Ernestas Poškus

Technical blog

"We must view with profound respect the infinite capacity of the human mind to resist the introduction of useful knowledge." - Thomas R. Lounsbury

| github | goodreads | linkedin | twitter |

ansible 2 / elasticsearch 2 / kernel 2 / leadership 1 / linux 2 / mnemonics 1 / nginx 1 / paper 40 / personal 5 / rust 1 / tools 2 /

Write Behind Logging

WC 541 / RT 3min


Design of the logging and recovery components of database management systems (DBMSs) has always been influenced by the difference in the performance characteristics of volatile (DRAM) and non-volatile storage devices (HDD/SSDs). The key assumption has been that non-volatile storage is much slower than DRAM and only supports block-oriented read/writes. But the arrival of new nonvolatile memory (NVM) storage that is almost as fast as DRAM with fine-grained read/writes invalidates these previous design choices.

Recovery Principles

A DBMS guarantees the integrity of a database by ensuring

These two constraints are referred to as durability of updates and failure atomicity, respectively.

DBMS must protect against:


The most well-known recovery method based on WAL is the ARIES protocol developed by IBM in the 1990s. ARIES is a physiological logging protocol where the DBMS combines a physical redo process with a logical undo process. During normal operations, the DBMS records transactions’ modifications in a durable log that it uses to restore the database after a crash.

A disk-oriented DBMS maintains two meta-data tables at runtime that it uses for recovery. The first is the dirty page table (DPT) that contains the modified pages that are in DRAM but have not been propagated to durable storage.

The second table is the active transaction table (ATT) that tracks the status of the running transactions.

With an in-memory DBMS, transactions access tuples through pointers without indirection through a buffer pool.

Runtime operations

Commit processing



Write-behind logging leverages fast, byte-addressable NVM to reduce the amount of data that the DBMS records in the log when a transaction modifies the database. Individual bytes in NVM can be accessed by the processor, and hence there is no need to organize tuples into pages or go through the I/O subsystem.

WBL reduces data duplication by flushing changes to the database in NVM during regular transaction processing.

Log is always (slightly) behind the contents of the database.

Runtime operations

Commit processing



WBL - write behind logging

NVM - non volatile memory

DBMS - database management system

DPT - dirty page table

ATT - active transaction table