| |||||||||||||
Enterprise distributions suck and free software rulesEnterprise distributions suck and free software rulesPosted Mar 8, 2011 2:06 UTC (Tue) by airlied (subscriber, #9104)In reply to: Enterprise distributions suck and free software rules by kragilkragil Parent article: Enterprise distributions and free software
Have you ever used Linus kernel in a production env, then upgraded it 2 years later?
good luck.
The sheer number of regressions esp in areas like the VM is quite large, the kernel needs stabilisation periods that upstream doesn't supply, and shoving into Debian stable doesn't do anything other than sticking people with a kernel that isn't going to improve or add features.
The featureset of no major regressions + new HW support is something people think is worth paying for, community distros don't cater to this market at all.
(Log in to post comments)
Enterprise distributions suck and free software rules Posted Mar 8, 2011 2:48 UTC (Tue) by dlang (✭ supporter ✭, #313) [Link]
yes I have, I have been doing so for 14 years now.
I can point at the availability of the systems that I manage and compare it to the availability of the systems running RHEL and make a very good case that my linus kernel on whitebox systems is better than their RHEL on top-tier branded systems
if you blindly install a RHEL upgrade/kernel on your system without testing it, you will have some pretty horrible problems. If you are doing the testing, then it doesn't really matter if you are testing RHEL or linus-kernel based systems (although, if you run into something strange, you can get debugging and fixes for the linus-kernel faster than I've seen happen with RHEL)
Enterprise distributions suck and free software rules Posted Mar 8, 2011 4:18 UTC (Tue) by raven667 (subscriber, #5198) [Link]
I have to say I really haven't had this problem, even when running triple digit numbers of systems I can't think of the last time i've had an RHEL update go bad, kernel or otherwise.
Enterprise distributions suck and free software rules Posted Mar 8, 2011 13:20 UTC (Tue) by SEJeff (subscriber, #51588) [Link]
RHEL 4.5 was a disaster of an upgrade:
https://bugzilla.redhat.com/show_bug.cgi?id=327591
Redhat took all of the upstream nfs client "fixes" that google primarily wrote and backported them to RHEL4. It did not go so well. The best thing I got out of working on this bug was ninja-level nfs debugging experience. See comment 8 for the secret sauce.
Enterprise distributions suck and free software rules Posted Mar 8, 2011 18:21 UTC (Tue) by ricwheeler (subscriber, #4980) [Link]
NFS client patches that google wrote?
Are you confusing google (no interest in NFS) with NetApp which employs Trond, the upstream client maintainer?
Enterprise distributions suck and free software rules Posted Mar 8, 2011 20:11 UTC (Tue) by vonbrand (subscriber, #4458) [Link] If you compare the stability of vanilla Linus kernels on whiteboxen to $ENTERPRISE kernels on top-tier machines, you don't know if the differences are due to the kernel's stability, broken(ish) support for funky hardware on the "top" machines, different workloads/stress, or even simply due to different definitions of "stable" in different environments.
Enterprise distributions suck and free software rules Posted Mar 12, 2011 21:00 UTC (Sat) by dlang (✭ supporter ✭, #313) [Link]
true, although according the 'prevailing wisdom' both selecting top tier vendors and going with enterprise kernels are choices that are supposed to increase reliability
that being said, I used the kernel.org kernel on the top tier boxes for a few years before deciding that they didn't provide any additional reliability.
as for the argument that reliability means different things, in the case I am referring to, reliability is measured and defined by the same people.
Enterprise distributions suck and free software rules Posted Mar 8, 2011 13:02 UTC (Tue) by kragilkragil (guest, #72832) [Link]
If nobody would do the Franken-kernel BS then all vendors would have to test with Linus RCs and everybody would be even happier. The enterprise kernel story is weak anyway you skin it and wastes so many hours of precious kernel devs lifes.
It needs to die and I hope it will sooner than later and I hope that the next few quarters RHs numbers won't be shiny and investors will punish them for wasting their money on inefficient development.
Enterprise distributions suck and free software rules Posted Mar 8, 2011 20:26 UTC (Tue) by airlied (subscriber, #9104) [Link]
you look like you have a business plan in you, or don't really understand.
The QA cycle is longer than the kernel release cycle. QAing a kernel requires a set amount of time, at least for people to be happy that it isn't considerably worse than the previous one. Some of the benchmarks that enterprise customers care about can nearly take longer than the test cycle to get scheduled, they require hw that isn't always available. Never mind tracking down regressions using a bisect on a test that takes a week to run and process the results of.
Enterprise distributions suck and free software rules Posted Mar 8, 2011 20:43 UTC (Tue) by lmb (subscriber, #39048) [Link]
QA cycles can be shortened. That is "merely" an engineering problem that can be solved: automation, parallelization, smarter tests, and better test plans. Not to mention better code review. There's a whole stack of books on continuous delivery & deployment out there in your library.
And even if the test cases were only run for, say, for every three upstream releases, I postulate that this would a) greatly reduce the chance that relevant regressions get introduced, b) even rev'ing the "enterprise" kernels every three upstream kernel releases would already be a huge boost over rev'ing them every 3 years.
I'd never have expected that, of all people, the _Linux_ folks would be the ones to claim that Linux/OSS can't work in an mission-critical environment but needs to be curtailed to a legacy "enterprise" model! I'm truly amazed.
Enterprise distributions suck and free software rules Posted Mar 9, 2011 4:22 UTC (Wed) by airlied (subscriber, #9104) [Link]
I think you are missing a large point, if you have competent admins, and I stress admins, (having one super-hero admin is a really bad business decision and you deserve your whitebox+Linus solution to fail hard when your admin has some life altering event), then you can totally deploy Linus tree into mission critical places. Google, facebook etc are prime examples of this, even though google might be a few kernels back they are getting closer to mainline. However if you have management or admins who like to spend time with their kids/families then you have to have some sort of support place you can call and someone you can blame. Now the company providing that service cannot provide bespoke Linus kernels every 3 months, it just isn't practical.
Enterprise distros also have a whole bunch of certifications (government, application) etc that it isn't feasible to redo every 3-6 mths it can takes a year or so to get some of them finished.
Enterprise distributions suck and free software rules Posted Mar 9, 2011 4:27 UTC (Wed) by rahulsundaram (subscriber, #21946) [Link]
"QA cycles can be shortened. That is "merely" an engineering problem that can be solved: automation, parallelization, smarter tests, and better test plans. Not to mention better code review. There's a whole stack of books on continuous delivery & deployment out there in your library."
If you can make this model work, you have a brilliant edge over the competition. Feel free to try.
"I'd never have expected that, of all people, the _Linux_ folks would be the ones to claim that Linux/OSS can't work in an mission-critical environment but needs to be curtailed to a legacy "enterprise" model! I'm truly amazed."
Linux can certainly work in a mission critical environment. The debate is not about that at all but whether the current enterprise model is legacy or necessary. I would say that it is possible to tweak the model and vendors occasionally do that but it is not going to go away unless some vendor decides to provide a sustainable alternative.
Enterprise distributions suck and free software rules Posted Mar 10, 2011 3:56 UTC (Thu) by BenHutchings (subscriber, #37955) [Link] The featureset of no major regressions + new HW support is something people think is worth paying for, community distros don't cater to this market at all. Actually, Debian does add hardware support, particularly for hardware that may be required during installation. However, we aren't nearly as thorough as Red Hat in doing this, mostly because we don't have the resources for regression-testing driver changes. I have been meaning to propose some sort of hardware partner program for Debian kernel and X maintainers to work with hardware vendors (or upstream maintainers with good collections of hardware) on testing driver updates. But I think it would have to be somewhat different from commercial partner programs.
|
|||||||||||||