Self-maintaining real-time data aggregation

FIELD OF THE INVENTION

0001The present invention relates to methods and computer systems for monitoring a workflow of a business or other organization. More particularly, the present invention relates to methods for aggregating information about multiple instances of an activity and for maintaining that aggregation.

BACKGROUND OF THE INVENTION

0002Computers, and in particular, computer database applications, are used by businesses and other organizations to monitor and record information about an organization's activities. Often, the organization will have various processes or activities that must be performed, and which recur frequently. Indeed, it is common for an organization to have numerous instances of an activity in various stages of completion at any given time. As one example, a business may sell goods based on orders received from customers. An activity of interest may be fulfilling those customer orders; each purchase order represents a separate instance of that activity. At any particular time, that business may have multiple instances of the activity (i.e., multiple orders from multiple customers) in various stages of completion. As but another example, a financial institution may loan funds to customers based on applications from those customers. An activity of interest may be the processing of a loan application to completion (e.g., approval or rejection), with each application representing a separate instance of the activity. At any particular time, there may be multiple loan application instances in various stages of processing. As yet another example, a governmental entity responsible for issuing permits may have multiple permit applications in various stages of being processed.

0003<patcit id="pcit0001" dnum="US20030149702A"><text>US 2003/0149702</text></patcit> discloses a database with batched accumulation data in an aggregation buffer.

0004In order to monitor numerous instances of an activity, many organizations store information about those activity instances in a database program. In particular, a record or other data object can be created for each instance of the activity. A separate field or other component of the record is then established to hold a value for some type of information common to each instance. Using one of the previous examples as an illustration, a business selling goods may create a separate database record for each customer order. Within that record may be separate fields for the time the order was received, where the order was received, what was ordered, when the order was shipped, etc. Such use of a database program is often conceptualized as a table. Each instance of the activity is assigned a separate row (or tuple) of the table. Each type of information common to multiple instances is then assigned a separate column of the table.

0005Although the values of individual fields in individual records may sometimes be needed, many organizations frequently need information about groups of records. Moreover, this information is often needed in real-time. For example, many businesses that sell goods need to know how many orders are currently pending, how many orders have been completed, and how many orders are in one or more intermediate stages of completion. Certain database programs are able to provide such reports by aggregating values within multiple records of the database. Without more, however, this is often an unacceptable solution where the database is very large.

0006As more and more records accumulate, the speed with which a database can be accessed drops significantly. For a large business such as a goods seller receiving hundreds or thousands of orders per day, the number of records can reach into hundreds of thousands or millions. Each time the database is queried, a finite amount of time is needed to search a disk drive or other storage device. Similarly, as new records are created and existing records updated, a finite amount of time is needed to create or update each of those records. As the number of records grows, the time needed to find a particular record increases. In a business or organization having hundreds (or thousands) of users and hundreds of thousands (or millions) of database records, the latency for database system access can become quite substantial. Moreover, with numerous users attempting access the same information within the database, deadlocks between users trying to access the same records can occur. If numerous users are inserting and updating records into a large database while other users are attempting to access the database in order to generate a summary of various fields, all users may experience less-than-satisfactory database performance.

0007<patcit id="pcit0002" dnum="US5794241A"><text>US 5,794,241</text></patcit> addresses this problem by introducing a method for dynamically disabling and enabling table locking in a relational database.

0008Another possible solution is to generate an On-Line Analytical Processing (OLAP) cube for data in a database. However, the processing required for generating OLAP cubes can also be quite time-consuming. If there are a significant number of database records, OLAP cubes can often only be generated on a daily (or sometimes hourly) basis. If an organization needs aggregated information in real-time, OLAP cubes will often not suffice.

SUMMARY OF THE INVENTION

0009The present invention addresses the above and other challenges associated with maintaining aggregated information about multiple instances of an activity. In at least one embodiment, the invention includes a method for maintaining aggregations of values contained by fields of multiple database records. The method includes creating multiple aggregation groups. Each group includes a plurality of aggregation records, and each aggregation record includes a value for an aggregation of values contained by fields of a different subset of the multiple database records. The method further includes selecting a first aggregation group upon insertion or update of a first of the multiple database records. The method also includes revising, based on one or more values within the inserted or updated first database record, and as part of a first aggregation group update transaction, the aggregation value of one of the aggregation records of the first aggregation group. Subsequent selection of the first aggregation group is prevented until completion of the first aggregation group update transaction. A second aggregation group is selected while the first aggregation group update transaction is being performed and upon insertion or update of a second of the multiple database records. Based on one or more values within the inserted or updated second database record and during the first aggregation group update transaction, the aggregation value of one of the aggregation records of the second aggregation group is revised. In other aspects of the invention, the aggregation groups are combined into a single table of aggregation records.

0010In at least another embodiment, the invention includes a method for maintaining aggregated data regarding multiple instances of an organizational activity, each instance of the activity having one of a plurality of process states. The method includes creating a plurality of records in an aggregated data table, each record containing an aggregated value for a subset of the multiple instances in the same process state during the same time period. The method further includes updating the aggregated data table to reflect deletion of data corresponding to instances in one of the process states outside of a preselected time window.

0011These and other features and advantages of the present invention will be readily apparent and fully understood from the following detailed description of preferred embodiments, taken in connection with the appended drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

0012<ul id="ul0001" list-style="none"><li><figref idref="f0001">FIG. 1</figref> is a block diagram showing processing of customer orders by a hypothetical wholesale business.</li><li><figref idref="f0002">FIG. 2</figref> is a portion of an instances data table for the business of <figref idref="f0001">FIG. 1</figref>.</li><li><figref idref="f0003">FIG. 3</figref> is a table aggregating data from various fields of the table of <figref idref="f0002">FIG. 2</figref>.</li><li><figref idref="f0004">FIG. 4</figref> shows update of an aggregated data table to reflect a new instances data record.</li><li><figref idref="f0005">FIG. 5</figref> shows update of an aggregated data table to reflect update of an existing instances data record.</li><li><figref idref="f0006">FIG. 6</figref> shows a deadlock between multiple program threads.<figref idref="f0007">FIG. 7</figref> shows a multi-partition aggregation table according to at least one embodiment of the invention.</li><li><figref idref="f0008">FIG. 8</figref> shows implementation of the aggregation table of <figref idref="f0007">FIG. 7</figref> according to at least one embodiment of the invention.</li><li><figref idref="f0009">FIG. 9</figref> shows a view combining the partitions of a multi-partition aggregation table according to at least one embodiment of the invention.</li><li><figref idref="f0010">FIG. 10</figref> shows one implementation of a stored procedure to assign partitions to program threads.</li><li><figref idref="f0011">FIG. 11</figref> shows another table providing aggregated data and in which a business milestone timestamp is used as an aggregation criterion.</li><li><figref idref="f0011">FIG. 12</figref> illustrates a table facilitating, according to one embodiment of the invention, deletion of stale aggregated data.</li><li><figref idref="f0012">FIG. 13</figref> is a flow chart showing logic for a trigger to maintain a real-time data aggregation table according to at least one embodiment of the invention.</li></ul>

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

0013The present invention can be advantageously used in combination with the methods, apparatus and systems described in <patcit id="pcit0003" dnum="US157968A" dnum-type="L"><text>U.S. Patent Application Ser. No. 10/157,968</text></patcit>, titled "Support for Real-Time Queries Concerning Current State, Data and History of a Process" and filed on May 31, 2002, the contents of which are incorporated by reference herein.

0014The present invention will be described by reference to Structured Query Language (SQL) instructions and other data analysis features found in the SQL SERVER™ 2000 relational database management system (RDBMS) software and associated Online Analytical Processing (OLAP) services software available from Microsoft Corporation of Redmond, Washington. Although some aspects of SQL instructions that may be used to implement certain embodiments of the invention are described herein, other instructions, programming algorithms and procedures used to implement the invention will be apparent to persons skilled in the art once those persons are provided with the description provided herein. General descriptions of SQL SERVER™ 2000 RDBMS software and associated OLAP services software can be obtained from various sources, including Inside Microsoft® SQL SERVER™ 2000 by Karen Delaney (2001 Microsoft Press) and Microsoft® SQL SERVER™ 2000 Books Online, available at <http://www.microsoft.com/sql/techinfo/productdoc/2000/>. The invention is not limited to implementation using SQL SERVER™ 2000 RDBMS software and associated OLAP services software, and may be implemented using other types of RDBMS and OLAP software.

0015The present invention will also be described by reference to RDBMS software (such as the aforementioned SQL SERVER™ 2000 software) operating on a server and accessed by one or more clients. Such configurations are known in the art and described in, e.g., the previously-incorporated <patcit id="pcit0004" dnum="US157968A" dnum-type="L"><text>U.S. patent application 10/157,968</text></patcit>. However, a client-server configuration is only one example of a manner in which the invention can be implemented. The invention can also be implemented in other physical system configurations.

0016<figref idref="f0001">FIG. 1</figref> is a block diagram showing processing of customer orders by a hypothetical wholesale business which sells goods to customers based on customer purchase orders. For convenience, the business will be referred to herein as "Business X." Business X receives purchase orders from multiple customers. Upon receipt of each purchase order, Business X inputs information about the order into a database maintained on a database server. Specifically, Business X creates a new record for the purchase order in the database, with fields for various types of information common to individual purchase orders. In the example, each record includes fields for purchase order number, date and time of receipt, city in which the customer is located, and quantity of goods ordered. Business X determines whether each purchase order will be accepted or rejected, and then updates one or more other fields in the database to reflect the acceptance or rejection. If a purchase order is accepted, a corresponding sales order is generated and sent to the Business X warehouse located in the city of the ordering customer. When goods are shipped to a customer, the time of shipment is input (via a client computer at one of the warehouses) into another field of the record. When goods are delivered, yet another field is updated.

0017Although the above example has been created to describe the invention, the example (including additional aspects described herein) is representative of many actual businesses, albeit in simplified form. Moreover, persons skilled in the art will appreciate that the concepts described with regard to hypothetical Business X are applicable to a broad spectrum of business and other organizational activities. Indeed, the invention could alternatively be described by a generic reference to an "organization" instead of "Business X." Similarly, instead of referring to purchase orders and various stages of order fulfillment, the invention could be described using generic terms such as "instance of an organizational process," a "state" of the organizational process instance, completion of an organizational process instance, etc. Although a simplified example of an actual business type is used in order to provide a more readable description, the invention is not limited to a particular type of organization or organizational activity.

0018<figref idref="f0002">FIG. 2</figref> is a portion of a table from the database of <figref idref="f0001">FIG. 1</figref>, and is maintained by Business X and used to store data regarding individual purchase order instances. This instances data table has individual records (e.g., rows) for each purchase order and individual fields (columns) for various types of data. In the example, "PO#" is a purchase order number. "RecvTime" is the date and time the purchase order was received, "City" is the city in which the warehouse closest to the customer issuing a purchase order is located, and "Quantity" is the number of items ordered. "ShipTime" is the time goods for a purchase order were shipped, and "DeliveryTime" is the time those goods were delivered. "ProcessState" is a variable used to describe the portion of the process a purchase order is currently undergoing. If a purchase order has been received but no goods have been shipped, the purchase order is "InProcess." If goods have been shipped but not delivered, the purchase order is "Shipped." If goods have been delivered, the purchase order is "Delivered." Although not shown, there could be other values for ProcessState (e.g., "Denied" of a purchase order was denied). Purchase orders for which goods have not yet been shipped have a <NULL> value in the ShipTime field. Similarly, purchase orders for which goods have not been delivered have a <NULL> value in the DeliveryTime field. As purchase orders are shipped and/or delivered, those fields are updated with the appropriate time values.

0019As can be seen from <figref idref="f0002">FIG. 2</figref>, and even with only a limited number of records, a table of raw data can become unwieldy when aggregated information is desired. If, for example, the total goods delivered from a particular warehouse is needed, it is necessary to check each row. A more useful table format for obtaining aggregated data is shown in <figref idref="f0003">FIG. 3</figref>. In particular, <figref idref="f0003">FIG. 3</figref> shows, by warehouse, the total number of purchase orders ("Count") that are currently delivered, in process or shipped. <figref idref="f0003">FIG. 3</figref> also shows, in the Quantity column, the total number of goods that each warehouse has delivered, shipped, and is processing for shipment.

0020<figref idref="f0004">FIG. 4</figref> shows how the table of <figref idref="f0003">FIG. 3</figref> would be updated to reflect receipt of a new purchase order number 135 for 30 units from Redmond. Specifically, the "InProcess" total for Redmond is increased by 30 and the Count total is increased by 1. <figref idref="f0005">FIG. 5</figref> shows updating the table of <figref idref="f0004">FIG. 4</figref> when purchase order number 135 is shipped. The "Quantity" and "Count" totals for Redmond/InProcess purchase orders are respectively decreased by 30 and 1, while the same fields for Redmond/Shipped purchase orders are respectively increased by 30 and 1.

0021Although the updates of <figref idref="f0004">FIGS. 4</figref> and <figref idref="f0005">5</figref> are relatively straightforward, problems can occur when multiple programming threads are simultaneously accessing the same aggregated data table. <figref idref="f0006">FIG. 6</figref> shows two program threads attempting to simultaneously perform transactions on the same table. For simplicity, it is assumed that there is a clustered index on City and on ProcessState (i.e., data rows are stored in order based on the clustered index keys), allowing the records to be accessed from top to bottom in the order shown in <figref idref="f0006">FIG. 6</figref>. In many database environments, multiple records are modified as part of a single transaction that accesses a table. Because of the processing overhead required for each transaction, batch processing multiple records is often more efficient, It is also a common practice to require that transactions be performed on an "all-or-nothing" basis. In other words, no records are modified unless all parts of the transaction can be completed. Otherwise, integrity of the table could be compromised.

0022These database processing constraints can cause deadlock. Deadlock occurs in <figref idref="f0006">FIG. 6</figref> when program thread A attempts to update three records in one transaction: update Redmond/InProcess and Redmond/Shipped to reflect that purchase order 134 has shipped, and update Seattle/InProcess to reflect new purchase order 136. Simultaneously, program thread B also attempts to update three records in one transaction: update Seattle/InProcess and Seattle/Shipped to reflect that purchase order 133 has shipped, and then update Redmond/InProcess to reflect new purchase order 135. In order to perform its transaction, thread A acquires an exclusive lock on Redmond/InProcess and Redmond/Shipped so as to prevent other threads from affecting those records while they are being updated by thread A. Thread A then attempts to acquire an exclusive lock on Seattle/InProcess. At the same time, thread B acquires an exclusive lock on the records Seattle/InProcess and Seattle/Shipped and makes a modification, and then tries to acquire an exclusive lock on Redmond/InProcess. Because thread A has already locked Redmond/InProcess, thread B cannot complete the first part of its transaction. However, thread B is able to lock Seattle/InProcess and Seattle/Shipped before thread A attempts to lock Seattle/InProcess. Thread A is thus prevented from completing the second part of its transaction. In effect, each thread is waiting for the other to finish, and neither can complete its transaction. Some systems will choose one of the threads as a deadlock victim, roll back any partially completed portions of the victim's transaction, and report an error to the victim. Although this allows the other thread to proceed, and although the victim can retry its transaction, system performance is nonetheless degraded. In particular, processing resources and memory are wasted undoing partially completed transactions, and then further wasted by repeating previously-completed portions of a transaction.

0023<figref idref="f0007">FIG. 7</figref> shows how deadlock is prevented according to at least one embodiment of the invention. A table of aggregated values is partitioned into separate tables. <figref idref="f0007">FIG. 7</figref> shows two partitions (separated by a bold line), and the presence of additional partitions is indicated by the vertical ellipsis. Each of the individual partitioned tables 0, 1, etc. contains the same records, but the values in those records will usually differ among the partitions. In the example of <figref idref="f0007">FIG. 7</figref>, partition 0 has records for RedmondlDelivered, Redmond/InProcess, Redmond/Shipped, Seattle/Delivered, Seattle/InProcess and Seattle/Shipped. Partition 1 also has records for Redmond/Delivered, Redmond/InProcess, Redmond/Shipped, Seattle/Delivered, Seattle/InProcess and Seattle/Shipped. However, the values of Quantity and Count for Redmond/Delivered in partition 0 are different from the values of Quantity and Count for Redmond/Delivered in partition 1, etc. Only a single program thread is permitted to access a particular partition at one time. To illustrate, threads A and B from <figref idref="f0006">FIG. 6</figref> are now shown in <figref idref="f0007">FIG. 7</figref>. In <figref idref="f0007">FIG. 7</figref>, thread A performs its transaction on partition 0, while thread B simultaneously performs its transaction on partition 1. Because the threads are no longer competing for access to the same records, deadlock is avoided.

0024In one embodiment of the invention implemented using SQL SERVER™ 2000 RDBMS software, the partitioned tables of <figref idref="f0007">FIG. 7</figref> are implemented as a separate SQL table, such as is shown in <figref idref="f0008">FIG. 8</figref>. The partition number is included as an additional field ("PartitionID"). This multi-partition aggregation table is maintained by a trigger on the instances data table of <figref idref="f0002">FIG. 2</figref>. As known in the art, a "trigger" is a special type of SQL stored procedure which is automatically invoked by an INSERT, UPDATE or DELETE statement. Whenever a new record is inserted into the instances data table of <figref idref="f0002">FIG. 2</figref>, the trigger causes a corresponding contribution to be made to one of the partitions of the multi-partition aggregation table of <figref idref="f0008">FIG. 8</figref>. Similarly, when a record in the instances data table (<figref idref="f0002">FIG. 2</figref>) is updated, the corresponding records of multi-partition aggregation table (<figref idref="f0008">FIG. 8</figref>) are also updated.

0025At any given time, an individual partition in the table of <figref idref="f0008">FIG. 8</figref> will not have complete information about Business X. One update to the instances data table (<figref idref="f0002">FIG. 2</figref>) may cause an update within partition 0, while another update to the instances data table may cause an update within partition 1, etc. Moreover, creation of a record in the instances data table may cause update of a record in the table of one partition of <figref idref="f0008">FIG. 8</figref>, and a later update of that same instances data record might cause update of records in a different partition. For example, adding new purchase order number 140 (Redmond, Quantity=200) to <figref idref="f0002">FIG. 2</figref> might cause the Quantity and Count fields of the <PartitionID=1>/<City=Redmond>/<ProcessState=InProcess> record in <figref idref="f0008">FIG. 8</figref> to be increased by 200 and 1. However, when goods for purchase order 140 are shipped, the Quantity and Count fields of the <PartitionID=3>/<City=Redmond>/<ProcessState=InProcess> record in <figref idref="f0008">FIG. 8</figref> might be reduced by 200 and 1 and the same fields of the <PartitionID=3/<City=Redmond>/<ProcessState=Shipped> record increased by 200 and 1. Accordingly, the partitions are combined to provide a complete data aggregation table for Business X. In particular, the following SQL code is used in at least one embodiment: <img file="EP1519283B1_D0001.tif" /> The first italicized name <i>("view_name")</i> is a name for a view in which the individual partitions are combined. The second italicized name <i>("multi_partition_table")</i> is the name of the multi-partition aggregation table of <figref idref="f0008">FIG. 8</figref>. The resulting table <i>("view_name")</i> would be in the form shown in <figref idref="f0009">FIG. 9</figref>. Because each partition of the multi-partition aggregation table contains relatively few records, the partitions can be combined relatively quickly. In effect, the summaries are summarized. In other embodiments, each partition can be implemented as a separate table and then joined in a CREATE VIEW command with the SQL JOIN statement.

0026So that each partition is only accessed by one program thread at a time, a thread must have possession of a virtual token in order to access a partition. In one embodiment of the invention, the token is obtained by a call to a special stored procedure named Get_Mutex. As previously discussed, the multi-partition aggregation table is maintained by a SQL trigger. Before a row within the multi-partition aggregation table can be updated, one or more instructions within the trigger must indicate which row is to be updated. As part of those instructions, the Get_Mutex stored procedure is called. The Get_Mutex procedure then returns a value for the PartitionID column of the table. While a thread is performing a transaction on records having that returned PartitionID value, other threads are prevented from obtaining the same value, and thereby prevented from accessing a partition while it is being updated by another thread.

0027Referring to a prior example, when a record for purchase order 140 is initially created in the instances data table (<figref idref="f0002">FIG. 2</figref>), the trigger is fired. That trigger then: <ul id="ul0002" list-style="none"><li>calls Get_Mutex and receives a value for PartitionID;</li><li>determines the row(s) within the partition to be updated;</li><li>identifies the row of the multi-partition aggregation table where PartitionID equals the PartitionID value returned by the Get_Mutex procedure, where City equals the value of City in the instances data record for purchase order 140, and where ProcessState equals "InProcess";</li><li>increases the value for Quantity in that row by the value of Quantity in the instances data record for purchase order 140; and</li><li>increases the value for Count in that row by 1.</li></ul>

0028As a further illustration, an existing instances data table record for purchase order 140 is later updated to change ProcessState from InProcess to Shipped. Additional logic in the trigger would then: <ul id="ul0003" list-style="none"><li>call Get_Mutex and receive a value for PartitionID;</li><li>determine the row(s) within the partition to be updated;</li><li>identify the row of the multi-partition aggregation table where PartitionID equals the value returned by the Get_Mutex procedure, where City equals the value of City in the instances data record for purchase order 140, and where ProcessState equals InProcess;</li><li>decrease the value for Quantity in that row by the value for Quantity in the instances data record for purchase order 140;</li><li>decrease the value of Count in that row by 1;</li><li>identify the row of the multi-partition aggregation table where PartitionID equals a value returned by the Get_Mutex procedure, where City equals the value of City in the instances data record for purchase order 140, and where ProcessState equals Shipped;</li><li>increase the value for Quantity in that row by the value for Quantity in the instances data record for purchase order 140; and</li><li>increase the value of Count in that row by 1.</li></ul>

0029Similar logic would identify and modify the appropriate rows in the multi-partition aggregation table when purchase order 140 ProcessState changes from Shipped to Delivered.

0030One implementation for Get_Mutex is shown in <figref idref="f0010">FIG. 10</figref>. First, a single-column table named RTA_Mutex is created ("create table RTA_Mutex"). The RTA_Mutex table has the same number of rows as there are partitions in the multi-partition aggregation table of <figref idref="f0008">FIG. 8</figref>; each field is then assigned one of the PartitionID values. In the example of <figref idref="f0010">FIG. 10</figref>, it is assumed that the table of <figref idref="f0008">FIG. 8</figref> includes 10 partitions (i.e., PartitioniD has integer values from 0 through 9). Next, the Get_Mutex stored procedure is created. After declaring the local variable @par, the local @par variable is assigned a value from one of the rows of the RTA_Mutex table. Specifically, the "select" statement assigns to @par the value of PartitionID from the RTA_Mutex table where PartitionID equals "@@spid%10". The system function @@spid returns the server process identifier of the current user process. In other words, the @@spid function returns a number identifying the program thread that called the Get_Mutex stored procedure. The modulo arithmatic operator (" % ") then returns the remainder of that program thread identifier divided by 10. The procedure then finds a row of RTA_Mutex where the value for PartitionID equals that remainder, and locks that row with an exclusive lock locking hint ( "(xlock)" ). The value of PartitionID in that locked RTA_Mutex row is then returned as the result of the Get_Mutex stored procedure. Because of the exclusive lock on that row, no other programming threads are allowed to access that row of the RTA_Mutex table, and are therefore unable to obtain that PartitionID value until completion of the transaction that previously acquired that PartitionID value.

0031By way of illustration, a program thread having an identifier of 231 calls Get_Mutex. The remainder of 231 divided by 10 is 1; the value returned to the programming thread by Get_Mutex is thus the value of the field in RTA_Mutex where PartitionID equals 1. Upon return of a value from Get_Mutex, program thread 231 is then able to update row(s) of partition 1 in <figref idref="f0008">FIG. 8</figref>. While thread 231 is updating partition 1, program thread 161 calls Get_Mutex. However, because the transaction of thread 231 is not completed, thread 161 is not able to access the row of RTA_Mutex holding a value of 1 for PartitionID. Thread 161 is then queued until the exclusive lock on the necessary RTA_Mutex row is released (i.e., the transaction of thread 231 completes). If additional threads attempt to access that RTA_Mutex row, they are also queued on a first-in-first-out (or other) basis. Notably, thread 161 does not proceed until a value is returned from its call to Get_Mutex, and thread 161 therefore does not attempt to access partition 1 while thread 231 is accessing partition 1. While thread 161 is waiting in the queue for the xlock to release, thread 154 calls Get_Mutex. Because there is currently no lock on the row of RTA_Mutex where PartitionID = 4, Get_Mutex returns a value (4) to thread 154.

0032<figref idref="f0010">FIG. 10</figref> is only one example of a manner in which a Get_Mutex procedure could be implemented. As another example, all threads attempting to access a partition could enter a first-in-first-out queue. Each thread in the queue could then be assigned the first available partition.

0033In one embodiment, the number of partitions is at least equal to, and preferably greater than, the number of processors on the database server.

0034As previously indicated, the size of the instances data table (<figref idref="f0002">FIG. 2</figref>) will grow over time as more and more purchase orders are received. However, the size of the aggregation table of <figref idref="f0009">FIG. 9</figref> (or of the multi-partition aggregation table of <figref idref="f0008">FIG. 8</figref>) will not increase unless additional aggregation fields are added (e.g., Redmond/Denied and Seattle/Denied). Although the size of the tables will not grow, however, the values of the Quantity and Count columns will increase. Often, an organization only requires aggregated data for events within a certain time "window." For example, Business X may require aggregated data for purchase orders currently in process, for orders that are currently being shipped, and for orders that were delivered within the last 24 hours. Moreover, Business X may want to further sort the data based on, e.g., the time of day in which events occur. Referring to <figref idref="f0011">FIG. 11</figref>, Business X managers wish to know how many purchase orders became in process in Redmond and in Seattle during the most recent 8:00 a.m. hour and have not been shipped (0 in the example), during the most recent 9:00 a.m. hour (also 0 in the example), during the most recent 10:00 a.m. hour (2 and 1), etc. Similarly, the managers also want to know how many purchase orders were shipped at 8:00 a.m. but not yet delivered, were shipped at 9:00 a.m. but not yet delivered, etc. The managers also want to know how many purchase orders were delivered from those locations during those time periods. However, this means that the aggregation table grows each hour. Unlike purchase orders that are "InProcess" or "Shipped," and which will ultimately become "Delivered," a "Delivered" purchase order does not transition to another state. Put differently, once a purchase order becomes "Delivered" at a particular hour on a particular day, a record will persist in the aggregation table for that hour/day unless some action is taken. Over time, the table would thus become large and slow to query. On the other hand, the managers are less interested in knowing the total number of purchase orders that have been delivered from a location over a longer period (e.g., in the last week, etc.). Although such information may be useful for some purposes, it is needed relatively infrequently.

0035<figref idref="f0011">FIG. 12</figref> shows how, in at least one embodiment of the invention, data for inactive purchase order instances (e.g., delivered purchase orders) may be purged from the real-time aggregation table when that data reaches a certain age. For simplicity, only a single partition is shown. However, and as persons skilled in the art will appreciate based on the following description, the embodiment of <figref idref="f0011">FIG. 12</figref> can be implemented in conjunction with the previously-described multi-partition embodiment.

0036As seen in <figref idref="f0011">FIG. 12</figref>, columns for TimeSlice and Hour have been added to the real-time aggregation table of <figref idref="f0009">FIG. 9</figref>. TimeSlice is the server date and time (with hour precision) that the instances data table (<figref idref="f0002">FIG. 2</figref>) is updated to reflect that an instance is completed. In the example, a purchase order becomes completed when ordered goods have been delivered to the customer. Hour is the hour of the day in which the event of interest occurs. In the example, the time component of TimeSlice is the same as Hour for a completed record, although this need not be the case.

0037As in the previous example, the data of the real-time aggregation table of <figref idref="f0011">FIG. 12</figref> is modified by a trigger which fires when data records of the instances data table are inserted or updated. The value of the TimeSlice column is automatically generated by the trigger according to two rules. If more updates are expected to a record (e.g., the goods have not yet been delivered), TimeSlice is NULL. Otherwise, a number is generated that represents the current time (with hour precision) and date. Notably, a non-NULL TimeSlice value for a completed purchase order does not prevent aggregation of data for completed purchase orders. Because the time component of TimeSlice only has hour precision, purchase orders delivered from the same warehouse during the same 1 hour period can be aggregated in a single record. For example, purchase orders 126 and 128 (<figref idref="f0002">FIG. 2</figref>) were delivered at 3:10 p.m. and 3:05 p.m., respectively. As seen in <figref idref="f0011">FIG. 12</figref>, the record for Jan 20 3 p.m./3/Seattle/Delivered shows a Count of 2 and a Quantity of 790, which correlates with data records for purchase order 126 and purchase order 128 in <figref idref="f0002">FIG. 2</figref>.

0038Each time the trigger modifies the real-time aggregation table of <figref idref="f0011">FIG. 12</figref>, the trigger also determines whether any of the completed instance aggregated data in the table is stale, i.e., no longer needed. In the example, it is assumed that aggregated data for instances completed more than 24 hours ago is stale and should be removed from the aggregated data table. Accordingly, each time the trigger modifies the table of <figref idref="f0011">FIG. 12</figref>, the trigger deletes records having a TimeSlice value in which the latest time in that time slice period is more than 24 hours prior to the current time. If the trigger updates the table of <figref idref="f0011">FIG. 12</figref> at 1:05 p.m. on January 21, the record for Jan 20 12p.m./12/Seattle/Delivered will be deleted, as all data in that record corresponds to purchase orders completed more than 24 hours ago. In at least one embodiment, the table of <figref idref="f0011">FIG. 12</figref> is displayed in a view which hides the TimeSlice data from the user.

0039<figref idref="f0012">FIG. 13</figref> is a flow chart showing logic for a trigger that maintains a real-time data aggregation table (such as in <figref idref="f0011">FIG. 12</figref>) in multiple partitions. In particular, the flow chart of <figref idref="f0012">FIG. 13</figref> shows logic that is followed by the trigger in each individual programming thread during a transaction to update one or more aggregated data records in a partition. Each thread updates an individual partition, with each partition having a format similar to the table of <figref idref="f0011">FIG. 12</figref>. During execution, the trigger accesses two system tables named "inserted" and "deleted" that are maintained in memory (e.g., RAM) by a SQL database server. The inserted and deleted tables are automatically generated by the database server, and temporarily store copies of data from a row affected during a preceding attempt to insert a record into or to update an existing record of an instances data table (such as in <figref idref="f0002">FIG. 2</figref>). Specifically, the inserted table contains values inserted into a row of the instances data table, and the deleted table contains values replaced in an updated row of the instances data table.

0040After beginning, the trigger first obtains the token at block 102. The trigger obtains the token by a call to a procedure such as Get_Mutex, as previously described. At block 104, the trigger then determines whether an instances data record is being updated. In particular, the trigger determines if the deleted table is empty. If the deleted table is not empty, a Quantity value in the deleted table is used at block 106 to decrease the Quantity value for the appropriate record of the partition. In particular, the trigger identifies the record in the partition where Hour equals the hour during which the instances data record was updated to its previous ProcessState value, where City equals the city of the updated instances data record, and where ProcessState equals the previous process state of the updated record. The Count value for that record is also decreased by 1. At block 108, the trigger then determines whether any records in the partition have a NULL value for TimeSlice and a Count value of 0. If so, those records are deleted.

0041At block 110, the trigger determines whether the instances data record being updated (or inserted) is complete. In other words, the trigger determines whether further updates are expected. This can be performed in various ways. For example, an "IsComplete" column could be added to each instance record and set to "1" when the update is final. In other embodiments, various process state values (e.g., "Delivered" or "Denied") could signal process completion for the instance. If the instances data record is not complete (i.e., the process is still active as to the purchase order instance being updated and further updates are expected), the Quantity value from the "inserted" table is used at block 112 to increase the Quantity value of the record in the aggregation table partition where TimeSlice equals NULL, Hour equals the hour the instances data record was updated to its current ProcessState value, City equals the city of the updated instances data record, and ProcessState equals the current ProcessState value of the updated instances data record. The Count value in that partition record is also increased by 1. The trigger then concludes.

0042If at block 110 the trigger determines that the instances data record being updated is complete, a TimeSlice value is generated at block 114. At block 116, the trigger deletes all records having a TimeSlice value outside of the selected time window (e.g., older than 24 hours). The trigger then identifies (or creates, if necessary) a partition record in which the value of TimeSlice equals the just-generated TimeSlice value, where Hour equals the hour of completion, where City equals the city of the updated instances data record, and where ProcessState equals the value from the completed instances data record ("Delivered" in the example). The Quantity value from the "inserted" table is then used to increase the Quantity value of the identified (or created) partition record. The Count value for that partition record is also increased by 1. The trigger then concludes.

0043In other embodiments, the trigger logic of <figref idref="f0012">FIG. 13</figref> could be modified to accommodate updates of multiple records in the instances data table. Instead of ending after blocks 112 or 118, the trigger would determine whether there are additional instance records being updated as part of a batch update of the instances data table. If so, the trigger logic would loop back to block 104. If not, the trigger would then conclude. In still other embodiments, the time window could be adjusted by a user.

0044Although the invention has been described using a hypothetical business type as an example, it should be remembered that the invention is not limited to a particular type of business, organization or activity. Moreover, the invention is not limited to implementations in which values are aggregated by summing values of fields in database records. As but one example, an aggregation table could be created in which the aggregated values represent averages of values of fields in database records. Accordingly, although specific examples of carrying out the invention have been described, those skilled in the art will appreciate that there are numerous variations and permutations of the above described systems and techniques that fall within the spirit and scope of the invention as set forth in the appended claims. These and other modifications are within the scope of the invention as defined by the appended claims.