Target is a Windows Server 2019 or 2022, the problem is always the same, the file copy over 10BaseT is going down and up and is not steady. Source is a Windows 10 machine, no AD is used. One NIC has an aquantia chip, the other one is an intel. The other way around there is no problem and also there is no bottleneck anywhere. Anyone got some idea?
Try this from a cmd as Administrator netsh interface ipv4 set global taskoffload=disabled netsh interface ipv6 set global taskoffload=disabled Reboot. To revert, replace "disabled" with "enabled" A different approach to obtain the same results is to go into the advanced settings for the NIC and disable LSO but this comes under different names depending on the NIC, there is a v2 which if exists is the one in use, there are flavours for IPv4 and IPv6 as well, all should be configured if you follow the hardware path. I am recommending the software registry (command line) approach which covers other offloading technologies but do not have significant impact in small implementation setups.
The speed is reasonable for a virtualised machine. Do you run Citrix without AD? What hpyervisor solution do you use? VMware Workstation, ESXi?
@ch100 This is all bare metal, from and to nvme, I just used a vmdk-file from my "workstation" because of its size.
Hard to say then. Check network switch speed and duplex setting assuming NIC is configured on default settings. There is also RSC - Receive Segment Coalescing State which you might try to disable. netsh interface tcp set global rsc=disabled It is trial and error, normally the settings which I provided you in the first post would make a difference in certain configurations. If those do not change your speed, then it is difficult to pinpoint the exact issue.
@ch100 For sure. I am not the only one having this issue but finding the cause is hard. Someone else fixed the issue for them by not using Windows but some arch version I think as the SMB host. RSC didn't help either but now the max speed is less then before. Yeah...
Hi dear @ch100 First of all, I hope you, all our colleagues of MDL and all your friends and family are well and healthy now and always... Is it possible to use this command to speed up any communication within a normal gigabit windows network? If yes, is there any way to implement this through the windows registry when it is using a setupcomplete routine on installation? Thanks in advanced. Regards @JeepWillys58
Cache of what? That is the target drive (local): Interesting, it has less headroom then expected, the same nvme in my system was faster I think... The target is completely free, nothing stored on it.
yeah,scratch that,always thought nvme depended on system ram for cache. I couldn't duplicate the issue with a larger than my memory file transfer.
For the ipv4 command the equivalent in registry is Code: Windows Registry Editor Version 5.00 [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters] "DisableTaskOffload"=dword:00000001 For ipv6 I don't know if there is a reliable way to configure it in registry, but if you do your own research you might come with your own solution. Please be aware that out of the box ipv6 is the primary protocol and if you only configure ipv4, the configuration might not apply if it is not done for ipv6 as well. There are a lot of variables and ipv4 and ipv6 can be prioritised further, but if you are looking for a simple solution and don't understand all dependencies, you are better off by leaving the default configuration and only change the settings which I provided if they resolve an issue. The configuration can be done in Group Policy as startup script but I don't want to be dragged into a discussion about an enterprise setup as this is not the purpose of MDL.
Because you mentioned SMB, you could try to enable/disable SMB1 and see if this resolves the issue. If anyone else reconfigured oplocks you might find some tuning available there. Normally oplocks should be enabled if they don't cause any issues and this is default. Sorry I don't have the settings readily available but you can find the relevant information in Microsoft documentation.
Windows always use RAM as read-cache and most likely as write-cache but this depends on the configuration of the drive in Device Manager. For most purposes and safety of data, the default settings are optimal. The DRAM on-board of hard-disks and/or RAID controllers is used independently of Windows controlled by the firmware although Windows can enable/disable it if the configuration is made available by the firmware. I don't think caching has anything to do with the issue described by OP as it is common to many configurations and it is not a fault as such.
I set up both machines in a homeserver style setup, changed nothing other than the cache-size in the NICs. SMB1 I don't touch either and not want to anymore. Both sides are an up to date Windows anyways. And I am not the only one with that problem. Maybe cabling is not optimal, maybe two brands of NIC isn't. Maybe it is just a setting in WinServer, but I don't think that anymore. Thank you for your suggestions. I think if not someone else had the same problem and had solved it, there is not much to be done left.