[Lf_carrier] CGL and the Linux Foundation

[John Cherry]

On Tue, 2007-05-15 at 06:43 +0300, mika.kukkonen at nsn.com wrote:
> >-----Original Message-----
> >From: lf_carrier-bounces at lists.linux-foundation.org 
> >[mailto:lf_carrier-bounces at lists.linux-foundation.org] On 
> >Behalf Of ext John Cherry
> (...)
> >I'm not sure if these are NTT specific, but TIPC was rejected by the
> >kernel community for quite some time.  Other projects that have not
> >gotten traction in the kernel community include live patching and boot
> >image fallback.
> >
> >TIPC was critical for high availability IPC.  It is actually the
> >backbone of a couple of clustering solutions as well.
> 
> Umm, John fails to explicitly mention here, that _TIPC_ now is in
> the Linux kernel, and has been a while. So it is one of the few
> successes actually.

I was actually mentioning TIPC as an example of a capability that was
not immediately accepted into the kernel, and may have never been
accepted into the kernel if it had not been for CGL.  TIPC was accepted
in large part because of the efforts and perseverance of a good friend
of mine....Mika!
 
> 
> >Live patching is a servicability/availability requirement makes it
> >possible to patch a binary while the system is running.  This is
> >important for online updates with zero downtime.
> 
> I do not think these will ever get mainlined to the kernel, as they
> are so intrusive and of little value to the "normal" (PC) user. But
> I have seen, that some distro vendors are pretty OK in maintaining
> these in their trees, so I do not think this is so bad situation.

Agreed, but this is an example of why CGL is important.  Live patching
is a capability that is maintained outside of the mainstream that is
absolutely needed for network equipment environments.  The CGL live
patching requirements do not specify a specific implementation of live
patching.

> 
> >Boot image fallback provides a way to quickly and gracefully back out
> >changes.  There are a plethora of issues with this and coming up with a
> >solution that is acceptable to all has been difficult.
> 
> Kexec() has been mainlined, what else you need? :-)

Good one.  Yes, kexec has been used for everything from boot image
fallback to a crash dumper.  When we looked at boot image fallback,
there were all kinds of issues with application state...like rolling
forward database transactions, etc.  Got really messy...really fast.

Cheers,
John