Packets Received Discarded going up around 300 per minute

[AlexRac]

Hi all, doing my routine latest Exchange 2019 upgrade to the latest version, I ran HealthChecker.ps1.

One or two warning errors where : Packets Received Discarded high, and Sleepy NIC Disabled: False.

Sleepy NIC Disabled: False --- Warning: It's recommended to disable NIC power saving options
More Information: https://aka.ms/HC-NICPowerManagement
Packets Received Discarded: 1002 --- Error: This value should be at 0.We are also seeing this value being rather high so this can cause a performance impacted on a system.

So I look it up (Google, reddit, here) and there is not really anything concluding. See at the bottom for the Sleepy NIC.

So I lookup the stats:

Get-NetAdapterStatistics  | Format-List -Property "*"
...
ReceivedDiscardedPackets : 3339

Some info and tests done:

Proxmox 8.3.0, all 8 VMs are all going to the same Broadcom Adv. Dual 10GBASE-T Ethernet port 1.

Windows 2012R2 has "ReceivedDiscardedPackets : 0" with Driver Date 2020-08-10 Version 63.83.104.18900.

Windows 2022 has around 300 more packets per minutes whatever these changes in the driver advance tab:

IPv4 Checksum Offload : Rx&Tx Enabled > Disabled.
Offload.Rx.Checksum : All > Disabled.
TCP Checksum Offload IPv4 : Rx&Tx Enabled > Disabled.
TCP Checksum Offload IPv6 : Rx&Tx Enabled > Disabled.
UDP Checksum Offload IPv4 : Rx&Tx Enabled > Disabled.
UDP Checksum Offload IPv6 : Rx&Tx Enabled > Disabled.
UDP Segmentation Offload IPv4 : Enabled > Disabled.
UDP Segmentation Offload IPv6 : Enabled > Disabled.

AND whatever the activity. I tryed tracking the number while copying 5GB files at 10Gbps, idle, and copying some 5GB files at 1Gbps, and it's always the same augmentation around 300 packets per minute.

OR whatever the driver version: (can't tell if the new version has anything new since I could not find a proper release notes)

Driver Date: 02/25/2024, Driver Version: 100.94.104.24800
Driver Date: 10/21/2024, Driver Version: 100.100.104.26600

Hardware wise, there are no errors reported by the iDrac9 system on the network cards, everything is at 0 errors.

Since these errors are only with the Windows 2022 servers and not on the Windows 2012R2 server, this might be more a driver issue?

The driver does not seems to support power management, so it's not possible to change the Sleepy NIC status, but all power options in Windows server are at high power usage.

I suspect that these Packets Received Discarded are related to sometimes when I type in the RDP session some character seems to not be sent, but there are no other suspicions.

Should I continue to search or, do you guys have those Packets Received Discarded (use the powershell command Get-NetAdapterStatistics | Format-List -Property "*") ???

Have a great day!

 

[guruevi]

Are you using the virtio on a bridge or a virtual function? Discards typically happen if you have a (TCP packet) queue that can’t be processed fast enough, you’d have to see when these things occur, in my environment we instrument with Prometheus, when the packet discards happen you can then correlate what else is going on, eg does it happen when you do a migration, or a backup (do you have fleecing enabled), or just your system running out of CPU/RAM, or just your application not responding.

 

[AlexRac]

Are you using the virtio on a bridge or a virtual function?

Yes everything goes through vmbr0, but on the Proxmox host, it's around 30 packets that are dropped each minutes, and on Windows 2022 servers VMs it's more like 200 to 400 packets that are dropped, and on Windows 2012R2 VM, it's 0 packet that are dropped.

ip -s link show vmbr0 (on Proxmox host)
09h56 dropped 84095
09h57 dropped 84124
09h58 dropped 84155
09h59 dropped 84182

Get-NetAdapterStatistics | Format-List -Property "*" (On Windows 2022 server)
09h57 ReceivedDiscardedPackets : 357832
09h58 ReceivedDiscardedPackets : 358211
09h59 ReceivedDiscardedPackets : 358638


I'll probably install netdata to get a broader look.