Using the DLM as a Distributed In-Memory Database


One Line Summary

Presentation of an approach for how to use the Linux Distributed Lock Manager (DLM) to implement persistent reservation support for highly available SCSI clusters efficiently.


In a fully redundant SAN setup both initiator and target systems are redundant. A popular approach to decide which initiator system should be active is to use a quorum disk and to control ownership of the quorum disk via persistent reservations. Although several open source SCSI target implementations exist for Linux, none of these subsystems supports this scenario today because persistent reservations are processed locally. One possible approach to keep the persistent reservation state in sync between the SCSI target systems is to use the DLM as a distributed in-memory database. The purpose of this talk is to present an efficient algorithm that synchronizes the persistent reservation state via the DLM.


High Availability, scsi, clustering, HA, persistent reservations, DLM, distributed lock manager

Presentation Materials



  • Bart Van Assche



    Bart Van Assche obtained a Ph.D. in distributed computing. He works for SanDisk on the ION Accelerator software. This software provides SAN functionality that is optimized for flash memory.
    Bart is also co-maintainer of SCST, the maintainer of the SRP initiator driver in the Linux kernel and is also an active contributor to the Linux kernel SCSI initiator core.