Le 01/08/2012 16:51, Weedy a écrit :
> On 01/08/12 04:57 AM, Emmanuel Deloget wrote:
>> Hello, 
>>
>> I understand that a lot of effort has been pushed in making 
>> Linux 3.3 the trunk kernel, and I understand that I probably 
>> missed long (IRC?) discussions on this very subject, but since 
>> 3.3.8 is going to be the last supported kernel in the 3.3.x 
>> branch it might be a good idea to move on to another, more 
>> recent kernel version - and to do it slightly better. Not 
>> that anything is really bad, but there were obviously better 
>> choices that 3.3 at the time it came up. 
> 
> 3.3 was probably chosen based on wifi driver quality at the time. I
> think we started with 3.3.2 or .3 so it was pretty new at the time.
> Because of that we (most probably) can't drop down to 3.0.

I agree that 3.0 lacked a lot of changes made to support 
wifi and other interesting features. This makes my remarks about
the existence of better choices a bit weird (but then, note that
3.4 was well in the way when 3.3.2 and .3 were released ;
according to the tag log on kernel.org, 3.3.2 and 3.4-rc3 were
released the same week ; that doesn't make 3.4 a better choice, 
since it has yet to be released). 

Even if 3.4 was not option, and that (in fact, and contrary
to my precendent assertions) 3.3 was a good pick, it would have 
been a good thing to switch to 3.4 as soon as it hit the 
schelves. I agree and I fully understand that changing a kernel
is not a trivial task - it's indeed one of the trickier and 
probably as risky as changing the compiler or the runtime. I
understand that there are many, many things to test and that
tests take a lot of time and energy - and time is not free. 
But IMHO, the migration task should have started as soon as
possible. Waiting for kernel patches to accumulate just makes
things harder and harder. 

> Trying to sync with a -rt kernel does seem like a good idea,

Yep. I may have some ideas on how to put things together
in order to import -rt support in OpenWRT in a sensible way, 
but this cannot be tested if there is no support for a
-rt compatible kernel in the first place. 

> -longterm seems unnecessary.

With products that are out for 4 to 6 years and that target
a large, consumer market (I'm talking about millions of units), 
knowing that the selected kernel will be community-supported 
for at least 2 years (and maybe more if another maintainer
takes the lead) is a huge, huge incentive to invest time and
money in this specific kernel. That's the raison d'être of
-longterm after all :)

Anyway, the lack of support for at least one -longterm 
(the latest) is still an annoying issue. It makes captilizing
on these kernels a problematic issue (and such capitalization
is of utter most importance for those who sell routers for
a living). OpenWRT is used in professionnal products, and some
of these products have a huge install base that needs to be
updated from time to time.

> For the record my production networks runs trunk (something around
> 31xxx, I forget right now). It's a VPN mesh consisting of over 100 nodes
> and minutes of downtime is measured in thousands of dollars. Ignoring
> the whole "REAL MEN USE TRUNK(tm)" the tagged releases are just too old
> to do anything fun with.

Doing fun things and making things work are orthogonal 
notions :) Doing fun things (I guess you are speaking of
taking advantages of new features) may even be counter
productive (ISP tends to prefer having a working router 
than one which is able to do fun things :))

<-->

Le 01/08/2012 20:40, Hartmut Knaack a écrit :> Hi,
> my impression is, that a kernel version makes it into trunk
> if it is either a long term kernel, or it brings essential
> new functions. For 3.3 this was most certainly the introduction
> of BQL code. Keeping in mind that our main targets are network
> routers, the bufferbloat issue probably concerns most of the
> maintainers.

This, along with the information provided by Weedy in his 
answer, recoup an informal discussion I had with colleagues 
of mine shortly after I hit the "send" button.

My orginal mail contains a bit of noise and some misinformed
statements, but the idea was never to derail the choices that
were made a few month ago. Its goal was more to discuss the
choice of *future* kernels, how they will be selected, and
objective criterion might be used to select them. I fully 
apology if the meaning of my fist mail was not clear enough 
and I hope that this short paragraph clarifies my discourse. 

Best regards, 

-- Emmanuel Deloget

_______________________________________________
openwrt-devel mailing list
[email protected]
https://lists.openwrt.org/mailman/listinfo/openwrt-devel

Reply via email to