Обсуждение: Re: [HACKERS] PostgreSQL - the Linux of Databases...
Marc wrote... > On Tue, 3 Mar 1998, Thomas G. Lockhart wrote: > > > > > It's OK Marc, us linux'ists weren't offended _too_ much by that quote :)) > > I'm such a trouble maker, but I find most Linux'ers such easy easy > prey *grin* I have this University full of Linux'ers I wonder why there are SO MANY Linux'ers? :-) > that you can spark > up just with a comment like "Linux != Unix"...which, it isn't, its a > Unix-like clone...but they can't seem to figure the distinction *rofl* Agreed... :-) But BSD isn't Unix either - not officially. [Waits for Marc to disagree, again...] Not to mention the fact that at least one release of Linux did go through full Posix certification and is thus allowed to be called Unix :-) Andrew ---------------------------------------------------------------------------- Dr. Andrew C.R. Martin University College London EMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk URL: http://www.biochem.ucl.ac.uk/~martin Tel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775
Thus spake Andrew Martin
> > up just with a comment like "Linux != Unix"...which, it isn't, its a
> > Unix-like clone...but they can't seem to figure the distinction *rofl*
>
> Agreed... :-) But BSD isn't Unix either - not officially. [Waits for
> Marc to disagree, again...]
Of course it is. It has direct lineage back the Bell Labs. There is
no AT&T code left in but you can most definitely say "BSD Unix" where
you can't say "Linux Unix." For many years Berkeley was the main
development hotbed for Unix. In fact, BSD was eventually fed back
into SVR4.
> Not to mention the fact that at least one release of Linux did go through
> full Posix certification and is thus allowed to be called Unix :-)
Posix != Unix. NT is a Posix system. So is OpenVMS.
BTW, which version of Linux was Posix certified and who paid for it?
--
D'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves
http://www.druid.net/darcy/ | and a sheep voting on
+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.
On Tue, 3 Mar 1998, Andrew Martin wrote:
> Marc wrote...
> > On Tue, 3 Mar 1998, Thomas G. Lockhart wrote:
> >
> > >
> > > It's OK Marc, us linux'ists weren't offended _too_ much by that quote :))
> >
> > I'm such a trouble maker, but I find most Linux'ers such easy easy
> > prey *grin* I have this University full of Linux'ers
> I wonder why there are SO MANY Linux'ers? :-)
Actually, I don't...Linux had a much quicker start into the Free
market...the *BSD crowd had to content with the almost year(?) of legal
deliberations as to whether or not they were even allowed to distribute
and work on it :( Linux had no such problems, since Linux had no
history...no roots :)
> Agreed... :-) But BSD isn't Unix either - not officially. [Waits for
> Marc to disagree, again...]
I believing the only "official" Unix is the one produced by the
company that this year has decided it wants to own the name, isn't it? :)
> Not to mention the fact that at least one release of Linux did go through
> full Posix certification and is thus allowed to be called Unix :-)
Actually, my understanding is that its allowed to be called a
Posix-compliant Operating System... :)
On Tue, 3 Mar 1998, D'Arcy J.M. Cain wrote:
> Thus spake Andrew Martin
> > > up just with a comment like "Linux != Unix"...which, it isn't, its a
> > > Unix-like clone...but they can't seem to figure the distinction *rofl*
> >
> > Agreed... :-) But BSD isn't Unix either - not officially. [Waits for
> > Marc to disagree, again...]
>
> Of course it is. It has direct lineage back the Bell Labs. There is
> no AT&T code left in but you can most definitely say "BSD Unix" where
> you can't say "Linux Unix." For many years Berkeley was the main
> development hotbed for Unix. In fact, BSD was eventually fed back
> into SVR4.
What he said *scrambles to save this for next time*
> > Not to mention the fact that at least one release of Linux did go through
> > full Posix certification and is thus allowed to be called Unix :-)
>
> Posix != Unix. NT is a Posix system. So is OpenVMS.
>
> BTW, which version of Linux was Posix certified and who paid for it?
Ummmm, I don't know the version, but I do know that this was the
case...whether they stayed Posix certified or not is another story, but I
do remember this...
On Tue, 3 Mar 1998, The Hermit Hacker wrote: > What he said *scrambles to save this for next time* I can not belive this thing... :-) Hey, guys, you had a tough time doing 6.3, right ? Now that you all said your necessary rant on the Linux vs. Others thing, please calm down b4 I join the thread :-) (oops, I think I just did) Cristian -- ---------------------------------------------------------------------- Cristian Gafton -- gafton@redhat.com -- Red Hat Software, Inc. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ UNIX is user friendly. It's just selective about who its friends are.
On Tue, 3 Mar 1998, Cristian Gafton wrote:
> On Tue, 3 Mar 1998, The Hermit Hacker wrote:
>
> > What he said *scrambles to save this for next time*
>
> I can not belive this thing... :-) Hey, guys, you had a tough time doing
> 6.3, right ?
>
> Now that you all said your necessary rant on the Linux vs. Others thing,
> please calm down b4 I join the thread :-) (oops, I think I just did)
You joined much much too late though...this has been going on
since, oh, day one :) And, most ppl involved in the rant know me and my
opinions (they aren't necessarily the same as what I use as my bait, of
course, but ya gotta admit, Linux'ers are just soooooooo easy to bait
*grin*)
On Tue, 3 Mar 1998, Andrew Martin wrote:
> Marc wrote...
> > On Tue, 3 Mar 1998, Thomas G. Lockhart wrote:
[clipa-clipa]
> > that you can spark
> > up just with a comment like "Linux != Unix"...which, it isn't, its a
> > Unix-like clone...but they can't seem to figure the distinction *rofl*
>
> Agreed... :-) But BSD isn't Unix either - not officially. [Waits for
> Marc to disagree, again...]
Nope - I'm not even sure SCO Open Server is UNIX - and afaik THEY now own
the trademark papers.
Be VERY happy that neither Linux nor BSD is a "real" unix. Those systems
are seriously restrictive and clumsy (I suspect SCO is close - sorry, I've
had to do a lot of tech-service work on SCO systems recently. Not even
Solaris is _THAT_ bad... (close though). Guess I just miss my GNU and BSD
tools too much *grin*)
if you were just to talk about programs, Linux is a superset of
everything (except Irix at this time). If you were to talk about
networking, BSD is the standard that Linux follows. Who wants STREAMS
anyways? If you're talking API interface (and here's where I bate
non-glibc users), GLIBC-2 is the standard for Unix98+. (I still don't see
why postgres doesn't support it... though I haven't gotten around to
writing a patch (or looking recently)...).
> Not to mention the fact that at least one release of Linux did go through
> full Posix certification and is thus allowed to be called Unix :-)
*heh*
Just being a nard...
G'day, eh? :)
- Teunis
On Tue, 3 Mar 1998, teunis wrote:
> If you were to talk about
> networking, BSD is the standard that Linux follows.
And follows badly, last I heard...Linux's networking support
doesn't perform as well as *BSDs, and, last I heard, has been rewritten
from scratch 3 times in the past 6 years or so...
Who wants STREAMS
> anyways? If you're talking API interface (and here's where I bate
> non-glibc users), GLIBC-2 is the standard for Unix98+. (I still don't see
> why postgres doesn't support it... though I haven't gotten around to
> writing a patch (or looking recently)...).
Key reason why we don't support it...nobody except for Linux
currently is using it...
Marc G. Fournier
Systems Administrator @ hub.org
primary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org
> non-glibc users), GLIBC-2 is the standard for Unix98+. (I still don't see
> why postgres doesn't support it... though I haven't gotten around to
> writing a patch (or looking recently)...).
*sigh* Postgres runs just fine on a bug-free version of glibc2. We've heard
rumors that 2.0.7-pre1 from Debian is close enough, but I can't duplicate that
on my RH5.0 production box with Cristian's RH glibc2-2.0.7 package.
btw Cristian, that library in /home/gafton has most files labeled as 2.0.6; is
that expected or are there possibly some more patches available? I tried
installing on my RH5.0 production system and still see the
select '1 min'::timespan;
problem. Haven't had any luck picking out the math code and duplicating the
problem in a 10 line program yet either :(
Also, v6.3 has some extensive new documentation which you will want to get
into /usr/doc, including 4 hardcopy and html manuals. There is a Makefile in
the source doc/ distribution to extract them. Let me know if you want more
details.
- Tom
On Tue, 3 Mar 1998, The Hermit Hacker wrote: > And follows badly, last I heard...Linux's networking support > doesn't perform as well as *BSDs, and, last I heard, has been rewritten > from scratch 3 times in the past 6 years or so... You're right about the rewrite thing. You're quite wrong about the benchmarks, though. Linux's tcp/ip stack is known to be _now_ the fastest around re: internal latency. But there are things that are balancing this when compared with *BSD. Things like Linux's nfs server which sucks big time or sockets creation time which only got better in the development releases. Things are relative. But again, when you say that it was rewritten so many times in the last 6 years you have to remember the Linux is barely six years old :-) Cristian -- ---------------------------------------------------------------------- Cristian Gafton -- gafton@redhat.com -- Red Hat Software, Inc. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ UNIX is user friendly. It's just selective about who its friends are.
On Wed, 4 Mar 1998, Thomas G. Lockhart wrote: > btw Cristian, that library in /home/gafton has most files labeled as 2.0.6; is > that expected or are there possibly some more patches available? I tried > installing on my RH5.0 production system and still see the I am doing a new one with lots of more patches included. Watch that directory... > Also, v6.3 has some extensive new documentation which you will want to get > into /usr/doc, including 4 hardcopy and html manuals. There is a Makefile in Already done that. Check out the new packages. Cristian -- ---------------------------------------------------------------------- Cristian Gafton -- gafton@redhat.com -- Red Hat Software, Inc. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ UNIX is user friendly. It's just selective about who its friends are.
>>>>> "Thomas" == Thomas G Lockhart <lockhart@alumni.caltech.edu> writes: Thomas> *sigh* Postgres runs just fine on a bug-free version of glibc2. We've Thomas> heard rumors that 2.0.7-pre1 from Debian is close enough, but I can't Thomas> duplicate that on my RH5.0 production box with Cristian's RH Thomas> glibc2-2.0.7 package. Thought you might like to know, Tom, with Cristian's glibc2-2.0.7 rpm I downloaded yesterday, I was able to get: mydb=> select '1 min'::timespan; ?column? -------- @ 1 min (1 row) And all time-related regression tests succeeded. That was with gcc-2.7.2.3, (-O3 -m486), _however_, when compiled with gcc-2.8.0 (-O3 -mpentium), all that nasty time stuff just crept back again, don't know why. BTW, something I'm a bit concerned about -- re: the ``~30 sec deficit'' in regression test timing results you mentioned a couple of days ago -- I'm also seeing it here between the official 6.3 and a Feb-15 snapshot (no, LOCALE's always undef) on both Linux and FreeBSD, and can consistently reproduce it, with the same test suite (the one from the official 6.3). And thanks for the docs! -Pailing
> Thomas> *sigh* Postgres runs just fine on a bug-free version of glibc2. We've
> Thomas> heard rumors that 2.0.7-pre1 from Debian is close enough, but I can't
> Thomas> duplicate that on my RH5.0 production box with Cristian's RH
> Thomas> glibc2-2.0.7 package.
>
> Thought you might like to know, Tom, with Cristian's glibc2-2.0.7 rpm I
> downloaded yesterday, I was able to get:
> mydb=> select '1 min'::timespan;
> ?column?
> --------
> @ 1 min
> (1 row)
>
> And all time-related regression tests succeeded.
That's good news, but I'm annoyed I haven't been able to get this result myself
yet. Did you use the redhat beta rpm for postgres, or did you do a clean install
from sources?
> That was with gcc-2.7.2.3, (-O3 -m486), _however_, when compiled with
> gcc-2.8.0 (-O3 -mpentium), all that nasty time stuff just crept back again,
> don't know why.
>
> BTW, something I'm a bit concerned about -- re: the ``~30 sec deficit'' in
> regression test timing results you mentioned a couple of days ago -- I'm also
> seeing it here between the official 6.3 and a Feb-15 snapshot (no, LOCALE's
> always undef) on both Linux and FreeBSD, and can consistently reproduce it,
> with the same test suite (the one from the official 6.3).
I ended up convincing myself that most of the time difference was the overhead
in compiling with USE_LOCALE turned on. The RH rpm is compiled with it on, and
most of my testing is done with it turned off, but I had turned it on to track
down problems with the money type. Anyway, my times are now within ~10sec of the
shortest time I had ever seen (of course, it's easy to shave time off when you
skip essential code; I'll take the 10sec hit :)
> And thanks for the docs!
Sure. Consider finding something which interests you to write about for the next
release :))
- Tom
On Wed, 4 Mar 1998, Thomas G. Lockhart wrote:
> > non-glibc users), GLIBC-2 is the standard for Unix98+. (I still don't see
> > why postgres doesn't support it... though I haven't gotten around to
> > writing a patch (or looking recently)...).
>
> *sigh* Postgres runs just fine on a bug-free version of glibc2. We've heard
> rumors that 2.0.7-pre1 from Debian is close enough, but I can't duplicate that
> on my RH5.0 production box with Cristian's RH glibc2-2.0.7 package.
Then consider that the minimum supported and ignore it... *grin*
Solution found! :)
[GNU takes very long time to fix things toujours]
G'day, eh? :)
- Teunis
(PS: Linux's kernel networking layer has been extensively rewritten over
the last year and a half.. and is (afaik) considerably faster... at least
in the 2.1 kernels... though IIRC it was rewritten also in the 1.3
kernels... The NFS probs were solved in 2.1 a while ago though)
>>>>> "Thomas" == Thomas G Lockhart <lockhart@alumni.caltech.edu> writes: Thomas> That's good news, but I'm annoyed I haven't been able to get this Thomas> result myself yet. Did you use the redhat beta rpm for postgres, or Thomas> did you do a clean install from sources? clean install from source, that's why I listed the compiler versions and flags. Isn't that what you've been trying to get it to work? >> That was with gcc-2.7.2.3, (-O3 -m486), _however_, when compiled with >> gcc-2.8.0 (-O3 -mpentium), all that nasty time stuff just crept back again, >> don't know why. >> >> BTW, something I'm a bit concerned about -- re: the ``~30 sec deficit'' in >> regression test timing results you mentioned a couple of days ago -- I'm Thomas> I ended up convincing myself that most of the time difference was the Thomas> overhead in compiling with USE_LOCALE turned on. The RH rpm is Thomas> compiled with it on, and most of my testing is done with it turned I thought about doing some more testing and possibly tracking it down, and see whether that was caused by some bugfixes somewhere, but without all the snapshots from perhaps Feb-20 all the way up to the official release, it's a bit hard. -Pailing
> Thomas> That's good news, but I'm annoyed I haven't been able to get this
> Thomas> result myself yet. Did you use the redhat beta rpm for postgres, or
> Thomas> did you do a clean install from sources?
>
> clean install from source, that's why I listed the compiler versions and
> flags. Isn't that what you've been trying to get it to work?
No. I've been developing on RH4.2 (making source builds on that), and trying to
use rpms for RH5.0 built at RedHat to verify the glibc2 performance. May have to
do a build from source to get to the bottom of things, but I'm hoping not...
> >> That was with gcc-2.7.2.3, (-O3 -m486), _however_, when compiled with
> >> gcc-2.8.0 (-O3 -mpentium), all that nasty time stuff just crept back again,
> >> don't know why.
> I thought about doing some more testing and possibly tracking it down, and see
> whether that was caused by some bugfixes somewhere, but without all the
> snapshots from perhaps Feb-20 all the way up to the official release, it's a
> bit hard.
Well, have you tried the CVSup static package on RH5.0 yet? Don't know if it
would work, but if it did it would allow you to get snapshots as you want; in
fact I'm doing that right now trying to track down a problem introduced sometime
after 980112 and before 980201. Downloading 980120 right now as I (sort of)
binary search through the possibilities.
- Tom
In message <34FD57AE.F4C716F2@alumni.caltech.edu>, "Thomas G. Lockhart" writes:
> > Thomas> That's good news, but I'm annoyed I haven't been able to get this
> > Thomas> result myself yet. Did you use the redhat beta rpm for postgres, or
> > Thomas> did you do a clean install from sources?
> >
> > clean install from source, that's why I listed the compiler versions and
> > flags. Isn't that what you've been trying to get it to work?
>
> No. I've been developing on RH4.2 (making source builds on that), and trying
> to
> use rpms for RH5.0 built at RedHat to verify the glibc2 performance. May have
> to
> do a build from source to get to the bottom of things, but I'm hoping not...
>
> > >> That was with gcc-2.7.2.3, (-O3 -m486), _however_, when compiled with
> > >> gcc-2.8.0 (-O3 -mpentium), all that nasty time stuff just crept back aga
> in,
> > >> don't know why.
> > I thought about doing some more testing and possibly tracking it down, and
> see
> > whether that was caused by some bugfixes somewhere, but without all the
> > snapshots from perhaps Feb-20 all the way up to the official release, it's
> a
> > bit hard.
>
> Well, have you tried the CVSup static package on RH5.0 yet? Don't know if it
> would work, but if it did it would allow you to get snapshots as you want; in
> fact I'm doing that right now trying to track down a problem introduced somet
> ime
> after 980112 and before 980201. Downloading 980120 right now as I (sort of)
> binary search through the possibilities.
>
> - Tom
>
>
Just thought you'd like another data point. I just installed Cristian's
glibc2-2.0.7 packages. Like Thomas, I still get:
postgres=> select '1 min'::timespan;
?column?
------------
@ 60.00 secs
(1 row)
Without recompiling. With recompiling, I get:
postgres=> select '1 min'::timespan;
?column?
--------
@ 1 min
(1 row)
I'm using gcc 2.7.2.3 -O2.
Tom Szybist
szybist@boxhill.com
> Just thought you'd like another data point. I just installed Cristian's
> glibc2-2.0.7 packages. Like Thomas, I still get:
>
> postgres=> select '1 min'::timespan;
> ?column?
> ------------
> @ 60.00 secs
> (1 row)
>
> Without recompiling. With recompiling, I get:
>
> postgres=> select '1 min'::timespan;
> ?column?
> --------
> @ 1 min
> (1 row)
>
> I'm using gcc 2.7.2.3 -O2.
Well, this narrows it down a lot! Wonder why it requires a recompile?? afaik there
isn't any static library linking involved...
- Tom
In message <34FD77B1.848B35E7@alumni.caltech.edu>, "Thomas G. Lockhart" writes: > > Just thought you'd like another data point. I just installed Cristian's > > glibc2-2.0.7 packages. Like Thomas, I still get: > > > > postgres=> select '1 min'::timespan; > > ?column? > > ------------ > > @ 60.00 secs > > (1 row) > > > > Without recompiling. With recompiling, I get: > > > > postgres=> select '1 min'::timespan; > > ?column? > > -------- > > @ 1 min > > (1 row) > > > > I'm using gcc 2.7.2.3 -O2. > > Well, this narrows it down a lot! Wonder why it requires a recompile?? afaik there > isn't any static library linking involved... > > - Tom > Could an include file account for this? Tom Szybist szybist@boxhill.com
On Wed, 4 Mar 1998, Thomas A. Szybist wrote: > Just thought you'd like another data point. I just installed Cristian's > glibc2-2.0.7 packages. Like Thomas, I still get: Maybe you were using my postgresql package which was a little older ? I have new rpms on ftp://ftp.redhat.com/home/gafton/pgsql Cristian -- ---------------------------------------------------------------------- Cristian Gafton -- gafton@redhat.com -- Red Hat Software, Inc. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ UNIX is user friendly. It's just selective about who its friends are.