If you're running it as a client, it's not actually decoding the mp3 - the server is.
I was speaking of borrowing back horsepower from the client to reduce network load. Playing MP3 or Ogg files locally costs roughly 10x less LAN traffic and is much less laggy. This applies much more so for video or 3D stuff. A 486 is generally not up to decoding MP3s, look toward a Pentium over 100MHz as a working minimum local-Ogg player.
Therefore unusable on even a 486DX4-120, the fastest 486 I know of. If it's sucking 100% CPU locally - make that 50% for a hypothetically faster CPU but IRL probably around 70% - add in the overhead of fetching from the net, try to display something at the same time, and you're a dead duck.
Y'don't get it at all... Tux fields the static stuff and handballs the dynamic stuff. Even on heavily dynamic web pages, many or most of the files and bytes are static (images, static frames and such) and that gets processed roughly an order of magnitude faster than IIS. Even the stuff the Tux handballs (to Apache, for example) can be processed damn fast (e.g. Zend scripting).
I suppose you heard that the worlds largest webmail site runs on a cluster of Windows 2000 machines.
Yup. More than twice as many machines as the same load needed under FreeBSD, and each more powerful, and it's never been as stable since, despite the addition of hardware load balancing.
As far as I remember a lot of skull sweat from RH gurus went into that Linux machine as well in mindcraft tests. Did it help?
The skull sweat went into making Linux run a test carefully designed to make Windows shine (four network cards, my ass! what kind of nutcase...? never mind), and it did help a fair bit on the day. But where it really helped is that Linux's performance in areas formerly ruled by Windows went through the roof. In particular the handling of multiple network streams to the same source improved, again, more than an order of magnitude within 3 months, and the `select' mechanism underwent the revamp that allowed the old clunker you so despise to reliably handle an unprecedented number of hits. Error 500s? We don' need no steenking Error 500s!
IIS has never been Ring 0.
Yeah? Look closer. And explain away its layer-violating `TCP cheating' (if you can) while you're there.
You've obviously haven't seen.NET technologies in person.
Oooh, yes I have. (-:
You've obviously only ever listened to the good side.
MySQL has always been the speed leader in Open Source databases.
Now there's an opening begging to be taken! (dons his asbestos undies)
MySQL has only ever been shown by anyone but MySQL AB* to be faster at very simple mostly-or-all-read-only tasks, even with row-level locking. PostgreSQL, for example, multiplexes writes - which effectively removes even row-level-locking delays. As soon as you pile on the load, get complex in your queries, or start doing a serious amount of writing, the bencharks go all wahoonie-shaped.
On top of that, innoDB is a bandaid to try bringing real SQL features to MySQL, which leads to two observations:
MySQL proper is misnamed (it's not SQL, certainly not complete ACID, only a SQL subset); and
even with innoDB its repertoire of ACID features is incomplete
I'd also be interested in seeing Firebird and SAP DB in any benchmarks done. And MS SQL Server, if you're in a place where the NDA-style thou-shalt-not-publish-benchmarks EULA clause doesn't matter. I do like choice, but most firmly believe that some of the attention falling on almost-GPL MySQL would be better spent if devoted to some of the more capable really-GPL alternates.
* now who (M<cough>ro<cough>t) does that remind you of?
True, as long as you don't want to run anything serious locally (like an MP3 decoder or 3D app). Or at a serious res (like 2000x1600x24-or-x32 for a large monitor).
Who the heck needs a single P2-500 junker serving static content?
More than 90% of the websites in existence. (-:
Not that a Tux-based webserver is limited to static content, it's just that the static content whips out the door fast enough to make the cables bulge. And if it can do that on a clunker, what can it do on a real server?
Quads, 8-way servers, clusters, buttloads of RAM and data, dynamic pages served by each side's engine of choice.
Mmmm. Single machines, do we go for the high-end balanced-and-tuned zSeries or one of SGI's monsters? Clusters... I don't suppose you noticed that the world's most popular search engine is a passel of Linux clusters? I vote for the 512-way server.
Last time I checked Windows ripped Linux apart on tests like these.
Only on tests where the majority of skull-sweat went into selecting a setup that favoured every one of Windows' strengths and suffered from every one of Linux's weaknesses. On a level playing field, or to put it another way, something `resembling reality', IIS's ass is grass.
And IIS 6 (surprise!) now includes a kernel module as well, so TUX has lost its only advantage.
Er, no? IIS has always been Ring 0, which is why vulnerabilities in it have always been a catastrophe. That hamster has already been whipped to death.
And when we consider ASP.NET for generating the dynamic content we pretty much see that in terms of technology and performance the "inept" MS is far ahead.
Catching up with Java (Real Soon Now) is supposed to be `far ahead'? Gimme a break!
Speaking of buttloads of RAM, that's the way to make Samba really party. It uses about 20-30% more RAM per instance for file sharing than Windows 2000, but OTOH can serve about 10x as many connections on the same hardware. So which would you rather do, pay for ten servers, ten sets of licences and figure out how to integrate them? Or pay for one server and double the RAM?
How would you hang more than one PS/2 keyboard off one machine anyway?
PS/2 is PS/2 is PS/2; you can plug two PS/2 mice or two PS/2 keyboards into a standard machine, diddle with the drivers to tell them about the alternate IO and IRQ addresses and you're away. You cal also get multi-PS/2 port adapters and a lot of the "high-speed serial port" adapters are close enough to 5V RS232 (which is what a PS/2 port is) to work.
it seems to me that you wouldn't really need more than one textmode console
That's not the problem. PS/2 keyboards (all keyboards by default) are routed through the console driver. Hit a key on any keyboard, it goes into the console. There is no way to distinguish which keyboard it came from. You cannot have multiple console drivers (although I believe there is a patch for it). You can, however, tell the console not to bind the USB keyboards, leaving them free for XFree86. Later versions of XFree86 are able to deal with USB keyboards directly (don't recall whether you need a patch or not) rather than through the console.
Not sure if Xfree86 has been fixed yet to not treat all scroll-wheels on each mouse as one, but as at about a year ago, it didn't distinguish. If it has, this here AOpen Optical OpenEye Wheel Mouse model O-35G is about to become multi-talented.
The sticky problems are still with the Linux Console code, in that it doesn't really believe in more than one console with keyboard. However, USB keyboards are easier than PS/2 keyboards (easier to not route through the console), and mice are a non-issue (again, do not route through GPM).
As an alt to an X terminal, try a diskless fanless thin-client box. Motium make those with some really special features, but I don't know if the whizz-bang ones are ready for sale yet.
I'd really be interested in seeing another independent test between the latest release of RH Linux (tweaked and patched ad nauseam) and Windows Server 2003 with IIS6. It seems to me so far that Linux is up for serious buttkicking again.
For that to be fair, you'd want to use the latest Linux as well as the latest Windows (that is to say, if you tweak one you should tweak both properly). In which case there is no question but that Windows + IIS will get their collective butts kicked all over the stadium by Linux +Tux + Apache. There are no more rabbits for Microsoft to pull from their corporate hats in this domain, but Linux whupped them good last time I looked (some dizzying number of requests per second on a single-CPU P2-500), and is only just getting up to stride.
Dumping IIS because of a few security holes is really fucking stupid for a ton of reasons that I don't even have time to go into.
So... you think you shouldn't be dumping it just for breaching standards, being slow, sucking resources and Being Written By Microsoft With Malice Aforethought?
I personally don't think you should be dumping it because of a few security holes, I think you should be dumping it for having lots of security holes.
these results (I mean TPC) are provided by HARDWARE MANUFACTURERS
And so...? The TCO study was provided by IDC, using carefully chosen figures as expressed by Microsoft when they requested the study. The name on the shi8ngle doesn't matter nearly as much as who actually provided the impetus.
Think back to MindCraft, where they did things like using 4x100Mb cards instead of 1x1000Mb like anyone else would, and spreading Windows' logging across four partitions. Sure, they finally managed to tweak good results out of it (and of course the Linux kernel was so much better in those areas three months later anyway), but who in their right mind is going to set up a production machine like that, or even know that that doing so would make it faster?
Likewise choosing high-end hardware specifically for your TPC and tuning it to within an inch of it's life is not what your average punter is going to do, and only vaguely related to what a good DBA will do. They're going to buy a stock machine, do a stock install, make a few of the more popular tweaks, and run that. In those circumstances, PostgreSQL will comprehensively pound MS SQL Server into the pavement.
Lots more out there, I'm sure... you can easily skip some when your brain melts down after scanning 200 entries and maybe clicking on 50 to see if they're as relevant as they seem. Still much easier than reinventing the wheel yet again.
Among the enterprise-level DBs, SQL Server beats the crap outta every other DB both in price/performance, price AND performance.
That depends on who classes what as `enterprise-level'. PostgreSQL eats SQL Server's lunch pretty much across the board except on carefully architected benchmarking systems (think MindCraft and four separate logging partitions). SQL Server is basically only warmed-over SyBase, even though a lot of the later warming-over was supervised by a very competent ex-DEC database bloke.
OBTW, don't trust Microsoft's costings for their competitor's systems as far as you can throw them.
For example, they sponsored a recent IDC study which assumed a software cost of USD$940/1009/340/6609/1390 (networking/file/print/security/web) for each Linux system (real cost $0/0/0/0/0, versus (they claim) $211/$3988/1665/5829/7107 for Windows 2000) - which factor alone made more than a slight difference to the resulting TCO figures. Microsoft-sponsored-IDC quote staffing costs for Linux at significantly higher than those for Windows, where in real life the reduction in headcount brought about by increased reliability prunes those costs to about 1/2 or 1/3. They also quote downtime costs for a Linux-based file server at over $4000 a year and nearly double that for a print server! I could do a complete reinstall from scratch every quarter and still cost less than that.
Being pessimistic and assuming that a Linux server collapses half as often as Windows, and nett staff costs are around half those of Windows, the bottom lines for Linux work out to: Network 64% File 54% Print 55% Security 56% Web 40% of Windows. Quite different to IDC's own conclusions.
IDC also do $3 worth of outsourcing for their WIndows file server. Good trick.
You can't import from a backup made by a different verion of Postgres.
I can. I have, many times.
The older version I was using was trying to parse the comment marks (---).
The command-line psql segfaults.
These lead me to suspect that your implementation was broken. I've never seen them happen.
not being able to delete a column, or change a column type.
ALTER TABLE [ ONLY ] table [ * ] ALTER [ COLUMN ] column { SET DEFAULT value | DROP DEFAULT }
ALTER TABLE [ ONLY ] table [ * ] RENAME [ COLUMN ] column TO newcolumn
I am always changing a column type to varchar(x) and pruning garbage off it (eg dollar signs and commas) and then converting it back to a numeric (double) field.
Hey, what? You can't store dollar signs and garbage in an integer or float, you shouldn't be trying to feed that to your db in the first place! If you want to do that kind of thing, amidst a `live' table is not the place for it: use a temp table and do it properly. Whis is probably why the PostgreSQL people didn't implement it.
I'd rather wait for MySQL to add the one thing I'm actually waiting for: stored procedures
I hope you've got a lifespan like Methuselah's, then. PostgreSQL does stored procedures in a variety of languages already. Your post does sound like a BASIC programmer grunting and squealing when presented with a real language that insists on him doing stuff like decalring variables, and has scoping etc, forcing him to do `work' (actually investing in manageability) that he didn't have to do before - at least, not up front and in small doses.
If you can show me how to do a base (kernel with very basic tools *ONLY*) install of any of those distros I would like to see it.
Boot the installer CD, type `rescue' and run that?
Selecting a minimal Mandrake system (circa 30MB) has been easy for a very long time. Choose individual packages, click on the recycle-looking icon, disable anything you don't like. Look in/root/drakx for a floppy image which will re-run that install for the next machine (you can write that into a CD's boot area if you don't want to boot from floppy).
Slashdot Mirror
Speak for yourself. I use it all the time.
I was speaking of borrowing back horsepower from the client to reduce network load. Playing MP3 or Ogg files locally costs roughly 10x less LAN traffic and is much less laggy. This applies much more so for video or 3D stuff. A 486 is generally not up to decoding MP3s, look toward a Pentium over 100MHz as a working minimum local-Ogg player.
Therefore unusable on even a 486DX4-120, the fastest 486 I know of. If it's sucking 100% CPU locally - make that 50% for a hypothetically faster CPU but IRL probably around 70% - add in the overhead of fetching from the net, try to display something at the same time, and you're a dead duck.
Published to the LKML. Sorry, off out the door to RAID a machine in Belmont.
No.
Are you just making your name up as you go along? (-:
Planning ahead, same as prepense. Which admittedly doesn't characterise Microsoft products very well. (-:
Y'don't get it at all... Tux fields the static stuff and handballs the dynamic stuff. Even on heavily dynamic web pages, many or most of the files and bytes are static (images, static frames and such) and that gets processed roughly an order of magnitude faster than IIS. Even the stuff the Tux handballs (to Apache, for example) can be processed damn fast (e.g. Zend scripting).
Yup. More than twice as many machines as the same load needed under FreeBSD, and each more powerful, and it's never been as stable since, despite the addition of hardware load balancing.
The skull sweat went into making Linux run a test carefully designed to make Windows shine (four network cards, my ass! what kind of nutcase...? never mind), and it did help a fair bit on the day. But where it really helped is that Linux's performance in areas formerly ruled by Windows went through the roof. In particular the handling of multiple network streams to the same source improved, again, more than an order of magnitude within 3 months, and the `select' mechanism underwent the revamp that allowed the old clunker you so despise to reliably handle an unprecedented number of hits. Error 500s? We don' need no steenking Error 500s!
Yeah? Look closer. And explain away its layer-violating `TCP cheating' (if you can) while you're there.
Oooh, yes I have. (-:
You've obviously only ever listened to the good side.
Now there's an opening begging to be taken! (dons his asbestos undies)
MySQL has only ever been shown by anyone but MySQL AB* to be faster at very simple mostly-or-all-read-only tasks, even with row-level locking. PostgreSQL, for example, multiplexes writes - which effectively removes even row-level-locking delays. As soon as you pile on the load, get complex in your queries, or start doing a serious amount of writing, the bencharks go all wahoonie-shaped.
On top of that, innoDB is a bandaid to try bringing real SQL features to MySQL, which leads to two observations:
I'd also be interested in seeing Firebird and SAP DB in any benchmarks done. And MS SQL Server, if you're in a place where the NDA-style thou-shalt-not-publish-benchmarks EULA clause doesn't matter. I do like choice, but most firmly believe that some of the attention falling on almost-GPL MySQL would be better spent if devoted to some of the more capable really-GPL alternates.
* now who (M<cough>ro<cough>t) does that remind you of?
True, as long as you don't want to run anything serious locally (like an MP3 decoder or 3D app). Or at a serious res (like 2000x1600x24-or-x32 for a large monitor).
More than 90% of the websites in existence. (-:
Not that a Tux-based webserver is limited to static content, it's just that the static content whips out the door fast enough to make the cables bulge. And if it can do that on a clunker, what can it do on a real server?
Mmmm. Single machines, do we go for the high-end balanced-and-tuned zSeries or one of SGI's monsters? Clusters... I don't suppose you noticed that the world's most popular search engine is a passel of Linux clusters? I vote for the 512-way server.
Only on tests where the majority of skull-sweat went into selecting a setup that favoured every one of Windows' strengths and suffered from every one of Linux's weaknesses. On a level playing field, or to put it another way, something `resembling reality', IIS's ass is grass.
Er, no? IIS has always been Ring 0, which is why vulnerabilities in it have always been a catastrophe. That hamster has already been whipped to death.
Catching up with Java (Real Soon Now) is supposed to be `far ahead'? Gimme a break!
Speaking of buttloads of RAM, that's the way to make Samba really party. It uses about 20-30% more RAM per instance for file sharing than Windows 2000, but OTOH can serve about 10x as many connections on the same hardware. So which would you rather do, pay for ten servers, ten sets of licences and figure out how to integrate them? Or pay for one server and double the RAM?
PS/2 is PS/2 is PS/2; you can plug two PS/2 mice or two PS/2 keyboards into a standard machine, diddle with the drivers to tell them about the alternate IO and IRQ addresses and you're away. You cal also get multi-PS/2 port adapters and a lot of the "high-speed serial port" adapters are close enough to 5V RS232 (which is what a PS/2 port is) to work.
That's not the problem. PS/2 keyboards (all keyboards by default) are routed through the console driver. Hit a key on any keyboard, it goes into the console. There is no way to distinguish which keyboard it came from. You cannot have multiple console drivers (although I believe there is a patch for it). You can, however, tell the console not to bind the USB keyboards, leaving them free for XFree86. Later versions of XFree86 are able to deal with USB keyboards directly (don't recall whether you need a patch or not) rather than through the console.
Not sure if Xfree86 has been fixed yet to not treat all scroll-wheels on each mouse as one, but as at about a year ago, it didn't distinguish. If it has, this here AOpen Optical OpenEye Wheel Mouse model O-35G is about to become multi-talented.
The sticky problems are still with the Linux Console code, in that it doesn't really believe in more than one console with keyboard. However, USB keyboards are easier than PS/2 keyboards (easier to not route through the console), and mice are a non-issue (again, do not route through GPM).
As an alt to an X terminal, try a diskless fanless thin-client box. Motium make those with some really special features, but I don't know if the whizz-bang ones are ready for sale yet.
For that to be fair, you'd want to use the latest Linux as well as the latest Windows (that is to say, if you tweak one you should tweak both properly). In which case there is no question but that Windows + IIS will get their collective butts kicked all over the stadium by Linux +Tux + Apache. There are no more rabbits for Microsoft to pull from their corporate hats in this domain, but Linux whupped them good last time I looked (some dizzying number of requests per second on a single-CPU P2-500), and is only just getting up to stride.
So... you think you shouldn't be dumping it just for breaching standards, being slow, sucking resources and Being Written By Microsoft With Malice Aforethought?
I personally don't think you should be dumping it because of a few security holes, I think you should be dumping it for having lots of security holes.
And so...? The TCO study was provided by IDC, using carefully chosen figures as expressed by Microsoft when they requested the study. The name on the shi8ngle doesn't matter nearly as much as who actually provided the impetus.
Think back to MindCraft, where they did things like using 4x100Mb cards instead of 1x1000Mb like anyone else would, and spreading Windows' logging across four partitions. Sure, they finally managed to tweak good results out of it (and of course the Linux kernel was so much better in those areas three months later anyway), but who in their right mind is going to set up a production machine like that, or even know that that doing so would make it faster?
Likewise choosing high-end hardware specifically for your TPC and tuning it to within an inch of it's life is not what your average punter is going to do, and only vaguely related to what a good DBA will do. They're going to buy a stock machine, do a stock install, make a few of the more popular tweaks, and run that. In those circumstances, PostgreSQL will comprehensively pound MS SQL Server into the pavement.
Lots more out there, I'm sure... you can easily skip some when your brain melts down after scanning 200 entries and maybe clicking on 50 to see if they're as relevant as they seem. Still much easier than reinventing the wheel yet again.
That depends on who classes what as `enterprise-level'. PostgreSQL eats SQL Server's lunch pretty much across the board except on carefully architected benchmarking systems (think MindCraft and four separate logging partitions). SQL Server is basically only warmed-over SyBase, even though a lot of the later warming-over was supervised by a very competent ex-DEC database bloke.
OBTW, don't trust Microsoft's costings for their competitor's systems as far as you can throw them.
For example, they sponsored a recent IDC study which assumed a software cost of USD$940/1009/340/6609/1390 (networking/file/print/security/web) for each Linux system (real cost $0/0/0/0/0, versus (they claim) $211/$3988/1665/5829/7107 for Windows 2000) - which factor alone made more than a slight difference to the resulting TCO figures. Microsoft-sponsored-IDC quote staffing costs for Linux at significantly higher than those for Windows, where in real life the reduction in headcount brought about by increased reliability prunes those costs to about 1/2 or 1/3. They also quote downtime costs for a Linux-based file server at over $4000 a year and nearly double that for a print server! I could do a complete reinstall from scratch every quarter and still cost less than that.
Being pessimistic and assuming that a Linux server collapses half as often as Windows, and nett staff costs are around half those of Windows, the bottom lines for Linux work out to: Network 64% File 54% Print 55% Security 56% Web 40% of Windows. Quite different to IDC's own conclusions.
IDC also do $3 worth of outsourcing for their WIndows file server. Good trick.
Not arguing, just pointing out that PostgreSQL replicates reliably [PDF] too. (-:
Welcome to Earth, I hope you didn't expect fairness.
I guess SlashDot is Real Life(tm) compatible.
So does PostgreSQL. When they polled their users, four percent claimed to run it on Solaris (4th most popular after Linux, Windows and *BSD).
Have one compatriot using it with great success and singing its praises but have never seen a competitive analysis.
I can. I have, many times.
These lead me to suspect that your implementation was broken. I've never seen them happen.
Hey, what? You can't store dollar signs and garbage in an integer or float, you shouldn't be trying to feed that to your db in the first place! If you want to do that kind of thing, amidst a `live' table is not the place for it: use a temp table and do it properly. Whis is probably why the PostgreSQL people didn't implement it.
I hope you've got a lifespan like Methuselah's, then. PostgreSQL does stored procedures in a variety of languages already. Your post does sound like a BASIC programmer grunting and squealing when presented with a real language that insists on him doing stuff like decalring variables, and has scoping etc, forcing him to do `work' (actually investing in manageability) that he didn't have to do before - at least, not up front and in small doses.
First one in months. I was beginning to wonder whether my karma cap was going to explode. What a relief! (-:
...get a wiretap order on Microsoft and spend some time every day sniffing the bazillion or so inbound reports on what is installed on whose machine.
Boot the installer CD, type `rescue' and run that?
Selecting a minimal Mandrake system (circa 30MB) has been easy for a very long time. Choose individual packages, click on the recycle-looking icon, disable anything you don't like. Look in /root/drakx for a floppy image which will re-run that install for the next machine (you can write that into a CD's boot area if you don't want to boot from floppy).