[mythtv-users] Fedora vs. CentOS - Thoughts

[Jarod Wilson]

Meant to reply to this a while ago. Better late than never...

On Nov 24, 2010, at 12:19 PM, Greg Oliver wrote:

> On Wed, Nov 24, 2010 at 10:52 AM, Jarod Wilson <jarod at wilsonet.com> wrote:
>> On Nov 24, 2010, at 1:17 AM, Greg Oliver wrote:
>> 
>>> On Tue, Nov 23, 2010 at 3:39 PM, Jarod Wilson <jarod at wilsonet.com> wrote:
...
>>> 
>>> Hi Jarod,
>>> 
>>> Since it looks like RH is not gonna distribute their patchlist with
>>> their kernel sources any longer, the benefit of having their kernel is
>>> gone.
>> 
>> How so, exactly? (Not that I want to be argumentative, I'd
>> actually really like honest feedback about the issues this
>> causes for folks).
> 
> Well, anyone (or any distro) can assemble a userspace with a vanilla
> kernel, but redhat's best feature (IMO) is their kernel.  They have
> the largest base of coders contributing to it,

Absolutely correct. :)

> and maintain some private patches as well.

Not many. The vast majority of what goes into the Red Hat kernels is upstream
or is headed upstream. Red Hat tries to keep the number of NOT upstream bits
to an absolute minimum.

> I *trust* their kernels for their stability.
> Their config tools may be pretty good too, but we never use them, so
> I do not really know.  What I do know is that we use Sun ACTA
> platforms exclusively in the department I work in, and since RH and
> Oracle like to argue over a few things we end up building kernels for
> particular hardware mixes every once in a while since we cannot wait
> for resolution.

The most common situation I see is that new hardware foo has just made
it to market, and RHEL won't support it until the next update release,
which may be as much as 5 months out, but its rare that happens anymore,
as hardware vendors actually think about Linux in advance these days,
and make sure their hardware works before its released. (For example,
when just a bit ago everyone was talking excitedly about sandy bridge
hardware being released, I'd had a sandy bridge box in my office for a
number of months already...)

> We do extensively use the cluster suite as well, so I guess you have
> to take my opinion with a grain of salt, because I guess without RH we
> would not have that either  :)

Haven't ever used it myself. Is it good or bad? :)

> While it is few and far between when we have to rebuild a kernel
> nowadays maybe it is not so bad.  Except for poor little ole me at
> home who wants to run a *patched* by redhat kernel because I trust
> them to keep my data safe due to their track record and extensive
> customer base of testing.  Trusting someone else to round up all of
> the patches (and I know not all of them are pushed to Linus' tree) is
> going to fall short with maybe the exception of Suse, and I really
> just do not like their whole facade they throw on it - for a console
> only server, I am sure it is just fine).

Its been a while since I've looked at a SuSE kernel (SLES, particularly),
but last I knew, Novell was playing it a LOT looser with patches than
Red Hat, adding all sorts of stuff to their kernel that never had much of
a chance of going upstream... I guess when you're struggling for market
share, you'll do anything customers ask for... :)

> Just my $.02  - there are a lot of days I curse RH (and
> Cisco/F5/Sun/Oracle  :)   ), but at the end of the day, I would have
> it no other way..  I do see their point though - I guess a main
> benefit to this is that the cloners have more of a job to do now - it
> should definitely weed out the weak ones  :)

The cloners who are purely cloning don't really have it any tougher. If
they're simply rebuilding the bits unmodified, there's no change. Now,
it should theoretically be a bit harder for those providing support on
top of their cloned bits, as they won't have as good an idea exactly
what changes have made it into the kernel and what haven't.

-- 
Jarod Wilson
jarod at wilsonet.com