On Sat, Jun 16, 2007 at 12:00:36PM +0200, Dawid Kuroczko wrote: > Anyway, what are your feelings with regard to using PostgreSQL > under AIX? <rant> "Don't". This is just my personal opinion, and I don't speak for my employer (especially in this case!), but my feelings are that AIX is an awful, bletcherous filthy mess that attempts to make UNIX look like a bad imitation of an AS400. Just for example, some packages come as RPMs and some as smit-installable packages, but neither packaging system knows anything about the other one, so you have the potential for all sorts of stupid errors that packaging systems were designed to _avoid_, without any of the flexibility of just compiling it yourself from source. This foolishness includes packages like, oh, gcc. Making shared libraries reminds you of the bad old a.out days on Linux; and it is distressing the IBM hasn't managed to update AIX to pre-1995 technology. That's ok, though, because some things (it seems) can't be made into shared libraries at all. So far as I know, for instance, we were never able to get PL/R working on AIX, because we just couldn't get the darn thing to compile. I seem to recall that building Perl in a way that was adequately flexible ended up taking weeks of troubleshooting and work on the part of people much smarter than I. Seneca Cunningham posted either on -general or -hackers (I forget which, but check the archives) a set of fantastically detailed contortions necessary to get some set of Postgres things working the way we wanted. AIX is designed to foil the simplest possible desires in an effort to solve some problem I haven't learned of yet. And the above assumes that everything works the way you would expect -- you know, in the way the manual says it does. I have no idea what parts of the system are actually tested by IBM's QA department before new AIX releases ship, but I can say with some certainty that parts of both libc and fsck -- fsck!! -- got overlooked in the past. At least when the Linux fanboys push on you the kernel patch of the millisecond, you know that you can look at the code or ask someone else about their experiences with it. In the case of IBM, what you get to ask is, well, IBM Support. Nice people. Excellent phone manners. Professional and worthy emails suitable for forwarding to management without even looking for scatalogical remarks. But prompt and effusive with technical detail? Not so much. The only slightly saving grace on AIX is a utility called topas, which does a nice curses-based display of various performance pieces. It makes most of the tools available on Linux seem primitive; but then, since most of the tools on Linux _are_ primitive, that shouldn't be surprising. And it's sort of awful that what you get for your license fee and inability to look at the code is one nice tool that works better than some free stuff, but doesn't work anywhere near as nicely as the RICHPse toolkit that was available starting with (IIRC) Solaris 2.5. Also, the _usual_ tools that you might be used to don't exist on AIX, so you have to learn topas, or fly blind. The large print giveth, and the small print taketh away. I loathe using AIX. I would run the other way before installing it on purpose. And given that IBM also supports both Red Hat and SuSE on POWER5 (and that for at least one non-Postgres workload we tried, Linux was actually faster on the hardware), I'd use Linux instead. > How do you feel it compares to other "big" Unixen as far as > PostgreSQL goes? As they say in alt.sysadmin.recovery, all operating systems suck. They don't usually say, however, that AIX, when pronounced as a word, is the only one that actually describes what it gives you. </rant> All best, A -- Andrew Sullivan | ajs@xxxxxxxxxxxxxxx I remember when computers were frustrating because they *did* exactly what you told them to. That actually seems sort of quaint now. --J.D. Baldwin
