Hi, last week Microsoft has finally released the ReFS stability fixes KB4098787 (from KB4077525) for Windows Server 2016 and Windows 10 that fixes ReFS lockup problems with Veeam repositories. I wrote about the changelog a few weeks ago. Another change is that this update will be downloaded and installed automatically from Windows Update (although you can still get the standalone package from the Microsoft Update Catalog). As you may remember, the Windows Server 2016 version of the original package has been recalled due to an unrelated quality issue. But, in the known issues section of the updated article, there’s another issue around antivirus compatibility now. Could this be the reason why some users saw no improvement after installing the original patch… Although the more likely reason is lack of RAM on a repository server, as Microsoft is still working on optimizing ReFS memory pressure, and this particular fix looks to be scheduled for the April update.
Cause:
ReFS deduplication functionality with Windows Server 2016 version 1709 was incompatible with Veeam B&R v9.5 advanced ReFS integration. Specifically, the issue is that block cloning API calls fail on deduplicated backup files with data integrity streams enabled which does look like a bug on Microsoft side. Thus, to make ReFS volumes with deduplication enabled actually usable for backup repositories, Veeam Software has decided to temporarily disable fast cloning functionality on such volumes in the Veeam Backup and Replication 9.5 Update 3. Then and after some weeks, the reason it took so long to fix is that there are multiple separate issues. It was to identify and separate all these individual issues when some were outside of ReFS even, for example, caused by NTFS-specific optimization in the OS memory management. And it also makes perfect sense now why only some of users were affected. For example, small ReFS volumes should be much less impacted by these bugs.
Microsoft, Veeam, ReFS and appliance:
In the “Last Word from Gostev“, he had talk about ReFS and about the topic of building cost-effective, all-in-one Veeam backup appliances. With his prior recommendations, he used to focus solely on the storage hardware aspect, explaining why a general purpose server with JBOD is the “way to go”. But he never really talked about the OS. According to him, Veeam’s SMB customers often complain on the forum how Windows Server license adds significant costs to a Veeam backup appliance, albeit because most don’t realize they also fully support Windows 10, which I can confirm is a great fit for purpose. But recently, there’s even better choice for higher-end backup appliances specifically, thanks to Microsoft releasing Windows 10 Pro for Workstations. Not only this edition gives you ReFS, but importantly it also supports server-grade CPUs like Xeon and in twice as many sockets (4, up from 2) – as well as three times more physical RAM (6 TB, up to 2 TB). Why does he put equal focus into the support of high-performance configurations, if we’re talking about what is mostly a backup repository? Remember that ingenious functionality Veeam provides that enables instant restore of any backup as a Hyper-V Virtual Machine. Effectively, this allows you to instantly recover a physical computer by running it directly from a backup file created with Veeam Agents (for Windows or Linux). In the next update, Veeam should support Windows 10 Hyper-V for this functionality as well effectively turning that all-in-one backup appliance into an all-in-one BCDR appliance, where this added compute capacity may actually come handy. It should offer many possibilities including ROBO environment (especially MSPs) etc.
ReFS – Microsoft releases cumulative update 4077525!
Microsoft ReFS new driver version fixed all issues!
General Windows Server 2016 licensing information, including licensing criteria and terminology can be viewed here.
Here Windows Server 2016 datasheet: Download
Veeam B&R 9.5 U3 – New protection feature here!
2 pings
[…] Continue reading » […]
[…] know, while overall ReFS stability has much improved, there’s still one major issue with the ReFS driver memory management, which causes kernel memory usage to spike on large backup file deletions, sometimes […]