Server 2025 - high CPU and RAM

[Cortec IT Dan]

For some reason I cannot reply with quotes.

pybfr - I will start considering it now, cheers.
benm2024 - Yeah that's the only reason I haven't done this already. Can I ask, what size volumes and how much RAM you have? Also, you still have certain regkeys set?

Guess I will try to wait for others feedback but this update has made it worse. Server hangs, you reboot and now all RAM is used immediately, rather than building up after restore point deletion.

[m.novelli]

My customers are using Dell Servers (various models like T360 / R550 / R7515) with ReFS and so far they didn't hit the issue

Volume size in the 40 - 100 TB range with good amount of RAM. I perform a full backup every weekend, not forever incremental, single backup job for every customer, generally 10 - 20 VM per customer

I'm talking about 5 Servers with Windows Server 2025

Marco

[Cortec IT Dan]

Marco - what is a "good amount of RAM", how often do you reckon restore points are being removed and how many at a time?

[m.novelli]

I'm rounding the rule of minimum 1 GB RAM per TB of ReFS: 32 GB RAM for customer with 42 TB usable (8 disks 8 TB RAID6) , 64 GB RAM for 76 TB usable (12 disk 8 TB RAID6)

The restore points are deleted every night, tipically just one restore point since I'm using a single backup job, for about 10 - 20 VM at time. Not really high intensive like others people here

Marco

[jokoenen]

Marco - What is the maximum single files size of the deleted backup job files?
Jo

[m.novelli]

2 TB maximum per single file

I have some customer with 3+ TB single files but their Servers are still on Windows Server 2022

Marco

[jokoenen]

Hi!
I tried it with 8 backup files each with a size between 3-4 TB and it worked quite well. Took around 25 minutes until finished and left 90 GB reserved for System. RAM usage around 17GB of 32GB total. CPU usage between 65% and 100% fluctuating but system allways responsive.
Jo

[Cortec IT Dan]

Jo,

Just to confirm, you deleted at minimum 24TB of data? and you have 32GB of RAM, what size refs volume?
Thank you

[jokoenen]

Volumes are 18TB in size each. Total deleted file size 32TB.

[Cortec IT Dan]

Jo/Marco/Ben
Any regs keys still in place in addition to the update? Tested again and this is deffo not fixed for me. Don't really understand why not.
Any upgrades of old OS to new or all new 2025 clean installs?

[m.novelli]

My Servers are a clean install, no reg key hack

Marco

[jokoenen]

Just checked: the RefsEnableLargeWorkingSetTrim is still set to 1. No other keys regarding the ReFS issues on our server.

[HenrikS.]

Installed December 2025 update last Wednesday.
Removed the registry overrides test, as they are not documented and I don't know exactly what they do.

Had another BSOD last night, ReFS bugs seems to still be there..

[Martin@SHH]

Ich habe es jetzt so gelöst, was die wöchentlichen full backups betrifft und indirekt dann auch die inkrementellen:
1. Regkey für das Disablen von Fast Cloning gesetzt: HKLM\Software\veeam\veeam backup and replication\RefsVirtualSyntheticDisabled als REG_DWORD = 1
2. Umstellung Backupjobs lokale VM's von synthetic full auf active full, da lokale NAS Performance-Engpass, es entfällt das Lesen auf der NAS, da die VM's frisch gesichert werden ohne bestehende Daten zu verwenden.
3. Remote backups externer Einrichtungen auf zentrale NAS mit schmaler DSL-Leitung weiterhin als synthetic full, da Engpass DSL-Leitung aber jetzt ohne Fast Cloning. Laufzeit bleibt gleich, da die Leitung der limitierende Faktor ist.

Ergebnis:
1. Ja, mein Active Full-Backup dauert jetzt statt 0,5h nun 3,25h, aber:
2. Mein Tape-Job auf LTO dauert jetzt nicht mehr 8,5h sondern nur 6,5h, also 2h gewonnen, Schreibrate bei LTO von 183MB/s auf 243MB/s gestiegen!
3. Keine Probleme mehr mit den cleanuptasks die cpu und RAM zulaufen lassen.

Für mich eine tragbare Lösung, da sie bei uns passt. Klar, die Vorzüge von ReFS sind damit praktisch obsolet, dafür endlich wieder ein stabiler Betrieb und schnelle LTO-Backups.

[RubinCompServ]

Have you considered using the Veeam apliance instead ?

I've considered it, but the inability to attach iSCSI storage is a deal-breaker.

[BenM2024]

Jo/Marco/Ben
Any regs keys still in place in addition to the update? Tested again and this is deffo not fixed for me. Don't really understand why not.
Any upgrades of old OS to new or all new 2025 clean installs?

Sorry to everyone wanting a reply... Microsoft Licensing has decided to throw a paddy after many years of just working.

Yup - on my 2025 server the ReFS reg keys and the software overrides from up-thread are still in place. If it works don't fix it! Of course that makes it difficult for anyone else to replicate the system.

Storage/RAM - I have 5Tb of Live data, Backued up to 1 NTFS iSCSI Volume and Backup Copy jobs sending that data to 4 iSCSI provisioned ReFS volumes of varying size, and retention periods - biggest is 17Tb. The server has 96Gb RAM - when the ReFS bug was at its height the server would take about 30 minutes to eat all that and crash. Sometimes it lasted a bit longer, sometimes it wopuld last only 5 minutes. I did bung some ram in from another machine for a while so it had 144Gb and that was swallowed up as well so I gave the ram back to where it came from!

Server OS - was 2016, upgraded to 2022, then, fatally, to 2025; no spare disk storage around to copy the data off and reformat to NTFS so I had to weather the storm!

After the software override reg settings things went back to being completely stable and working as they did back with server 2022. That was 6 weeks ago (or was it 5?); I have since upgraded Veeam B&R/VeeamONE to v13 with no real issues (finger trouble, if anything) and installed the December Patch Tuesday patch. Everything remains stable.

[jokoenen]

Bad news: today I let Veeam backup copy job delete the obsolete files and the server was not responsive anymore after approx. 1 minute.
Last time I did the deletion in advance using MS Explorer and had no problems.
Jo

[HenrikS.]

Agree, I restored the override reg keys after last BSOD/Reboot.
The server crashed again last night.. So far my experience with the December 2025 is that it is worse than November 2025 patch.

[JaySt]

i gave up. Reinstalled with server 2022.

[Cortec IT Dan]

So I also gave up and installed 2022, same issue!
PLEASE, someone tell me if I am missing something or doing something stupid.

[JaySt]

ok that's strange (and interesting). So you see exactly same behavior?

[s.middlebrook]

i gave up. Reinstalled with server 2022.

So I also gave up and installed 2022, same issue!

This is my task next week, Server 2022 to be reinstalled...

I've been battling with getting Postgres SQL repaired, as each BSD that goes into meltdown with the SQL Server Backups, and low and behold once that's sorted another BSD, and everything back to where it was.

Lost the will with Server 2025 in this type of environment.

[Gostev]

So I also gave up and installed 2022, same issue!
PLEASE, someone tell me if I am missing something or doing something stupid.

Then it can only be inflicted by 3rd party software or be some kind of hardware problem, as Server 2022 never had this issue.
Discussed in this thread is the net new issue caused by changes in ReFS stack that were made in Server 2025.

[karsten123]

is it please possible that veeam will write a kb, where everything is documented?
what is going to break for sure with Windows Server 2025 and for which infrastructure it is save to use.

[Gostev]

Discussed in this thread is the only known issue with Server 2025 to my knowledge and there's a KB article about it already (previously linked in this thread). If you feel something is missing from this KB article, it's best to provide feedback directly from the article so the responsible team receives and implements it.

[karsten123]

of course you are right, Anton. it is 2792.

STAY AWAY FROM REFS!!!

[Didi7]

Unbelievable that this thread is already more than one year old and people still struggle with Windows Server 2025.

[BenM2024]

>> people still struggle with Windows Server 2025.

To be fair, its not every 2025 installation using ReFS storage that has the problem. IIRC Veeam have one specific VM/Guest/Role that shows the issue - not a whole raft of them.

I have just got back after leaving the Backups running over the holidays - everything ok.

In terms of what happened over the hoidays - incrementals, synthetic fulls, retention related deletion jobs; all completed without a hitch. So for me at least, the issue is still acknowledged and worked around as no-one seems to have come up with a definitive root cause (other than "something" Microsoft did in ReFS 2025 broke deleting big files).

[Didi7]

To be fair, I use Windows Server 2025 with ReFS and VBR only in very small environments. I planned to use Windows Server 2025 in our big infrastructure environment as well, but suddenly ...

VSA

... was released and I couldn't say NO

[Cortec IT Dan]

Then it can only be inflicted by 3rd party software or be some kind of hardware problem, as Server 2022 never had this issue.
Discussed in this thread is the net new issue caused by changes in ReFS stack that were made in Server 2025.

I get what you are saying but its the same 3rd party software (screen connect) and hardware as when I was running server 2019 and had no issues.
Maybe its a different cause with the same symptoms... but it freezes when deleting restore points.