kernel-packages team mailing list archive

[Scott Moser <smoser@xxxxxxxxxx>]

So after some discussion, it seems that what we have is not entirely unreasonable.
There would/could be some gain (possbly on the order of 70-80M in /lib/modules).  That includes removal of wireless modules, graphics drivers, sound, and some other misc drivers.  I'll attach my revised list.

So there is definitely some space to gain from a -server image, however that comes at a development and recurring maintenance cost.
Below the 2 images are "cloud" and "maas", but can be considered "virtual server" and "physical server" for all practical purposes.

The options then are:
a.) keep things as they are, having 2 images for download.  Essentially a "ubuntu server + -virtual kernel" for virtual environments and a "ubuntu server + -generic kernel" for hardware.
b.) add new '-server' flavor,  and reduce those 2 images to 1.  This incurs cost of additional size for virtual users.
c.) add existing -generic flavor everywhere.

It seems that we have 2 different things for good reason.  The drivers
that are unnecessary in a virtual environment total to be quite large.
I think the best thing to do at the omemnt is keep the status-quo (a)
and reduce other differences to the 2 images.

We will reduce the delta so that the 2 are really the same other than
"hardware enablement" packages (kernel , grub/uefi ... ) and document
that cleanly.

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1415634

Title:
  RFC: replace linux-virtual with linux-server / tune kernel  packages

Status in linux package in Ubuntu:
  Confirmed

Bug description:
  We currently build 2 meta packages for easy consumption and automatic
  kernel upgrades.  These are linux-virtual and linux-generic.

  The -virtual meta package was created with the general purpose of
  "support virtual environments", and then extended to include "and
  common cloud workloads" (adding things like rdb and kvm).

  Would it be possible to find a happy medium where we had enough
  drivers to enable network and block devices and essential server
  devices but maintained a reasonable install size ?  That would enable
  us to create one set of images for use in cloud enviroments be they on
  bare metal or hypervisors.

  As an example of sizes collected from a cloud image, we have:

  release | kernel        | apt inst| initrd | /lib/modules
   vivid  | linux-virtual | 125M    |  8M    |  34M
   vivid  | linux-generic | 358M    | 27M    | 193M
   trusty | linux-virtual | 119M    |  7M    |  31M
   trusty | linux-generic | 337M    | 24M    | 184M

  'apt inst' is as reported by apt install after 'apt-get --purge ^linux-.*'
  vivid version 3.18.0.11, trusty version 3.13.0.44.51.
  both packages share the linux kernel binary which in this case is 5.6M on trusty and 6.3M on vivid.

  For a cloud image with default install in the 700M range, the
  difference between -virtual and -generic is considerable.  So it
  clearly has its value.

  However, this value comes at the cost of specialization.  If a user
  installs a server via MAAS or via ISO, they get linux-generic and have
  some set of modules/kernel function.  If they run a cloud image, they
  have a different set.  This is less than desireable as we'd like to
  say that both cases are "Ubuntu Server".  It also means that we have
  to build "maas images" which are primarily "cloud images with a
  hardware kernel".

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1415634/+subscriptions