|Abstract:||The Sarbanes-Oxley Act inspired research on long-term high-integrity retention of business records, based on the long-term immutability guarantees that WORM storage servers offer for files. Researchers recently proposed a Log-compliant DBMS Architecture (LDA) that extends those immutability guarantees to relational tuples, using an approach that imposes a 10-20% performance penalty on TPC-C benchmark runs.
In this paper, we present the transaction log on WORM (TLOW) approach for supporting long-term immutability for relational tuples. TLOW incurs less than 1% runtime overhead on TPC-C benchmarks with Berkeley DB, which is much less than for LDA. TLOW requires no changes to the DBMS kernel, and audit time is comparable to that of LDA: 2.7% of transaction time, i.e. ten days for a yearly audit on the platform we used. We also introduce the audit helper (AH) add-on to TLOW, which decreases the cost of a yearly audit on our platform to two hours. We provide a proof of correctness for TLOW, which exposes a subtle threat. The proof also illustrates a non-obvious problem with LDA, which we show how to correct.