DRBD: The Definitive Guide to High Availability Block Storage and Enterprise Replication

DRBD: Key Takeaways & Cheat Sheet

What is DRBD & How Does It Work?

• Network RAID 1: DRBD is basically software that mirrors block storage (hard drives, SSDs, partitions) over a network between Linux servers so you don't need expensive SAN hardware.

• Kernel Integration: It hooks directly into the Linux kernel I/O stack right below file systems, keeping it super fast and transparent to your apps.

• App Agnostic: Since it works on raw binary blocks, DRBD doesn't care what file system or app you run on top of it.

• Simple Management: You control it using high-level CLI tools like drbdadm which talks to lower-level kernel utilities under the hood.

• Primary vs. Secondary Roles: Only the node marked Primary allows reads/writes; the Secondary node receives updates continuously but locks down direct access to prevent data corruption.

Replication Protocols & Transport Speed

• Three Replication Modes: Supports Protocol A (Asynchronous for long-distance/WAN), Protocol B (Semi-synchronous), and Protocol C (Synchronous for local clusters with zero-data-loss/RPO=0).

• RDMA Support: To skip TCP network bottlenecks and high CPU usage on fast networks (10G+), DRBD can run over RDMA for zero-copy transfers with sub-40-microsecond latencies.

• Near Bare-Metal Speeds: Benchmarks on enterprise NVMe drives show DRBD holding around 170k IOPS for 4K random writes with minimal latency penalty compared to raw hardware.

Cluster Safety & High Availability

• Built-in Quorum: Uses majority voting rules to avoid split-brain situations during network partitions. If a node loses quorum, it freezes or throws I/O errors to keep data safe.

• Diskless Tiebreakers: You can add a lightweight third node without local storage to act purely as a voting tiebreaker to maintain quorum cheaply.

• DRBD Reactor vs. Pacemaker: While Pacemaker is the old-school heavy cluster manager, DRBD Reactor is a lightweight alternative that hooks into systemd and DRBD's native events for easy setup.

Common Enterprise Use Cases

• MySQL / MariaDB HA: Sits below the database file system to mirror InnoDB redo logs synchronously, allowing instant failover and crash recovery with zero lost transactions. Outperforms Ceph in TPS and latency stability.

• Windows & SQL Server (WinDRBD): Ports the DRBD engine natively into the Windows kernel by emulating SCSI disks, bringing block replication to Microsoft SQL Server without needing expensive SANs.

• Persistent Message Queues: Offers a simpler Active/Passive setup for RabbitMQ and ActiveMQ, skipping the heavy CPU/memory overhead of Raft consensus queues.

• Kubernetes via LINSTOR & Piraeus: LINSTOR orchestrates DRBD volumes for Kubernetes storage claims via CSI. Combines fast local NVMe performance with dynamic cloud-native storage management.

Long-Distance Disaster Recovery (DRBD Proxy): Buffers, compresses, and encrypts replication data over wide area networks (WANs) so local applications don't wait on cross-region network latency.

Works cited

1. Case Studies - LINBIT, https://linbit.com/case-studies/

2. DRBD 8.4 en - LINBIT, https://linbit.com/drbd-user-guide/users-guide-drbd-8-4/

3. Distributed Replicated Block Device - DRBD, https://docs.kernel.org/7.1/admin-guide/blockdev/drbd/index.html

4. DRBD 9.0 en - LINBIT, https://linbit.com/drbd-user-guide/drbd-guide-9_0-en/

5. DRBD Deep Dive - Medium, https://medium.com/@TechInternals/drbd-deep-dive-fd1ec54130dd

6. DRBD - LINBIT, https://linbit.com/drbd/

7. An Introduction to Distributed Replicated Block Device (DRBD), https://www.opensourceforu.com/2015/12/an-introduction-to-distributed-replicated-block-device-drbd/

8. Troubleshooting DRBD Performance Issues - Knowledge Base, https://kb.linbit.com/drbd/troubleshooting-drbd-performance-issues/

9. Exploring DRBD Replication with Real World Scenarios - Medium, https://linuxgd.medium.com/exploring-drbd-replication-with-real-world-scenarios-52cab4d0d221

10. RDMA & What It Means for Data Transfer & Replication - LINBIT, https://linbit.com/blog/rdma-what-it-means-for-data-transfer-replication/

11. RDMA Performance - LINBIT, https://linbit.com/blog/rdma-performance/

12. Independent Performance Testing of DRBD by E4 - LINBIT, https://linbit.com/blog/independent-performance-testing-of-drbd-by-e4/

13. IOPS vs Throughput: What's the Difference in Storage Performance?, https://www.starwindsoftware.com/blog/iops-vs-throughput/

14. drbd.conf - DRBD Configuration Files - Ubuntu Manpages, https://manpages.ubuntu.com/manpages/noble/man5/drbd.conf-9.0.5.html

15. How to fix DRBD recovery from split brain - IPSERVERONE, https://kb.ipserverone.com/knowledge-base/how-to-fix-drbd-recovery-from-split-brain/

16. Using the DRBD Quorum Feature as an Alternative to Fencing in a, https://linbit.com/blog/using-the-drbd-quorum-feature-as-an-alternative-to-fencing-in-a-pacemaker-cluster/

17. DRBD9 and LINSTOR User's Guide | PDF - Scribd, https://www.scribd.com/document/449145247/User-Guide-DRBD-9-pdf

18. How to resolve split-brain in DRBD9 - DEV Community, https://dev.to/kvaps/how-to-solve-split-brain-in-drbd9-15no

19. DRBD recovery from split brain - Artem Sidorenko, https://www.sidorenko.io/post/2012/03/drbd-recovery-from-split-brain/

20. Detecting split-brain - IBM® Netezza, https://www.ibm.com/docs/vi/psfa/7.1.0?topic=administration-detecting-split-brain

21. How To Recover from a DRBD9 Metadata Split Brain Situation, https://www.neteye-blog.com/blog/2019/07/02/how-to-recover-from-a-drbd9-metadata-split-brain-situation/

22. Resolving split-brains | drbdtop, https://linbit.github.io/drbdtop/guides/resolving-splitbrain/

23. DRBD Reactor & Pacemaker: A Side-by-Side Comparison - LINBIT, https://linbit.com/blog/drbd-reactor-pacemaker-a-side-by-side-comparison/

24. Configuring High Availability for MySQL Databases Using DRBD, https://linbit.com/blog/configuring-high-availability-for-mysql-databases-using-drbd/

25. STAC Summit, Software Updates, & Testimonials - LINBIT, https://linbit.com/blog/stac-summit-software-updates-testimonials/

26. Windows Replication Software - WinDRBD - LINBIT, https://linbit.com/sds-for-windows/

27. MySQL now provides support for DRBD, https://dev.mysql.com/blog-archive/mysql-now-provides-support-for-drbd/

28. https://linbit.com/blog/deploy-a-highly-available-mariadb-service-using-linstor-drbd-reactor-tech-guide/

29. windrbd/TODO at stable - GitHub, https://github.com/LINBIT/windrbd/blob/stable/TODO

30. [DRBD-announce] WinDRBD 1.1.4 released - LINBIT Mailing Lists, https://lists.linbit.com/pipermail/drbd-announce/2023-January/000624.html

31. Differences Between WinDRBD and Linux DRBD - Knowledge Base, https://kb.linbit.com/windrbd/differences-between-windrbd-and-linux-drbd/

32. Becoming More Accessible to Windows Users - LINBIT, https://linbit.com/blog/becoming-more-accessible-to-windows-users/

33. Open Source High Availability Storage - LINBIT, https://linbit.com/high-availability/

34. Apache ActiveMQ High Availability Architecture Guide 2026 - meshIQ, https://www.meshiq.com/blog/activemq-high-availability-architecture/

35. ActiveMQ vs RabbitMQ: Protocols, HA, Licence, Support - AceMQ, https://acemq.com/blogs/activemq-vs-rabbitmq/

36. Part 3: RabbitMQ Best Practice for High Availability - CloudAMQP, https://www.cloudamqp.com/blog/part3-rabbitmq-best-practice-for-high-availability/

37. Using DRBD to Build a Reliable & High-performing 2-node HA, https://linbit.com/blog/using-drbd-to-build-a-reliable-high-performing-2-node-ha-solution/

38. Disaster Recovery and High Availability 101 - RabbitMQ, https://www.rabbitmq.com/blog/2020/07/07/disaster-recovery-and-high-availability-101

39. Open Source Kubernetes Persistent Volumes Storage - LINBIT, https://linbit.com/kubernetes/

40. LINBIT Technology Overview - simplyblock, https://simplyblock.io/glossary/linbit/

41. LINSTOR User Guide - LINBIT, https://linbit.com/drbd-user-guide/linstor-guide-1_0-en/

42. Kubernetes Storage Performance Comparison Rook Ceph ... - ITNEXT, https://itnext.io/kubernetes-storage-performance-comparison-rook-ceph-and-piraeus-datastore-linstor-e9bc2859a8f0

43. Piraeus Datastore, https://piraeus.io/docs/v2.11.0/

44. Piraeus Datastore - Persistent Volumes for Kubernetes - YouTube, https://www.youtube.com/watch?v=WWiBwMZHZDY

45. Storage on Talos Linux with Linstor and DRBD - Pim Widdershoven, https://www.pimwiddershoven.nl/entry/storage-on-talos-linux-with-linstor-and-drbd/

46. piraeusdatastore/piraeus: High Available Datastore for Kubernetes, https://github.com/piraeusdatastore/piraeus

47. Performance Tuning for LINSTOR Persistent Storage in Kubernetes, https://linbit.com/blog/performance-tuning-for-linstor-persistent-storage-in-kubernetes/

48. DRBD Proxy | EuroStack Directory Project, https://euro-stack.com/solutions/drbd-proxy

49. WinDRBD + Proxy: Block-Based Replication for Windows Servers, https://linbit.com/blog/windrbd-proxy-for-windows/

50. DRBD Proxy - LINBIT, https://linbit.com/drbd-proxy/

51. When to Use DRBD Proxy - Knowledge Base - LINBIT, https://kb.linbit.com/drbd/drbd-proxy/drbd-proxy-when-to-use/

52. CASE STUDY: - LINBIT, https://linbit.com/wp-content/uploads/2020/08/SSI_Schaefer_Case_Study.pdf

53. HA & DR Solutions for Educational Organizations - LINBIT, https://linbit.com/blog/educational-organizations-choose-linbit/

54. LINBIT SDS Triples Operations for RTM in the Financial Sector, https://linbit.com/wp-content/uploads/2025/11/RTM_Case_Study-2025.pdf

55. Payliance - LINBIT, https://linbit.com/case-study/payliance/

DRBD Enterprise Storage Replication Guide

Overview Architecture & I/O Replication Protocols Quorum & Split-Brain Use Cases & Benchmarks Success Stories

LINBIT Docs ↗

High Availability Block Storage

Distributed Replicated Block Device (DRBD)

DRBD is an open-source, kernel-level storage replication engine developed by LINBIT. Simplistically, DRBD functions as a network-based RAID 1 array. It mirrors block devices (NVMe drives, hard disks, or LVM volumes) across multiple servers in real time, ensuring zero data loss and transparent application failover without expensive proprietary SAN hardware.

RPO = 0

Zero Data Loss Consistency

Synchronous block replication guarantees that remote memory and disk reflect every written transaction instantly.

< 40 µs

Ultra-Low Latency via RDMA

Kernel bypass over RoCE and InfiniBand eliminates TCP software bottlenecks and context switches.

Cross-Platform

Linux & Windows (WinDRBD)

Native kernel drivers for Linux and Windows Server enable seamlessly unified cross-OS storage replication.

Kernel Architecture & I/O Flow

This section illustrates how DRBD integrates directly into the Linux kernel I/O stack underneath the file system layer. By intercepting write requests before they hit physical media, DRBD forks data concurrently into local disk storage and network transmission buffers. Use the interactive controls below to simulate real-time application writes and explore device state handling.

Interactive I/O Stack Simulator

Select a target role for Node 1 to observe how the kernel handles incoming file system I/O write calls.

Kernel Event Log:

System ready. Node 1 is currently PRIMARY. Write operations permitted.

Linux Kernel I/O Data Flow Diagram

Application Layer

MySQL / MariaDB / MS SQL / ActiveMQ

↓ VFS / Block Cache System Call

DRBD Kernel Module (`/dev/drbd0`)

Virtual Block Device Layer (Major 147)

I/O FORK ACTIVE

Local Data Path

NVMe / LVM Volume

Idle

Network Transport Path

TCP/IP or RDMA Fabric

Idle

Peer Node (`Standby Host`)

DRBD Kernel Engine → Remote Physical Disk

Replication Protocols & Distance Simulator

DRBD supports three distinct replication synchronicity levels to balance local transactional speed against wide-area disaster recovery constraints. Choose a protocol or drag the network latency slider below to understand how network delay impacts overall write performance and data integrity guarantees.

Network Latency & RPO Calculator

Inter-Site Latency (RTT): 1.0 ms (Campus/LAN)

0.1 ms (Local NVMe/RDMA) 10 ms (Metropolitan) 50 ms (Global WAN)

Recommended Protocol:

Protocol C (Synchronous)

Guarantees RPO=0. Writes are acknowledged only when stored safely on both local and remote non-volatile media. Ideal for HA clusters within a single data center or campus.

Application Latency Overhead: ~1.05 ms / write

Recovery Point Objective (RPO): Zero Data Loss (RPO=0)

Protocol	Synchronicity	Acknowledgment Point	Primary Deployment Target
Protocol A	Asynchronous	Acknowledged when written locally & placed in local TCP buffer.	Long-distance WAN & DR sites (paired with DRBD Proxy).
Protocol B	Semi-Synchronous	Acknowledged when received into remote memory buffer.	Metropolitan networks with moderate latency constraints.
Protocol C	Synchronous	Acknowledged ONLY after non-volatile remote disk flush completes.	Local HA clusters for mission-critical databases requiring zero data loss.

Cluster Resilience, Quorum & Split-Brain Engine

A "split-brain" scenario occurs when a network partition forces two independent nodes into the Primary state, leading to divergent and corrupt datasets. DRBD natively solves this through strict majority quorum voting combined with diskless tiebreakers. Simulate a network partition below to observe automated protection policies in real time.

3-Node Cluster Topology

QUORUM HEALTHY (3/3 Votes)

Node A

PRIMARY

Storage Node

Node B

SECONDARY

Storage Node

Node C

DISKLESS

Tiebreaker Node

Cluster Quorum Logic:

Total votes: 3. Majority required: 2. With all nodes connected, Node A maintains Primary state securely. Node C acts as a cost-effective diskless tiebreaker without needing local storage hardware.

Manual Split-Brain Recovery Workflow

If misconfigurations ever cause a split-brain condition, DRBD blocks traffic and enters a safe StandAlone state. Administrators resolve this by declaring an authoritative survivor and executing the discard workflow on the victim node.

# 1. On Victim Node (Discard divergent writes):

drbdadm secondary <resource>

drbdadm disconnect <resource>

drbdadm -- --discard-my-data connect <resource>

# 2. On Authoritative Survivor Node (Trigger sync):

drbdadm connect <resource>

→ DRBD executes bitmap resynchronization, restoring UpToDate status without full volume recopying.

Enterprise Use Cases & Performance Benchmarks

DRBD serves as the transactional engine behind critical database engines, Windows workloads, persistent message streams, and Kubernetes clusters. Explore the tabs below to view implementation architectures and benchmark comparisons against traditional distributed storage systems.

MySQL & MariaDB High Availability

DRBD intercepts InnoDB storage engine fsync() system calls, guaranteeing synchronously mirrored log files before acknowledging transaction commits.

• DRBD Reactor Integration: Utilizes Linux cgroups to instantly "freeze" database execution during transient network partitions, allowing sub-second recovery ("thaw") upon reconnect without cold restarts.
• High Throughput Efficiency: Delivers over 81% of non-replicated bare-metal performance, bypassing application-layer logical binlog lag.

Microsoft SQL Server via WinDRBD

WinDRBD brings DRBD’s kernel-level engine natively to Windows Server, enabling enterprise-grade high availability without costly SQL Server Enterprise licenses or proprietary hardware SANs.

• SCSI Hardware Emulation: Presents replicated volumes as standard SCSI disks to Windows Plug-and-Play Manager, granting full NTFS/ReFS compatibility.
• Dynamic Object Instantiation: Prevents split-brain by hiding raw underlying partitions from Windows Volume Manager on secondary nodes.

Cross-Platform DR Capabilities

Because WinDRBD shares core replication logic with Linux DRBD, organizations can mirror production Windows SQL Server instances directly to off-site Linux DR hosts.

[Windows SQL Server] --(WinDRBD Protocol C)--> [Linux Standby Target]

Persistent Message Queue Orchestration

Deploying DRBD block storage under persistent message brokers like RabbitMQ and Apache ActiveMQ avoids the severe CPU overhead associated with multi-node application consensus algorithms (e.g., Raft Quorum Queues).

• Single-Node Processing Speed: The primary message broker writes directly to local NVMe media with minimal consensus overhead.
• Zero Message Loss Failover: Upon node promotion, the secondary broker mounts the mirrored drive and inherits the exact binary state of all unacknowledged queues.

Broker Architecture Comparison

Native Raft Quorum Queues: High CPU/RAM overhead, multi-node network chatter, reduced message throughput.

Active/Passive with DRBD: Maximum single-node write performance, zero application consensus lag, hardware-agnostic failover.

Kubernetes & Cloud-Native Storage (LINSTOR)

Integrated into Kubernetes through the CNCF sandbox project Piraeus Datastore, LINSTOR dynamically provisions DRBD-backed volume replicas across cloud nodes.

• CSI Plugin Integration: Translates Persistent Volume Claims (PVCs) into localized DRBD block resources automatically.
• Storage Provider Flexibility: Supports Thick/Thin LVM and ZFS pools.

Enterprise Success Stories & Reference Links

Read how global enterprises leverage DRBD to cut infrastructure costs, replace proprietary SAN arrays, and achieve uncompromising uptime. Click any case study reference card to view detailed documentation.

Intralogistics Read Case Study ↗

SSI Schaefer

Replaced proprietary SAN hardware with DRBD warm-standby clusters to automate rolling software patches across 24/7 warehouse logistics centers without risking downtime or data loss.

FinTech Cloud Read Case Study ↗

RTM (Financial Network, Brazil)

Migrated central payment network infrastructure (RSFN/PIX) to LINSTOR and DRBD over NVMe media, achieving a 4x IOPS improvement, 50% latency reduction, and 80% cost savings per terabyte.

AWS Payment Gateway Read Case Study ↗

Payliance

Deployed synchronous cross-Availability Zone DRBD replication across AWS EC2 instances, bypassing escalating EBS IOPS costs while guaranteeing strict financial regulatory compliance.

Higher Education Read Case Study ↗

Educational Assessment Testing

Backed mission-critical university online exam platforms with MySQL on DRBD clusters, guaranteeing 99.999% uptime and zero exam record loss during hardware failures.

DRBD Technical Explorer — Interactive Guide based on official LINBIT research documentation.

DRBD User Guide ↗ DRBD Proxy Docs ↗ Piraeus CNCF ↗