only a small subset of the user population. A more likely scenario is that every user's needs can be met over time but each user may not have the unrestricted access to the server resources that he or she would desire at any time.

[0004] To best meet the of the server user population, it is important for a business or organization to understand how users use the resources. Unfortunately, many businesses and organizations simply make assumptions about users and their use of the resources. Such assumptions, however, are often partially or even entirely false. Any solution that is developed, therefore, relies on the false assumptions. As a result, they are often simplistic and unsatisfactory.

[0005] Rather than rely on assumptions, a better approach is to collect usage data. One way to learn how resources are used is to survey users regarding their use. Such an approach however, requires not only appropriately drafted questions but also complete and accurate answers to the questions from the users. Users may try to answer the questions the completely and accurately but may over- or under-estimate their usage. User perceptions may be as inaccurate as assumptions.

[0006] Even if user-reported usage data may be collected, it may be difficult for the business or organization to understand the data and more importantly, how to allocate resources based on the data. Certain groups of users may have similar needs that can be satisfied with an appropriate resource allocation strategy but it may be difficult for the business or organization to discern the groups from the data. Even if the groups can be discerned, the business or organization must decide how to allocate resources to the groups. In an attempt to satisfy at least a portion of users in each group, businesses and organizations typically consider the needs of each group without consideration of the businesses' or organizations' needs. There is a need for a user categorization system and method that can facilitate the identification of user groups based on actual server usage data.

[0007] The present disclosure is directed to a user categorization system and method. The user categorization system and method relies on actual server data to discover which users are consuming the finite resources of the server and in what proportions. Users are categorized according to their consumption of resources. The user categorization system and method further ascribes a relative business value to each user group to facilitate the allocation of resources among groups. In an example embodiment, users of the server resources use the SAS programming language and the server resources execute SAS applications that support the SAS programming language.

[0008] The user categorization system and method connects an executed computer program to a business-defined classification of applicability to purpose. The system and method employs a double level of abstraction to link specific programming styles, first to a general solution case (“design pattern”), and then to link the general solution idiomatically to the business case. Users are clustered into categories, which result from the application of rules or functions that answer “how, what and how many” particular design patterns (or measure attributes) were used by the user.

[0010] In an example embodiment, the system and method connects an executed computer program to a business-defined classification of applicability to purpose. The program employs a double level of abstraction. Programming styles are first linked to a general solution case (“design pattern”), and then the general solution is linked idiomatically to the business case. Users are clustered into categories based on the application of rules or functions that determine “how, what and how many” particular design patterns (or measure attributes) were used by the user.

[0011] The mechanism used for the solution is a system using a taxonomy of programming techniques, called “design patterns,” the resource use of each executed computer program component associated to it as “measures,” and defining an idiomatic “cross reference” between the design patterns and figurative business meaning used to solve business management defined problems. In the definition of the taxonomies, the emphasis is on the business's perspective of problems and solutions.

[0012] First, a data cube is created to define a high dimensional array of values using attributes of the target system. This delineates the universe in which the users operate. Attributes are collected that could be used to describe the user's programming behavior. In an example embodiment, these attributes fall into three broad categories: program language writing style; computer resource consumption; and business organizational classification (dept and manager, etc). Enumerated lists are created for each attribute. These lists define the “vocabulary” and cardinality of the attributes. Some of the attributes (notably, resource consumption) remain to be used as “measures,” but are also grouped into lists of “dimensions” by way of clusters or bands of values.

[0013] The specific categories into which the attributes are organized are malleable and may be defined by the business need. Sometimes categories are mutually exclusive, but it is not required. The categorization technique is grounded in demographic clustering analysis. A preference assignment hierarchy for attributes found in multiple categories may be established based on business value preference or by ordering the measures.

[0014] An aggregation of all the attributes and measures by each of the attribute categories is performed. This step provides the ability to define additional measures as a ratio of the whole. This is in addition to the absolute sums used as measure values. The attributes are then matched together such that it is possible to identify (for example) one user, one program block, and the resources associated to that specific interaction, also known as a “tuple.” Higher order groups of collections are then also possible.

[0015] The first key component is to identify design patterns in the program blocks. Design patterns are an abstraction to the general case of a computer program or part of a computer program that solves a specific commonly occurring problem or key computation function. These patterns are later used as a proxy to associate the program block to a business use and the business value.

[0016] The description and definition of design patterns is flexible as long as the abstraction by which the higher concepts are derived from in the usage and classification of literal concepts of the program continue to hold true. A pattern applies to a particular situation, and it is through this applicability that the pattern is used idiomatically to represent a crucial key to the business process, especially when a small but representative sample is recognized as the clue to understanding a larger whole. The range of situations in which a pattern can be used is called its context. A multitude of patterns may be identified that apply to the same context.

[0017] The second key component is in defining, creating, and using an idiomatic cross reference between the design patterns and figurative business meaning. Once the idiomatic definition of a user's programming is added as an attribute to the data cube, conventional data mining techniques may be employed to extract previously unknown “interesting patterns” such as groupings of data records (cluster analysis), discovery of unusual records (anomaly detection) and “interesting relations” or dependencies (association rule mining). In this way, it is possible to extract information from the data and transform it into an understandable structure for further use managing the server environment. The following illustrates a technique that may be used. The elements are by no means required or representative of an exhaustive list but they are illustrative.

[0018] The data sets use conventional database table design, organized into rows and columns. The columns are also conventionally defined and are classified as keys, dimensions, or measures. (Kimball & Ross—The Data Warehouse Toolkit, 2nd Ed [Wiley 2002])

[0019] Multiple intermediate data sets are combined to yield three primary data sets. A time match algorithm matches elements attributes and sums the measures for the time domain. For instance, a SQL step may run for three hours, and the Computer Resource table is measured in five minute increments.

[0020] In this case, each of the measures is summed to the timeframe of the SQL step to correctly attribute the measures to the step. It also works in reverse where multiple program steps occur in the time increment on the Computer Resource table. The algorithm method is not important nor is the specific time increment amount. The goal is an accurate assignment of measures to be attributed to any given program step.

[0021]

TABLE 1
Programming Language Attributes Dataset
Dataset
Program Language Attributes
	Key	UserID
	Key	ProcessID
	Key	Date timestamp
	Dimension	Program Block ID
	Dimension	Program Data Input ID
	Dimension	Program Data Output ID
	Dimension	Program Step Summary Code
	Dimension	Program Step Options
	Measure	Counts

[0022]

TABLE 2
Computer Resource Consumption Dataset
Dataset
Computer Resource Consumption
	Key	UserID
	Key	ProcessID
	Key	Date timestamp
	Dimension	Program Block ID
	Dimension	Program Data Input ID
	Dimension	Program Data Output ID
	Measure	CPU Amount Used
	Measure	I/O Amount Used
	Measure	Disk Storage Used

[0023]

TABLE 3
Business Organization Classification Dataset
Dataset
Business Organization Classification
	Key	UserID
	Key	ProcessID
	Key	Date timestamp
	Dimension	Department ID
	Dimension	Supervisor ID
	Dimension	Program Data Output ID
	Dimension	Program Step Summary Code
	Dimension	Program Step Options
	Dimension	User Profile Category

[0024] A design pattern represents a summary of what a program block accomplishes. It may be defined loosely or in great detail. It does not need to conform to any standard outside of what the programmer might intend. For example, there are many ways to solve a statistical mathematical problem in the SAS programming language. There may first be a sample taken, one or more statistical tests are run on each sample, then one or more regressions (or other statistical calculation) are applied and finally, one or more statistical tests are run on the results. Each design pattern may be defined using a conventional context-sensitive parse tree. The technique of recognition of the pattern accurately in practice and, more importantly, the idiomatic association of one or a group of design patterns to the arbitrary business meaning facilitates the allocation of resources according to user categorizations. The mechanism of the match may be accomplished easily, as simple as a cross-reference lookup table that returns the meaning as an encoded symbol or value. A simple example follows.

[0025]

TABLE 4
First Level of Abstraction
	Resource Use Level	Maps to
Program Step	(arbitrary units)	Design Pattern
UNIVARIATE	More than 7; Less than 20	|
REG	Any	|-> Simple Statistic Use
LOGISTIC	More than 3; Less than 10	|
UNIVARIATE	More than 20	|
REG	More than 10	|-> Average Statistic Use
LOGISTIC	More than 10; Less than 50	|
Any Combination	Sums more than 35	|
of above		|

[0026]

TABLE 5
Second Level of Abstraction
	Resource Use Measure	Maps to
Design Pattern	(arbitrary units)	User Profile
Any Statistical Model Activity	In Any Quantity	|
		|-> Modeler
Any Statistical	Sums to more than 15	|
Computation Activity		|-> Statistical Use
Simple Statistic Use		|
Average Statistic Use		|
AND		|
Exclusive of Previous		|
Classification Set		|
Data Summarization	More than 30	|
Data Manipulation	More than 15	|-> Data ETL Use
Average Statistic Use	Less than 9	|
Report Creation	Less than 9	|
AND		|
Exclusive of Previous Two		|
Classification Sets		|

[0027] A variety of management reports based on the results may then be generated. A sample report is shown in the following table.

[0028]

TABLE 6
User Categorizations
UserID	Department	User Profile Assignment
abd123	e45676	Modeler
sde456	f56788	Statistical
fgy789	g56789	Data_ETL

[0029] Referring to FIG. 1, a block diagram illustrating categorization of users according to an example embodiment is shown. Users may be categorized according to their resource use and based on the needs of the business, resources may be allocated to users in based on the categories. For example, the business may decide to allocate more resources to smaller groups of users that perform functions identified as providing the greatest business value and to allocate fewer resources to larger groups of users that perform functions that are considered less vital to the business.

[0030] While certain embodiments of the disclosed system and method for user categorization are described in detail above, the scope of the invention is not to be considered limited by such disclosure, and modifications are possible without departing from the spirit of the invention as evidenced by the claims. For example, design patterns and measurement attributes may be varied and fall within the scope of the claimed invention. The number and types of user categorizations may be varied and fall within the scope of the claimed invention. One skilled in the art would recognize that such modifications are possible without departing from the scope of the claimed invention.