Massive packet loss/NIC issues with new Dell 7070 Ultra in FOG



  • Hi.

    Our company recently purchased a few Dell 7070 Ultra to start preparing our environment for a change to this PC/Setup in our production environment.

    So far i’ve successfully captured and deployed a test image from 1 of my 2 test PC’s of this model, my problem is that the 2:nd PC of the exact same model & batch suddenly after loading the bzimage/kernel file (throughout the rest of the imaging process) has massive packet losses/response times, ranging from ~500-5000+ms, with a lot of dropped packages all together. Resulting in imaging taking a weekend instead of ~5 min. Imaging works correctly also with our legacy hardware running undionly.kkpxe/BIOS

    Once in Windows/UEFI/anywhere else than FOG the PC has standard response times and everything works perfectly, showing that the NIC seems to work fine…

    The problem starts before any imaging/capturing begins - as soon as the kernel is loaded, pointing probably to a driver issue, what confuses me is why the first PC works like a charm every time in that case…

    I’ve manually upgraded the kernel to Kernel.TomElliott.4.19.64.64 (from included .48 kernel) - no difference.

    At this stage i’d like to try further PC’s from this model, but it will take a couple of months before that’s possible.

    So my question is, do you have anything else to point me in a direction to troubleshoot further, or is there a newer kernel/drivers that might simply work better? It is a brand new model and even a completely new series from Dell after all…

    I’ve tried changing ports & cables between the PC’s that work and doesn’t work, it’s always this specific PC that doesn’t work with any combination of cables etc… I’ve had one imaging that suddenly 99% of the process seemed to work and i managed to deploy the image to the PC that time, but that’s once in about 50+ tries, randomly during imaging it might start working with ~1ms for 5-10 sec and then it stops working again, sometimes (maybe 10% of the time) if i pull the ethernet cable out for a couple of seconds and put it back in it works for the first 5-10 seconds as well… Really feels like a driver issue.

    Do you think it’s a new/other kernel version that should solve this, or a newer FOG version altogether or something else?

    FOG 1.5.7 stable, ARM (FOG test environment on Raspberry Pi 4, 4gB, Raspbian Buster, latest updates as of last week)

    Imaging/capturing; Dell 7070 Ultra, i5 8365U, 8gB RAM, UEFI, with default ipxe.efi & .48 x64 & .64 x64 FOG/Tom kernels

    NIC; Intel I219-LM


    UPDATE; new BIOS released 21st October, 1.1.2. Updated to this version, same issue remains on this PC. Fresh win 10 install from microsoft - network works perfect, Debian Buster, FOG 1.5.7 etc same issue, works perfect still on 1st PC.

    Today (22:nd October) i got information from our branch office that they’ve managed to run FOG with our help on the 2 Dell 7070 ultras there and they work fine as well! Very strange that this single PC would behave like this and only in Linux env.


    Any help greatly appreciated, thanks!
    Best regards,
    Robin, IT Specialist. (With client PC environment responsibility among other things)



  • @Quazz said in Massive packet loss/NIC issues with new Dell 7070 Ultra in FOG:

    https://bugzilla.redhat.com/show_bug.cgi?id=1652865

    Some other people having a similar issue, problem possibly fixed on newer kernel version.

    @no0NE Go to the Kernel Update page and try grabbing the Kernel 5.1.16

    Thanks for the info!
    I’m quite sure it’s related to the kernels, i’ve tried ubuntu 19.10 kernel 5.3.0-18 and it works fine there… (what made me confused and starting this thread was due to the same PC’s - 1 to start with and now 3 tested all worked fine with same firmwares, OS’s, network/cables etc…!)

    I forgot to tell i also tried the TomElliot 5.1.16 (“mac/nvme fix”) kernel with the same results during my kernel troubleshooting this monday.

    So i can’t DOA the PC to Dell, but i’ll try to call their support to ask them what may differ between the physical PC’s settings wise, hardware, firmwares etc.


  • Moderator

    https://bugzilla.redhat.com/show_bug.cgi?id=1652865

    Some other people having a similar issue, problem possibly fixed on newer kernel version.

    @no0NE Go to the Kernel Update page and try grabbing the Kernel 5.1.16


  • Developer

    @no0NE are there any power saving features available on the nic under windows that are enabled? sometimes those get stuck in windows and the linux side doesn’t have the ability to re-enable the performance mode.



  • UPDATE; new BIOS released 21st October, 1.1.2. Updated to this version, same issue remains on this PC.

    • Fresh win 10 install from microsoft - network works perfect. Also, all Ubuntu version i’ve tried, latest 19.10 works fine (kernel 5.3.0-18)

    • Debian Buster, PXE FOG 1.5.7 etc same issue, works perfect still on 1st PC.

    Today (22:nd October) i got information from our branch office that they’ve managed to run FOG with our help on the 2 Dell 7070 ultras there and they work fine as well! Very strange that this single PC would behave like this and only in Linux env. If it weren’t for it working fine in Windows i’d DOA it, but i’ll reach out to Dell for any help on what this may be caused by.



  • @george1421

    Thanks for the feedback.
    We switched to Dell from another brand, so we don’t have any adapters laying around, but i have a Dell DA200 USB C dongle with ethernet at home, good idea to try that, i’ll bring it tomorrow just to test! :)

    I quickly tried some version of Ubuntu live ( i think 18.04) on it and it worked that time last week. But i downloaded Debian Buster now and tried as well just to make sure as much as possible, it’s the same problem there now! Only on one of the PC’s, Debian Buster on “the first” PC of the same model continues to work fine with the same cables & BIOS here as well… Thanks for pushing me to actually double check that again!

    I got to DOA submit this PC to Dell soon, but unsure how they’ll see this as it’s working in Windows, but i’ll hope they have enough goodwill with us being a new Dell customer with a big order going in.
    Before doing that, i’m reinstalling windows 10 manually now from USB / Win 10 media creator just to make sure that works still or now.

    Btw, we began our environment change focused on the 3060, then the 3070 came and we were preparing purchasing of that model, but we slowly trickled into optimizing our physical environment with the 7070 Ultra instead and landed on this model to purchase now. It’s quite different from 3060/3070 since it’s built on laptop parts to begin with. It’s still quite expandable and configurable, but no full size PCI-e etc ;)


    UPDATE

    After talking with a Dell tech who helped me brainstorm a few things and we doublechecked the revision of the NIC, which is the same on both/all machines (Rev 11 of the I219-LM Intel ethernet NIC), we didn’t get any wiser, so basically still at the same spot.

    What i’m concluding is that it seems to work on newest Kernels even on this troubling PC, so i’m leaning to have to build my own kernel and it should sort itself out (what i’d probably had done by now, if it weren’t for all the other PC’s already working…)



  • Moderator

    @no0NE I’d still recommend trying a different cable/switch port just to rule it out completely, even if the first device seems to work properly on it.

    I also found the following on this topic that’s worth checking out: https://wiki.hetzner.de/index.php/Low_performance_with_Intel_i218/i219_NIC/en


  • Developer

    @no0NE So that leads us to the suggestion George made earlier. Boot some kind of Linux Live OS and see if it causes the same problems on this hardware (but not on the other one).



  • @Sebastian-Roth said in Massive packet loss/NIC issues with new Dell 7070 Ultra in FOG:

    @no0NE Have you had the first device (working fine) on the same switch port and with the same cable that you now see the problem with the second “faulty” device?

    Absolutely sure the firmware is the exact same version on both?

    Yes & Yes, sadly. Other than that, i’ve tried several other straight & crossover cables to be fully sure. I’m running out of ideas to troubleshoot, hence this post… :) Good suggestions if i hadn’t already checked, thanks! :)


  • Moderator

    @no0NE said in Massive packet loss/NIC issues with new Dell 7070 Ultra in FOG:

    The Dell 7070 Ultra is a desktop

    Ok its similar to the 3060s we use then, just the 7000 series. (off topic, but dell has kind of made a mess with series numbers now its hard to tell what is what).

    So do you have a Dell usb c dongle from a 7400 or 3530 you could test with? It needs to be a Dell usb C dongle (ethernet adapter) so its supported by the firmware for pxe booting. I guess you could use a WD15 or WD19 dock. The idea is to try to use another network adapter than the built in one.

    I can tell you the Intel I219 network adapter has been supported since the 4.12.x linux kernel.

    Something else to try is to boot a linux live OS and see if you still have networking issues. FOS Linux runs on a current linux kernel so if its supported by linux then FOS Linux should be able to communicate with it.


  • Developer

    @no0NE Have you had the first device (working fine) on the same switch port and with the same cable that you now see the problem with the second “faulty” device?

    Absolutely sure the firmware is the exact same version on both?



  • @george1421
    Thanks for the reply!

    As you say, the issue is not on the server side, but rather on the client.

    The Dell 7070 Ultra is a desktop built with laptop components (google it, it’s quite cool ;) ). It’s powered through USB-C from our monitors (65w), together with DP & USB hub functionality, but the ethernet is connected separetly directly to the ethernet jack of the PC (Intel I219-LM network card).

    I ping and scan the traffic going to the PC from my admin PC, the FOG server, the other PC of the same model, can start the ping in windows and all’s fine, but as soon as it boots into FOG the packet losses starts to occur.

    To isolate as much problem sources as possible, i’ve also connected ethernet directly between FOG server and this PC and the packet losses is still the same - massive!
    I’ve also tried with dumb & smart/L3 switches and results are the same…

    Thanks for trying to help brainstorm how to find the error source since i might’ve missed something, but from these points i think i’ve eliminated your possible fault points.


  • Moderator

    @no0NE My personal idea is that its a target system firmware issue.

    Since you are talking about a 7070, is that a portable or a desktop?

    I really don’t think its the FOG server at this point, but instead something on the endpoint side.

    If this device is a portable are you using a usb c to ethernet dongle?

    Where are you seeing the packet loss?

    If you have these devices connected to an enterprise managed switch, are you seeing errors on the port counters?

    If this is a managed enterprise switch see if the green ethernet (802.3az) is enabled on the switch port. If it is disable it to see if it resolves the issue.



  • Just to add; most of the times trying to start the imaging process on “PC #2” it will fail after trying to check in for a while with an error stating “no route to host” or similar connection related errors (due to all the packet losses…) with an automated reboot after 1 min. sometimes though it slowly gets to the PartClone stage and then is stuck there at ~0.5-1mB/s with continued massive packet loss.

    I’d like to try it out on further PC’s of this model, it feels like a NIC/Hardware issue if it’s only this PC, but the first PC works repeatedly every time (20+ times). And it works without issues in Windows/UEFI…

    The packet loss against this PC is the same when pinged from any PC on the network, incl. from the FOG server itself.

    Latest available UEFI/BIOS for this model (v 1.0.2)


Log in to reply
 

414
Online

6.3k
Users

13.7k
Topics

129.3k
Posts