From: Dr J Pelan (J.Pelan_at_gatsby.ucl.ac.uk)
Date: Fri Mar 03 2006 - 16:42:20 CET
Date: Fri, 3 Mar 2006 15:42:20 +0000 (GMT) From: Dr J Pelan <J.Pelan@gatsby.ucl.ac.uk> Message-ID: <Pine.LNX.4.61.0603031228460.25274@godel.gatsby.ucl.ac.uk> Subject: Re: [suse-sles-e] [OT] Production and Support
On Thu, 2 Mar 2006, Alexei_Roudnev wrote:
> I mean, that to be _production grade_, something must be well tested
> under REAL load. Real load means one of
> - serious stress testing and failure testing (which is pretty tricky for
> the cluster systems - for example, you must test ALL possible
> failure scenarios.
> - real load in many installation and enough time to reveal problems
> because of _they really happen_.
The rationale of your philosophy is entirely understood. I am just
pointing out that it is not a policy that can be generally advocated
because it relies on people NOT following the policy. You want other
people to do the bulk of your testing for you. Someone has to go first.
In addition, the criteria are vague - how much production usage is good
enough? - and the information you have is imperfect - you have no real
idea how many people have deployed it, the circumstances under which they
use it or indeed the failures they experience and tolerate. This is pretty
much guess work for the lower profile packages.
While knowing that other people use package X can give you that warm,
fuzzy feeling, a more objective decision can be made based on testing
yourself (call it self-certification) under loads which approach and
exceed your desired load. You have to do this anyway. Even then your
confidence can never be 100%. I have encountered serious, show-stopping
bugs in filesystem software that was deployed and heavily loaded at at
least 1K sites for 8 years! The risks are never wholly transparent even in
very widely deployed software.
Let us not forget that bugs are still being found (and introduced) in, for
example, XFS, NFS, ext2 etc. etc. They may or may not affect daily
production use. The lesson is be prepared for failure.
> With OCFSv2, we have for now:
> - good testing by Oracle for Oracle related data ONLY.
[snip]
> It means that it is _relatively_ safe for Oracle in production, safe for
> _Oracle style usage_ (example - keep shared /usr/share directory - usage
> profile is exactly the same as for ORaCLE_HOME), and is not safe for
> such things as file server with numerous concurrent access sessions (NFS
> sever for example).
[snip]
Exactly my point, there are non-Oracle uses which can be investigated.
Each deployment case must be judged on its own merits.
I have more faith in Oracle. I am sure that their in-house development
testing and bone-fide filesystem certification process will exercise a
vast amount of the code paths and scenarios essential for general use.
OCFSv2 hasn't been plucked from the air either, it is partly based on
existing code and sits on extant layers like JBD, buffer cache etc. It has
a better pedigree and design team than many other packages (they've even
discovered an i/o scheduler fault). But don't take it on faith - test it!
> It have nothing with SUPPORT - which is just MARKETING TERM, nothing more
> (but based on, correct, real testing).
[snip]
Support means different things to different people. It can range from
self-help services through to a form of insurance cover but it is
certainly more than a marketing term. You use this list - that's informal
support. You use the SuSE service packs and patches - they're not
marketing. Official vendor support can be meaningful and essential.
As support contracts are a revenue stream and providing support is a cost,
the availablity and depth of support is a reflection of the confidence
that the vendor has in the product. They are best placed to decide.
This is all getting rather philosophical and off-topic though ;-)
-- John P. --------------------------------------------------------------------- To unsubscribe, e-mail: suse-sles-e-unsubscribe@suse.com For additional commands, e-mail: suse-sles-e-help@suse.com
This archive was generated by hypermail 2.1.7 : Fri Mar 03 2006 - 16:42:54 CET