@sebastian-roth I have no idea too… Changing hardware is not an option, too many PCs… Currently I would keep using fog 1.3.x and kernel 3.17.3, so far it could meet my requirement. Maybe I would try to merge the initrd later. Anyway thank you so much for your support!!!

@sebastian-roth still failed at the same position…

@sebastian-roth The one extracted from ISO

@sebastian-roth I tried the 1st option, but the PC hung after the language selection screen showed up, both keyboard & power button were frozen. Didn’t see any error message, so I don’t know if it is realted to HID driver or not, but anyway I cannot do the testing, should be driver missed or conflicted.

@sebastian-roth Oops, I missed the result. I tested the kernel & initrd from ISO in two booting ways - PXE and USB. The results were quiet similar, 36 passed + 14 failed & 34 passed + 16 failed.

@sebastian-roth Yes, 3.17.3 is not perfect. I did a quick testing and found starting from 4.0 kernel (fog kernel + fog initrd + PXE) the chance of this issue raised to around 50%. I tested the installed Ubuntu 16.04 system 60 times, no any failure. I also retested the latest CentOS 7 which used 3.10 kernal and systemd, but the failure rate was around 50% too.

As you said, we already did many compiling and testing, it’s fainful :-D, and got no significant changes. This should not be the right way. The recent testing result strengthened my gut feeling that this is a system level issue. Ubuntu team found and resolved this issue somehow, but CentOS didn’t. Kernel, driver, systemd and some other unknown parts involved in this issue. So I think the best solution should be converting the installed Ubuntu 16 into PXE boot in some way. What’s your suggestion?

@sebastian-roth Sorry, I was out again, the whole week. I retested 3.17.3+PXE 60 times, 2 failed and 58 passed.

@sebastian-roth It seems here is a misunderstanding, the official Ubuntu 16.04 kernel I tested was not from installation DVD, but from an installed system.

I tested the kernel+initrd of installation DVD, it’s similar to kernels built by you, say about 50% failed.

The installed kernel is not same as installation DVD. The installed kernel + initrd-ubuntu.xz has about 75% successful rate in 20 times reboot.

I tried to PXE boot the installed kernel (/boot/vmlinuz-4.4.0-62-generic) and initrd (/boot/initrd.img-4.4.0-62-generic), but the rootfs failed to mount. I uploaded screen capture screen-capture-rootfs-failed.jpg https://drive.google.com/open?id=0Bx_soHaLoSYETXhEeUVBRVllNVE

It seems the handler of reboot/shutdown is different in your initrd and Ubuntu initrd. The Ubuntu reboot/shutdown is handled by systemctl. Now I wonder if my reboot/shutdown issue is a kind of system-level issue, both kernel/drivers/utilities have their roles.

How can I use the initrd of the installed Ubuntu in PXE? I think this may have a good chance to resolve my issue.

@sebastian-roth log uploaded https://drive.google.com/open?id=0Bx_soHaLoSYETXhEeUVBRVllNVE dmesg-bzImage-4.4.0-62-generic-with-new-initrd.log

I did 50 times reboot/shutdown each. Here is the result.

3.17.3 kernel - PXE boot    46 passed, 4 failed.
bzImage-4.4.0-62-generic and latest initrd - PXE boot          27 passed, 23 failed.
Ubuntu-16.04 - CD boot      50 passed, 0 failed.

Seems 3.17.3 also has a very little chance of failure, but acceptable for me.

Is it possible to use Ubuntu 16.04 kernel and generate an initrd from it? It’s the best one.

@sebastian-roth The new initrd had a little improvement, I did 20 times reboot/shutdown, 70% passed. I’m not sure if it is just a result of probability.

The irqpoll has no any effect, and caused a hung during the boot process before network initializing messages. The passed ratio was still around 70% in 10 times reboot/shutdown.