International Business Machines Corporation

Rajesh Sambandhan

Manoj Kumar Agarwal

Amitava Kundu

Curt Lee Cotner

Prasan Roy

Mangesh V Shanbhag

Sequences of hierarchical records are aggregated and summarized. A capture log that includes a plurality of operations of a workload is received. A first data structure that models transaction types as sequences of nodes is created. The nodes identify operations in the workload. A present operation and a transaction identifier are read from the capture log. The transaction identifier is dissociated from a first node that identifies a prior operation. The transaction identifier is associated with a second node that identifies the present operation. In a second data structure that associates nodes with transaction identifiers, the first node is dissociated from the transaction identifier and the second node is associated with the transaction identifier. A summary of the workload is generated based, at least in part, on the first and second data structures. The summary includes signatures of transaction types and counts of instances of the transaction types.

Aggregating and summarizing sequences of hierarchical records

Database operation records are sequentially read from two or more log files. If the transaction identifier is new and the record is not an end-of-transaction record, an open transactions list entry is created. If the transaction identifier is new and the record is an end-of-transaction record, a transaction type list entry is created or updated. If the transaction identifier is not new and is not an end-of-transaction record, an open transactions list entry is updated. If the transaction identifier is not new and the record is an end-of-transaction record, a transaction type list entry is created or updated. When all log file records are read, analytical comparison between the information associated with two or more of the log files in data fields in the transaction type list entries is performed.

Comparing database performance without benchmark workloads

Identifying matching transactions. First and second log files contain operation records of transactions in a transaction workload, each file recording a respective execution of the transaction workload, the method comprising. A first record location in the first file and an associated window of a defined number of sequential second record locations in the second file are advanced one record location at a time. Whether each operation record of a complete transaction at a first record location has a matching operation record at one of the record locations in the associated window of second record locations is determined. If so, the complete transaction in the first file and the transaction that includes the matching operation records in the second file are identified as matching transactions.

Matching transactions in multi-level records

A predefined number of matches is identified between records in a first file and records in a second file. For the matches, determine the span of the actual range of record positions in the second file relative to the positions of the operation records in the first file within which all matches were found. If the actual span is smaller than the span of a current defined range of record positions by at least a first threshold value, decrease the span of the current defined range. If the actual span is within a second threshold value of the span of the current defined range, increase the span of the current defined range. If an amount above a third threshold value of operation records in the first file are not matched to operation records in the second file, increasing the span of the current defined range.

Match window size for matching multi-level transactions between log files

A method for identifying matching transactions between two log files where each transaction includes one or more statements. Each log file record records the execution of a statement and includes a transaction identifier. Each record in turn in one log file is compared to an advancing window of records in the other log file. A first table contains associations of statements to transactions and transactions to statements for records in the window. If a match is found between a record in the one file and a record in the window, information associating partial transactions in the one file to potential transactions of the records in the window is added to a second table. If an end-of-transaction record is read from the one file, a best match is found between the ended transaction and the potential transactions based on information in the first and second tables.

Growing community of inventors

Bengaluru, India

Rajesh Sambandhan