On Thu, Sep 14, 2017 at 8:31 AM, Alexander Pyhalov <[email protected]> wrote:

> On 09/13/17 11:01 PM, Peter Tribble wrote:
>
>>
>> Which begs the question - is systems innovation done and dusted?
>>
>
> Hi.
> I speak from user perspective, 'the things I want to see in illumos'.
>
> 1) Simple zone/system sharing. I just want to create an image and share
> it, so that it could be run on bare metal or in zone.


So, like the Universal Archives (or whatever it's called) that Solaris has.

I've been looking at using the aci format (from Linux container space)
to do this.


> And I hope that it will be able to run on any more-or-less up-to-date
> illumos distribution. And it would be good to do something with to make GZ
> and NGZ more loosely coupled, so that I shouldn't update them at the same
> times (running fresh zones on older kernels).


Tribblix has this, partly. I have alien zones (run a different distro in a
zone)
and images like docker (from my mvi project).


> Docker had us in this area.
>

They have a slightly eaiser time of it, because their stability boundary
is the kernel syscall layer, rather than libc. (Even so, it isn't actually
true
that you can take a docker image and move it to a different system and
guarantee it behaves identically. It'll work most of the time, though.)

Perhaps it's time we considered stabilising syscalls or looking at libc
evolution a bit differently.


> 2) Zone/VM live migration. Don't tell me "it's too hard, use clusters". I
> want to have an illumos-based cluster for virtualization purposes,
> VMware/Linux KVM/even OpenVZ do it, why can't we?
>

I'm not sure you could do that easily for zones. Although this is actually
tied
in with suspend/resume (of a process), and potentially migrating the
suspended
process to a different node or restarting it after a reboot. which used to
work but
I haven't seen for a while.

Although I see a need for higher-level tooling to manage a distributed set
of
zones (as opposed to moving zones about).


> 3) RAIDZ reconfiguration. I want to enlarge pool without introducing new
> logical vdev.
>
> 4) ZFS shrinking. I've provisioned too much space for this FS on my
> storage and want to shrink volume. What should I do?
>

Yes. But not our problem, as OpenZFS is a separate project. I suspect
they'll
get around to this in time. I've heard various suggestions to avoid the full
BP rewrite problem.


> 5) Be usable as OpenStack node (at least compute node).
>

I started to get OpenStack working once, but it's an enormous pile of
different projects and I couldn't justify the effort as I couldn't see a
situation in which I would actually use it personally.


> 6) Customizable FMA agents. I want to have generic API for writing custom
> agents, which would, for example, allow me to write an agent to send me a
> SMS when DBMS service has suddenly gone to maintenance (in ~ 10 lines of
> shell code).
>

That's a good example of a small project that could give huge benefits. The
whole
SMF/FMA notification piece needs a good rework. (And implementing SMF
monitor methods while we're at it).


> As I understand, 1/2/5 is already implemented (perhaps, partially) in
> Solaris 11.
> --
> Best regards,
> Alexander Pyhalov,
> system administrator of Southern Federal University IT department
>



-- 
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/

------------------------------------------
illumos-discuss
Archives: 
https://illumos.topicbox.com/groups/discuss/discussions/T83f198c8597cf8e3-M94aec48fdae0b1f2806a5809
Powered by Topicbox: https://topicbox.com

Reply via email to