Re: EC2 image boot time improvements
Hi Tim,
On 9 February 2016 at 15:29, Tim Sattarov <stimur@gmail.com> wrote:
> Hello everyone,
>
> I was comparing boot time of Amazon Linux and official Debian images.
> And here what I've found:
> 1) Default kernel has 30 second delay for Xen framebuffer initialization
I was hit by that yesterday:
admin@ip-172-31-36-245:~$ sudo systemd-analyze
Startup finished in 37.499s (kernel) + 1min 5.007s (userspace) = 1min 42.507s
In dmesg:
[ 5.944108] xenbus_probe_frontend: Waiting for devices to
initialise: 25s...20s...15s...10s...5s...0s...
[ 30.850956] xenbus_probe_frontend: Timeout connecting to device:
device/vfb/0 (local state 3, remote state 1)
[ 30.858357] xenbus_probe_frontend: Device with no driver: device/vbd/51712
[ 30.863125] xenbus_probe_frontend: Device with no driver: device/vif/0
> 2) Networking service has another 30 seconds delay I addressed here:
> https://github.com/andsens/bootstrap-vz/pull/263
I do remember when you sent this patch. Thank you.
> Currently I use custom AMIs built with patched bootstrap-vz and using my
> own kernel, stripped down to only what is needed for EC2 and it has much
> better results. But I would love to see those changes in the official
> images as well.
>
> But for the kernel: I believe there should be separate flavour for ec2
> platform = new package and feature request to Debian Kernel Team.
>
> Any thoughts ?
Ubuntu used to have an EC2 flavor of their kernel, but looks like they
aren't using it anymore since Maverick (10.10)[1]. So I'm not sure if
we should do this nearly six years later.
This Xen framebuffer initialization delay can also be observed on
Oracle Compute Cloud and probably on any Xen HVM installation. As it
looks like you solved this, maybe we can integrated this change to the
default kernel, instead of packaging a new stripped down version for
an specific provider?
> Since I'm quite new to debian-cloud, what would be the best way to do it ?
> I will open bug for bootstrap-vz to include that fix or may be there is
> better way to fix it.
>
>On 9 February 2016 at 15:34, Tim Sattarov <stimur@gmail.com> wrote:
> Heh, I've just noticed this:
> https://github.com/andsens/bootstrap-vz/releases/tag/v0.9.9-squeeze
> So I guess first question can be closed ? :)
Your pull request was merged back then and AFAICT, the need for a new
tag/release had no impact whatsoever, because James uses the upstream
code (possible with some changes on his fork, which are used to be
merged later). What happened is that he hadn't published a new EC2
image after it was merged.
"v0.9.9-squeeze" was tagged because we wanted to make clear that this
is the last version which Squeeze is supported, as you can see on
#280[2].
[1]: https://wiki.ubuntu.com/KernalTeam/EC2Kernel#Kernel_Release_Notes
[2]: https://github.com/andsens/bootstrap-vz/issues/280
--
Tiago "Myhro" Ilieve
Blog: https://blog.myhro.info/
GitHub: https://github.com/myhro
LinkedIn: https://br.linkedin.com/in/myhro
Montes Claros - MG, Brasil
Reply to: