Taurus: A Parallel Transaction Recovery Method Based on Fine-Granularity Dependency Tracking
نویسندگان
چکیده
Logging is crucial to performance in modern multicore main-memory database management systems (DBMSs). Traditional data logging (ARIES) and command logging algorithms enforce a sequential order among log records using a global log sequence number (LSN). Log flushing and recovery after a crash are both performed in the LSN order. This serialization of transaction logging and recovery can limit the system performance at high core count. In this paper, we propose Taurus to break the LSN abstraction and enable parallel logging and recovery by tracking fine-grained dependencies among transactions. The dependency tracking lends Taurus three salient features. (1) Taurus decouples the transaction logging order with commit order and allows transactions to be flushed to persistent storage in parallel independently. Transactions that are persistent before commit can be discovered and ignored by the recovery algorithm using the logged dependency information. (2) Taurus can leverage multiple persistent devices for logging. (3) Taurus can leverage multiple devices and multiple worker threads for parallel recovery. Taurus improves logging and recovery parallelism for both data and command logging. Our evaluation on a 32-core machine with four persistent devices shows that Taurus can improve the data (command) logging performance by up to 3.8× (2.67×) and reduce recovery time by 4.9× (9.8×) compared to baseline algorithms.
منابع مشابه
Aries: a Transaction Recovery Method Supporting Fine-granularity Locking and Partial Rollbacks Using Write-ahead
متن کامل
A Performance Study of Locking Granularity in Shared-Nothing Parallel Database Systems
Locking granularity refers to the size of a lockable data unit, called a "granule", in a database system. Fine granularity improves system performance by increasing concurrency level but it also increases lock management overhead. Coarse granularity, on the other hand, sacrifices system performance but lowers the lock management cost. This paper explores the impact of granule size on performanc...
متن کاملA hybrid approach for database intrusion detection at transaction and inter-transaction levels
Nowadays, information plays an important role in organizations. Sensitive information is often stored in databases. Traditional mechanisms such as encryption, access control, and authentication cannot provide a high level of confidence. Therefore, the existence of Intrusion Detection Systems in databases is necessary. In this paper, we propose an intrusion detection system for detecting attacks...
متن کامل3-RPS Parallel Manipulator Dynamical Modelling and Control Based on SMC and FL Methods
In this paper, a dynamical model-based SMC (Sliding Mode Control) is proposed fortrajectory tracking of a 3-RPS (Revolute, Prismatic, Spherical) parallel manipulator. With ignoring smallinertial effects of all legs and joints compared with those of the end-effector of 3-RPS, the dynamical model ofthe manipulator is developed based on Lagrange method. By removing the unknown Lagrange multipliers...
متن کاملNew Maximum Power Point Tracking Technique Based on P&O Method
In the most described maximum power point tracking (MPPT) methods in the literatures, the optimal operation point of the photovoltaic (PV) systems is estimated by linear approximations. However, these approximations can lead to less optimal operating conditions and significantly reduce the performances of the PV systems. This paper proposes a new approach to determine the maximum power point (M...
متن کامل