Troubleshooting a Network Issue

[mikeyosm]

Hello

 

I'm running UNRAID 6.4.1 with a Windows 10 VM and Nvidia 1050ti Passthrough with an uptime of 16 days so far which is great. I am however experiencing an odd network issue that manifests itself every so often maybe 1 or 2 times a day. The UNRAID host is set to 9000 MTU and my Windows 10 VM is using Jumbo Frames with the virtual NIC negotiating at 10Gbps. Generally, network performance is great and I'm constantly reach the max transfer rate of my SSDs when moving data between the Windows 10 VM and UNRAID network shares. So what's the prob?

 

Well, completely at random, the Windows 10 VM network connection will suddenly change to 'limited connectivity'. As soon as It does this, I putty to the UNRAID host and ping the gateway address of the router and this fails. Obviously, I can't ping any external DNS or IP addresses from the VM either, however, internal IP's on the 192.168.x.x subnet resolve OK. It's almost like the UNRAID host is losing the gateway address completely so I can't ping the router or resolve external addresses.

 

I am using a homeplug 1Gbps solution that uses my electricity cables to carry the network connection around the house, but I don't think this is the issue as I have tried numerious vendors and the issue exists regardless of which one I use. Now, the issue could be the router right? Wrong, when the issue starts on the UNRAID host/W10 VM, any devices connected via WIFI still maintain Internet connectivity so the router is still up and servicing DNS requests etc with no issues. I must also add, that the issue resolves itself after 1 or 2 minutes with no intervention from myself, but to resolve it quicker, I unplug/replug the network cable from the UNRAID host and hey presto the network is back.

 

I want to narrow down the issue if I can to maybe the host/VM and/or the homeplug adaptor, so can anyone provide some tips on what extra logging I can enable to show me a complete diagnostics of where the network is failing?

 

Thank you for your help.

[John_M]

If you use Jumbo frames on one device on a network segment you have to use it on all devices on that segment. Are you doing that? I can't work out from your description the exact topology of your network but in my experience Jumbo frames are more trouble than they are worth outside of a small, closed environment, because their use breaks backwards compatibility. If you set everything to an MTU of 1500 does it all work?

[mikeyosm]

22 hours ago, John_M said:

If you use Jumbo frames on one device on a network segment you have to use it on all devices on that segment. Are you doing that? I can't work out from your description the exact topology of your network but in my experience Jumbo frames are more trouble than they are worth outside of a small, closed environment, because their use breaks backwards compatibility. If you set everything to an MTU of 1500 does it all work?

 

Jumbo frames MTU 9000 is set on the UNRAID host Interface (eth0) and on all Windows VMs (NIC connection rate set to 10G and Jumbo frames enabled).

The Homeplug does not have jumbo frames enabled (it only support up to 1Gbps). I am not using any other network switches.

 

I use jumbo frames so I can transfer files over the network between VMs at 200+ MB/s as I'm using fast SSDs and an NVME drive. I have not set everything to 1500 MTU for that reason and I cant see why I need to sacrifice network performance for stability when most of the time the network is fine except for this very random occasions as I explained.

[John_M]

If the Homeplug doesn't support Jumbo frames but your unRAID eth0 is using them you have a problem.

 

23 hours ago, John_M said:

If you set everything to an MTU of 1500 does it all work?

 

[bonienl]

Using jumbo frames isn't unfortunately a simple plug-n-play set up. To make jumbo frames work all devices in the chain, such as switches, routers and other clients must support jumbo frames too (note that jumbo frame can mean different sizes to different equipment, officially any MTU larger than 1500 bytes is considered a jumbo frame).

 

MTU sizes are negotiated whenever a TCP session is setup with a remote client, in general this works fine but firewalls are notorious to break this negotation scheme. UDP sessions don't use negotation, and will fail if both endpoints or the underlying network doesn't support jumbo frames.

 

In short, yes you can use jumbo frames but you need to know exactly what you are doing. Read: network know-how.

Is it beneficial? That depends, most modern NIC cards have hardware off-loading and no issues to fill-up a 1G link using the standard MTU size. Moving to 10G links may give some improvement with jumbo frames.