Re: Adjusting to DB2

From: Serge Rielau <srielau_at_ca.ibm.com>
Date: Thu, 27 Oct 2005 17:59:49 -0400
Message-ID: <3sd0v5Fnltg6U1@individual.net>

Mark Townsend wrote:
> The porting teams at Oracle do indeed do their own QA. They also do
> their own install docs. If you look at the number of products supported
> on the number of platforms, it works out to about 2.8 heads per product
> per platform. Not really alot. I'm sure the DB2 LUW team has more than
> 2.8 Sun or HP engineers involved in their ports.
I think what we are finding out here is an impedance mismatch in organization and lingo that makes any comparison attempty quite futile. When I say it takes a few weeks to port DB2 that is a port to a new, never before supported platform (like e.g. MAC OS). Once a port is completed human resource consumption is rather limited. Platform exploitation (as in supporting MS latest, startfromscratchyetagain-APIs or exploitation of new Linux features) in my lingo are not part of porting. That's just regular devlopment (done by the right set of people).
Don't want to prolong the debate her, just adding pieces to the puzzle.

> The reasons for the lag are very simple. We run massive regression tests
> before we release (and also during development), on huge grids of
> machines (some 1400 CPUs or more).
>
> The hardware used for these tests are centralized and shared by all the
> server tech product teams - so that's database, app server, grid
> control, collab suite etc. So in DB2 land, that would be like all the
> DB2 dev groups sharing the servers with the WepSphere group and the
> Tivoli group as well.

OK, got that. Doesn't work that way in SWG. Resource sharing is across Releases (e.g. fixpack QA has to be interleaved)

>> A product that uses good encapsulation does not require the same 
>> amount of QA across all combinations of platforms because only low 
>> level function can diverge. 

> We would disagree. 27 years of supporting every platform under the sun
> (no pun intended) has taught us that this is a very, very dangerous
> assumption to make. Even the compilers on different platforms can do
> vastly different things with the same code line. And not to pour too
> much gas on the fire, but I think Oracle development has a great deal
> more experience in supporting a single code line on multiple platforms
> than the DB2 LUW developers do.

Well, decisions and experiences are local. As Mark A. quite correctly pointed out DB2 has a good track record w.r.t. OS specific bugs. In my personal experience running any SQL Compiler QA against very few selected (i.e picky) platforms provides near 100% confidence. Ironically Windows is a very good candidate (drops dead at the slightest cold). AIX a rather bad one (too lenient). This data is readily available through root-cause analysis of bugs found internally and externally.

Next to all of my problems with platforms in QA have been QA tooling specific. (In DB2 developers co-own QA areas (such as SQL features) with the QA team).
I never had to work on an APAR that was platform specific. I found a couple of c-compiler bugs, all exposed during initial porting to a new platform.

It is up to anyones guess why things are the way they are, but statistical data for DB2 does support that DB2's bugs are non platform specific to a point that QA can affort to differenciate. I doubt that it requires 25 years for such data to become significant if it were present.

Cheers
Serge

-- 
Serge Rielau
DB2 SQL Compiler Development
IBM Toronto Lab

Received on Thu Oct 27 2005 - 16:59:49 CDT