Marco Russo and Alberto Ferrari track account balances:
Computing an inventory level or an account balance at a given time is a common requirement for many reports. However, when the source data contains all the transactions since the initial zero state, the calculation requires a running sum from the beginning of the data history until the day considered. While easy to implement, a calculation like this can be extremely expensive depending on several factors: the number of cells to compute in the report, the data volume of the transactions, and the cardinality of the dimensions.
The usual approach to optimizing this type of calculation is to introduce a snapshot table that pre-calculates the value of each date for all the dimensions required. Because of the resulting data volume, this solution can be very expensive both in terms of processing time and in terms of resulting memory consumption. A tradeoff is to limit the cardinality of the time available for the snapshot, for example by creating a monthly or quarterly snapshot instead of a daily snapshot. However, this approach limits the analysis of inventory or balance amount trends, and it removes any detail below the snapshot cardinality.
This article shows how to implement a hybrid approach that minimizes the snapshot cost without losing analytical capabilities. This provides outstanding query performance for the reports.
To an extent, this replicates what accountants do with general ledger operations: after an accounting period ends, you summarize results. Then, you can start from the summarized end point instead of needing to read each journal entry. It also provides a convenient time to perform audits and ensure that the numbers are what they should be.