ReFS. Known issues

Challenge

Backup and Backup Copy synthetic operations like retention (merge) or synthetic full creation on ReFS based repositories could lead to following issues:

1. Memory exhaustion on repository servers with ReFS
2. Synthetic operations take considerable amount of time
3. Repository server becomes unresponsive or significantly slow

All of the above could lead to unexpected/faulty backup/backup copy jobs termination.

Cause

As described in Microsoft article KB4016173 the root cause resides in existing design of ReFS metadata management.

Solution

Based on experience of multiple installations and tests performed by our RnD and Solutions Architects teams recommendations below proved to be most efficient in resolving problems described in “Challenge” section:

Operating System
 

Use the following OS versions to avoid any known ReFS issues described in “Challenge” section:

• Windows Server 2019 (not officially supported until Veeam Backup and Replication 9.5 Update 4. Lab use only)
• Windows Server 2016 patched to at least September 2018 updates (KB4343884 or later)
• Windows 10 Pro for Workstations

Note: refs.sys driver file should be 10.0.14393.2457 or later

Hardware
 

Backup repository should meet minimum system requirements. In particular:

• CPU: at least 1 core per each concurrent backup proxy task and at least 1 core for each two concurrent repository tasks recommended.
• RAM: Backup repository server running other backup roles should have added amount of RAM for each role.
• Storage: 64K ReFS volume formatting is recommended

3rd Party Software
 

Uninstall the following 3rd party software that have been reported to cause ReFS stability and/or performance issues:

• Antivirus software (except Windows Defender)
• Microsoft Configuration Manager

Note: general recommendation is to run crystal clear image of Windows installed from original ISO or MSDN resource to ensure no bloatware is installed.

More Information

We continue to closely collaborate with Microsoft ReFS team on improving stability and performance of ReFS. To monitor current state please refer to our forum discussion.