We're using contrib/dbmirror in production, and it works fine, if your definition of fine is "ok". We run the mirroring process every five seconds, and have a few triggers and whatnot written to facilitate a hot-swap failover.
Our transactional volume isn't high enough yet to cause us problems (less than 30 a minute), but for now, this is ok. I'm tracking the "real" pgreplication stuff, and occasionally take a desultory trip into WAL land, when I can grab a minute.
If you know UNIX there is hardly anything you can do with Windows NT that is any more secure than UNIX.
Oh, really? Tell me, then, how do I revoke root's ability to even open a file? How about allowing root to read only, and allowing all members of a given set of users the ability to do nothing but append? I'd get into file records, but according to Unix dogma, nobody ever needs those.
Security is complicated because computers are complicated. Sweeping issues under a "not our problem" carpet doesn't address them. This isn't by the way an argument from the position that NT is great -- it's clearly not. But it's security system derives from VMS, and VMS was pretty great.
The idea that security is adequately managed by three bytes and two ints is plainly wrong -- I would think that Unix advocates would want to improve upon the system's woeful history.
I could not disagree more. Having a superuser AT ALL is a security mistake. How many exploits would Unix have suffered if there was no possiblity of universal privilege escalation? The issue isn't so much with interactive users as it is with the kernel just stepping aside when presented with uid 0.
Blaming the sysadmin for Unix security holes when the entire security infrastructure is what's b0rken is blaming the victim. Shamingly, this is a situation where NT is radically more advanced than most (all?) Unixes.
The problems with AFS are that it's almighty bizarre to Joe Average Sysadmin, and that the clients aren't as widespread -- I was going to use "mature," but that's not really a word any rational person would use to describe NFS in/any/ respect -- as the clients for NFS. I love AFS, have proselytized for it in the past, but it's just hard to overcome sysadmin laziness, perceived difficulty of administration and installation, and the network effect of much crappier protocols.
this is wrong for many reasons, but perhaps the simplest is: think, the VM is probably written in C... how is it going to be faster than C?
Uh, maybe because the processor isn't executing C? The VM and the C compiler are both producing machine code. It's perfectly feasible for a smart VM -- say, one with runtime code rewriting -- to run faster than a static C binary.
What's the last C project of significant size that compiled and ran without changes on more than one platform? Why do you think GNU autoconf exists? Just for kicks? The reason so much software is written in C is ubiquity, not portability.
It's a typical X Windows canard: "well, we COULD have said nifty feature, but we don't WANT it. But if YOU want it (by the way WE think it's GAUCHE), go ahead an implement it, SUCKER." It's the second most popular pro-X blather, after "does YOUR windowing system have NETWORK TRANSPARENCY?"
Technically, X has indeed progressed to where it's almost as sophisticated at NeWS or NeXT. But nobody on the business end of X ever sees much of that benefit, as applications are by necessity coded for the typical Sun 3/60 with a 4-bit framebuffer, or alternately, for a Sourceforge abandoned el-cheapo "themable" C++ toolkit that requires a trip down a dependency path hell to make building mh look simple. Or maybe they're built against some nifty new microsubversion of GTK, or QT, or or or or or. OR.
This weekend I bought a Powermac to replace a PC running Gentoo Linux. Guess which system was easier to set up subpixel font rendering on? Guess which system allows me printer output that looks EXACTLY as the display does? Guess which system applies my usability configuration COMPLETELY across the board? And guess which system has a million almost identical but actually lethally incompatible text files in a million nonstandard locations controlling every single behavior of the windowing system BUT the one you want to use?
"But I COULD do it, if I thought it wasn't so LAME. By the way, did I mention that MY window system has NETWORK TRANSPARENCY?"
Sweet creeping zombie Jesus. At least the world of computing can take comfort in the fact that this trainwreck of a project will never ever see the light of day.
sed, awk, find, vi, Perl, the C pre-processor, the C language, the C++ language, termcap, curses, ncurses, Unix terminal "handling" in the general case, lpr, lp, CUPS, lpr-ng, bash, zsh, tcsh, NFS, NFSv3, AFS...
The list goes on. Up with aggressive mediocrity! Worse is better! Welcome to Unix, here's your accordion.
Fair enough. Right now, the fastest processor in the world is the Pentium 4 3.06ghz: 1130/1103 (int/fp). For pure floating-point horses, it's the Itanic 2 743/1427 (int/fp).
So a 2.5ghz 970 would be close in performance to both of today's fastest shipping processors. It's likely that the P4 and Itanic will be 15-20% faster in six months, so IBM will still be lagging in the performance hunt. However, it's striking how much closer to the peak performers this chip will move IBM -- and, by extension, Apple.
For comparison's sake, the P4 Xeon @ 1.8ghz pulls 703/717 (int/fp) on SPEC CPU2000.
Assuming a linear scaling in SPEC performance, we can look forward to a 2.5ghz 970 scoring about 1294/1460, which is pretty respectable. Not a world beater (especially for 2H03), but a far cry from the abominable performance of the current G4.
HPaq is definitely up to their necks in Itanic (which, btw, is already faster than EV7), but even if they weren't, I find it hard to believe that they'd have the process to shrink EV7 further and clock it higher.
The truly scary thing is that there's plenty of headroom in the P4 core. Intel is talking about 7ghz on their next shrink. One hopes that AMD is able to keep up, if only to keep the price of gross nasty fast microprocessors within the grasp of the common man.
So the P4 is *1.4 times* as fast on integer maths, and two percent slower at floating point. Ouch. Care to rethink that statement about the EV7? Sure, it's 64bit, and scalable, and a much nicer architecture, to boot. But I wasn't arguing address space size, suitability for large-scale NUMA or SMP clustering, or ease of writing compiler backends, was I?
Peace, 'jfb
PS: Even more horrifying for fans of elegance in their microarchitectures:
The only Yamaha hardware in my house are two SG-2000 and one SA-2100 guitars. They rock. The 2000s, in particular, are total monsters of rock Wall Of Distortion Marc Bolan T-Rex Bowie machines. The best Japanese guitars ever made.
He has to stop drinking in order to qualify for the astronaut training. And it was non-alcoholic champagne, to celebrate Barney's selection as the civilian astronaut.
"De fault! The two sweetest words in the English language!"
Slashdot Mirror
> The Scientologists don't think they're a cult, either.
On the other hand, the Scientologists could probably design a working OoOE processor.
*rimshot*
'jfb
This took me three tries to get. Nicely done.
'jfb
I didn't know the slashdot crowd was so into French sculpture.
...
You learn something new every day
'jfb
We're using contrib/dbmirror in production, and it works fine, if your definition of fine is "ok". We run the mirroring process every five seconds, and have a few triggers and whatnot written to facilitate a hot-swap failover.
Our transactional volume isn't high enough yet to cause us problems (less than 30 a minute), but for now, this is ok. I'm tracking the "real" pgreplication stuff, and occasionally take a desultory trip into WAL land, when I can grab a minute.
'jfb
If you know UNIX there is hardly anything you can do with Windows NT that is any more secure than UNIX.
Oh, really? Tell me, then, how do I revoke root's ability to even open a file? How about allowing root to read only, and allowing all members of a given set of users the ability to do nothing but append? I'd get into file records, but according to Unix dogma, nobody ever needs those.
Security is complicated because computers are complicated. Sweeping issues under a "not our problem" carpet doesn't address them. This isn't by the way an argument from the position that NT is great -- it's clearly not. But it's security system derives from VMS, and VMS was pretty great.
The idea that security is adequately managed by three bytes and two ints is plainly wrong -- I would think that Unix advocates would want to improve upon the system's woeful history.
'jfb
I could not disagree more. Having a superuser AT ALL is a security mistake. How many exploits would Unix have suffered if there was no possiblity of universal privilege escalation? The issue isn't so much with interactive users as it is with the kernel just stepping aside when presented with uid 0.
Blaming the sysadmin for Unix security holes when the entire security infrastructure is what's b0rken is blaming the victim. Shamingly, this is a situation where NT is radically more advanced than most (all?) Unixes.
'jfb
... or Monty Burns quoting Ahab?
'jfb
The problems with AFS are that it's almighty bizarre to Joe Average Sysadmin, and that the clients aren't as widespread -- I was going to use "mature," but that's not really a word any rational person would use to describe NFS in /any/ respect -- as the clients for NFS. I love AFS, have proselytized for it in the past, but it's just hard to overcome sysadmin laziness, perceived difficulty of administration and installation, and the network effect of much crappier protocols.
'jfb
... which is the great failure of the POSIX security metaphor.
'jfb
"Sighting" in this context means "seen while alive".
'jfb
this is wrong for many reasons, but perhaps the simplest is: think, the VM is probably written in C... how is it going to be faster than C?
Uh, maybe because the processor isn't executing C? The VM and the C compiler are both producing machine code. It's perfectly feasible for a smart VM -- say, one with runtime code rewriting -- to run faster than a static C binary.
'jfb
Portability? In C? Are you joking?
What's the last C project of significant size that compiled and ran without changes on more than one platform? Why do you think GNU autoconf exists? Just for kicks? The reason so much software is written in C is ubiquity, not portability.
'jfb
"Uh, what?"
It's a typical X Windows canard: "well, we COULD have said nifty feature, but we don't WANT it. But if YOU want it (by the way WE think it's GAUCHE), go ahead an implement it, SUCKER." It's the second most popular pro-X blather, after "does YOUR windowing system have NETWORK TRANSPARENCY?"
Technically, X has indeed progressed to where it's almost as sophisticated at NeWS or NeXT. But nobody on the business end of X ever sees much of that benefit, as applications are by necessity coded for the typical Sun 3/60 with a 4-bit framebuffer, or alternately, for a Sourceforge abandoned el-cheapo "themable" C++ toolkit that requires a trip down a dependency path hell to make building mh look simple. Or maybe they're built against some nifty new microsubversion of GTK, or QT, or or or or or. OR.
This weekend I bought a Powermac to replace a PC running Gentoo Linux. Guess which system was easier to set up subpixel font rendering on? Guess which system allows me printer output that looks EXACTLY as the display does? Guess which system applies my usability configuration COMPLETELY across the board? And guess which system has a million almost identical but actually lethally incompatible text files in a million nonstandard locations controlling every single behavior of the windowing system BUT the one you want to use?
"But I COULD do it, if I thought it wasn't so LAME. By the way, did I mention that MY window system has NETWORK TRANSPARENCY?"
'jfb
Bzzzzt.
5 /smalltalk1.html
"The very first Smalltalk system was a thousand lines Basic program, which successfully computed 3+4 in October 1972."
From http://www.cosc.canterbury.ac.nz/~wolfgang/cosc20
'jfb
What, Perl didn't have ENOUGH syntax before?
Sweet creeping zombie Jesus. At least the world of computing can take comfort in the fact that this trainwreck of a project will never ever see the light of day.
'jfb
sed, awk, find, vi, Perl, the C pre-processor, the C language, the C++ language, termcap, curses, ncurses, Unix terminal "handling" in the general case, lpr, lp, CUPS, lpr-ng, bash, zsh, tcsh, NFS, NFSv3, AFS ...
The list goes on. Up with aggressive mediocrity! Worse is better! Welcome to Unix, here's your accordion.
'jfb
Fair enough. Right now, the fastest processor in the world is the Pentium 4 3.06ghz: 1130/1103 (int/fp). For pure floating-point horses, it's the Itanic 2 743/1427 (int/fp).
So a 2.5ghz 970 would be close in performance to both of today's fastest shipping processors. It's likely that the P4 and Itanic will be 15-20% faster in six months, so IBM will still be lagging in the performance hunt. However, it's striking how much closer to the peak performers this chip will move IBM -- and, by extension, Apple.
'jfb
For comparison's sake, the P4 Xeon @ 1.8ghz pulls 703/717 (int/fp) on SPEC CPU2000.
Assuming a linear scaling in SPEC performance, we can look forward to a 2.5ghz 970 scoring about 1294/1460, which is pretty respectable. Not a world beater (especially for 2H03), but a far cry from the abominable performance of the current G4.
'jfb
HPaq is definitely up to their necks in Itanic (which, btw, is already faster than EV7), but even if they weren't, I find it hard to believe that they'd have the process to shrink EV7 further and clock it higher.
The truly scary thing is that there's plenty of headroom in the P4 core. Intel is talking about 7ghz on their next shrink. One hopes that AMD is able to keep up, if only to keep the price of gross nasty fast microprocessors within the grasp of the common man.
'jfb
Let's go to the numbers. From spec.org:
P4, 3.06ghz: 1130/1103 (int/fp)
EV7, 1.15ghz: 795/1124 (int/fp)
So the P4 is *1.4 times* as fast on integer maths, and two percent slower at floating point. Ouch. Care to rethink that statement about the EV7? Sure, it's 64bit, and scalable, and a much nicer architecture, to boot. But I wasn't arguing address space size, suitability for large-scale NUMA or SMP clustering, or ease of writing compiler backends, was I?
Peace,
'jfb
PS: Even more horrifying for fans of elegance in their microarchitectures:
Itanic^2, 1.0ghz: 810/1431 (int/fp)
The only Yamaha hardware in my house are two SG-2000 and one SA-2100 guitars. They rock. The 2000s, in particular, are total monsters of rock Wall Of Distortion Marc Bolan T-Rex Bowie machines. The best Japanese guitars ever made.
'jfb
Funny, then, the the single highest performing microprocessor in the world is one of those blown Civics, isn't it?
'jfb
Anywhere between five and fifteen times faster?
'jfb
He has to stop drinking in order to qualify for the astronaut training. And it was non-alcoholic champagne, to celebrate Barney's selection as the civilian astronaut.
"De fault! The two sweetest words in the English language!"
'jfb
Dentist: These predate stainless steel, so you can't get them wet.
Best episode ever.
'jfb