Home
| Calendar
| Mail Lists
| List Archives
| Desktop SIG
| Hardware Hacking SIG
Wiki | Flickr | PicasaWeb | Video | Maps & Directions | Installfests | Keysignings Linux Cafe | Meeting Notes | Linux Links | Bling | About BLU |
On Thursday 03 February 2005 13:07, Gordon Marx wrote: > On Thu, 03 Feb 2005 13:02:49 -0500, Steve Seremeth > > <blu_discuss at seremeth.com> wrote: > > And it's _critical_ that QA run the same bits as what's installed in > > your production environment on the same hardware whenever possible. > > Even if it means you're stuck on AIX version I don't even know > what....thanks FAA. I fully agree with Steve, and I'll reiterate with a war story. On a contract, I had a Digital Alpha at home for development. The lab at work had a PC running Solaris 7. The deployed system in Indiana was a Sparc. I had ported a database system and had implemented locking for supporting about 200 simultaneous users. My 64 bit version worked fine, and kudos to the original developer in producing software that was nearly 64-bit clean. Only 1 line of code had to be changed for 64-bit. The QA team did the appropriate QA. We then deployed it to the Sparc (the day I reported to work elsewhere). It promptly crashed. The reason was that in the b-tree indexes, the key-value was text and caused the pointer to not be on a 4-byte boundary. Solaris on the PC does not care about unaligned access. Linux and Tru64 on the Alpha do automatic fixups (this is an option that can be turned off). Solaris on Sparc always traps. Luckily my sponsor at Raytheon was sick, and they sent me home since no one else could escort me. I dialed in to the spark, and using ex, I changed all the code to memcpy(3) the long index value from the key table to an aligned long. I strongly recommended that the company acquire the identical hardware/software for QA, and they took my advice. (Note that I did not write the key lookup routine. It was originally written by Al Stevens). Had we properly QA'd the code, I could have re-engineered the key-building software to force the index value to properly align. I've seen other problems where the QA and deployment systems were similar hardware/software, but may have differed in configuration, and it was this difference in configuration. A well known fast food chain missed a payroll because of this. -- Jerry Feldman <gaf at blu.org> Boston Linux and Unix user group http://www.blu.org PGP key id:C5061EA9 PGP Key fingerprint:053C 73EC 3AC1 5C44 3E14 9245 FB00 3ED5 C506 1EA9
BLU is a member of BostonUserGroups | |
We also thank MIT for the use of their facilities. |