>why is Apache still spawning processes for every request that comes in... don't they realize the overhead of that??
I'm not sure if you're serious about that or not, however
Since Apache2 you could change multiprocessing modules. The prefork MPM does the equivalent of apache1 and makes a number of forked off worker processes.
Apache1/Apache2-prefork don't use one process per request unless you specifically request it to. Worker processes are retired after a number of requests (usually a couple hundred) as this helps to contain the impact of any memory leakage that may have been accidentally introduced etc
Apache1/Apache2-prefork doesn't wait for a request to spawn the new worker processes, It spawns them in the background and hands incoming requests over to them. An incoming request doesn't have to wait for a worker process to spawn
Though you can argue that windows is reasonably performant in general, one primitive that linux that's far faster is process creation. Processes are meant to be cheap in unix systems, and used as such.
Separating things out into multiple processes helps to contain the effect of any bugs. Worker processes can crash all they want and won't impact the service as a whole.
I tend to use prefork MPM on servers for the isolation, unless I'm expecting a tonne of traffic, but by all means use the worker MPM, which uses a large amount of threads inside a single process.
One thing that I find somewhat counterproductive is that browsers do not save files sent over SSL in their caches.
It's sensible, I suppose, to assume that if something's sent over an SSL channel that it's sensitive and therefore shouldn't be saved, but it would give a speed and bandwidth efficiency hit which would deter usage of SSL for everyday browsing.
You could, of course, have the HTML transmitted over SSL and the supporting images over plain HTTP, but then the browser will scare people by warning that not all content on the page is secure..
I think browsers should start looking at encrypting their cache files, so that stuff such as SSL can be accommodated without breaking caching.
To be honest, I think that's a major loophole in the GPL. You have to be able to prove that the code's been modified before you can compel the company to release it, and that's often a lot easier said than done without the source code (which you won't have if they havn't released it)
Telephone calls use 64KBps. Provisioning for peak times there is relatively straightforward - you just multiply the number of peak users by that figure.
Peer to peer apps, on the other hand, open up multiple connections in order to use all the bandwidth available to them. Additionally, this means that TCP "back-off" mechanisms help less with them than with single-connection apps.
End result of all this is that unless there's massive overprovisioning, p2p apps threaten to fill any pipe the ISP throws at it, and adversely affect other apps while doing so.
We don't really need to go to the lengths of using XML as the storage format.. We just need a standard - Any standard.
One reason, I think, people tend to favour plaintext with formatting conventions is that you need to be able to "hoist" a system - you need the programming language to be readable without elaborate tools because those elaborate tools won't always be available, and will themselves need to be programmed in something. This suggests plaintext as a way of doing things, because you can work with it with only the most primitive tools.
Keeping the whitespace usage convention consistent in the storage format also helps when generating diffs and such - reformatting the code between programmers versions would cause unnecessary delta instructions
So.. We've decided to store stuff in plain text and with a consistent structure in terms of whitespace. What we have not established, however, is the necessity of actually viewing it in this format. Computers are more than powerful enough to show code in formats that take full advantage of the available screen width, use proportional fonts, elastic tabstobs and coloured highlighting and rewrap stuff to fit an 80-column convention on a common storage format when saving the file.
I really don't see why we need to compromise anything here.
Regardless of the nature of the software development team, the nature of competition remains the same.
Stagnation is costly - delaying a product launch drives people to the alternatives (both due to the alternatives updating faster, and due to the lack of progress seen by the outside world).
Of course, buggy software is costly in terms of reputation, as well, so the end question becomes at what point will the delaying of the release cost us more market share then the remaining bugs will.
Unfortunate from a purists eyes, but it's just the way things go.
I was too young to remember ever not having a computer.. but I still have fond memories of its awkward rubber keys, the way you had to hook it up to the hi-fi to load the programs off a tape, and best of all, it's game of frogger.
Strange that in many ways it's the limitations of such machines that you fondly remember..
Dense Wavelength Division Multiplexing isn't so much a protocol as it is an improved method of encoding.
The main obstacle to adoption, as far as I'm aware, is the crosstalk incurred at the amplifiers.
Most fiber-optic connections these says make use of amplifying L.A.S.E.R.s wherein the incident EM photons induce the emission of photons of identical frequency from atoms which are in an energetic state. However, due to the finite power of the pumping source, and the finite population of the atoms used as lasing medium, there can be problems with crosstalk - Transmitting a high level on one frequency depletes the population of energised atoms in the lasing medium and causes the amplification ratio of the other frequencies to drop.
I read a while back about one type of L.A.S.E.R. amplifier where a single frequency was injected transversely to the path of the intended amplified radiation. This would make each frequency have a constant "big" competitor for the energised atoms, and thus drastically decrease the magnitude of this crosstalk.
Personally, I find the concept of mainframe development rather attractive, as I do any architecture substantially different from what I'm used to. I'd really like to get to know how to use and program these machines.
Problem is, I've no idea how to go about this. It wasn't offered as a module at university, and I don't exactly have one lying around I can play with.
I recall reading about how IBM donated a mainframe to an english university (reading? Can't remember) for tuition purposes, but I don't exactly want to take a second degree to go about this.
One thing that strikes me is that backward compatibility on mainframes is legendary (with many programs written for a system 360 still running without modification. This would suggest the use of old machines for training. Would there be any objection to companies donating their retired mainframes to academic institutions for this purpose?
So, I note that a lot of the time, recently, when downloading music files, some of the hosts have the correct version and others the decoy.
This is dangerous, though, because it exposes the possibility of distribution of other types of file (eg application binaries) with shady bits from crackers inserted.
Peer-to-peer networks should look at this as a significant security risk and devise ways around it.. I think bittorrent could be made the most resistant, as there's generally a progenitor host that chunks could be validated from.
We have good filtering methods available now. I find bayes to be very effective, for example.
What I can't get over is the nagging feeling that this whole anti-spam effort is rather unstructured. I'd certainly like to chain together multiple filters, combine their results, and control all of this from within my client, but combining filters like this at the moment involves lashing them together with a script, but then what client plugins there are only control a single filter and probably wouldn't work with a funky scripted filter combination.
Lets define some standard CORBA interfaces for various types of mail filter, and a simple method of connecting them so that mail classifications and corrections can be properly distributed..
That's a good rule for preventing your own mobile phone usage annoying people. However, you have no way of predicting whether your intended recipient is in a similarly considerate place.
Hey, for all you know your recipient might be being held hostage by someone who's snapped and pledged to kill the next person whos damn phone rings.
How can you prevent this? Perhaps text the recipient to ask permission to call, and warn them to move to a better location? Can't see that catching on.
Slashdot Mirror
I'm not sure if you're serious about that or not, however
I tend to use prefork MPM on servers for the isolation, unless I'm expecting a tonne of traffic, but by all means use the worker MPM, which uses a large amount of threads inside a single process.
I'll do it.
One thing that I find somewhat counterproductive is that browsers do not save files sent over SSL in their caches.
It's sensible, I suppose, to assume that if something's sent over an SSL channel that it's sensitive and therefore shouldn't be saved, but it would give a speed and bandwidth efficiency hit which would deter usage of SSL for everyday browsing.
You could, of course, have the HTML transmitted over SSL and the supporting images over plain HTTP, but then the browser will scare people by warning that not all content on the page is secure..
I think browsers should start looking at encrypting their cache files, so that stuff such as SSL can be accommodated without breaking caching.
To be honest, I think that's a major loophole in the GPL. You have to be able to prove that the code's been modified before you can compel the company to release it, and that's often a lot easier said than done without the source code (which you won't have if they havn't released it)
Telephone calls use 64KBps. Provisioning for peak times there is relatively straightforward - you just multiply the number of peak users by that figure.
Peer to peer apps, on the other hand, open up multiple connections in order to use all the bandwidth available to them. Additionally, this means that TCP "back-off" mechanisms help less with them than with single-connection apps.
End result of all this is that unless there's massive overprovisioning, p2p apps threaten to fill any pipe the ISP throws at it, and adversely affect other apps while doing so.
..they're buying the old Concorde airframes and launching them from the US Navy's new railgun?
We don't really need to go to the lengths of using XML as the storage format.. We just need a standard - Any standard.
One reason, I think, people tend to favour plaintext with formatting conventions is that you need to be able to "hoist" a system - you need the programming language to be readable without elaborate tools because those elaborate tools won't always be available, and will themselves need to be programmed in something. This suggests plaintext as a way of doing things, because you can work with it with only the most primitive tools.
Keeping the whitespace usage convention consistent in the storage format also helps when generating diffs and such - reformatting the code between programmers versions would cause unnecessary delta instructions
So.. We've decided to store stuff in plain text and with a consistent structure in terms of whitespace. What we have not established, however, is the necessity of actually viewing it in this format. Computers are more than powerful enough to show code in formats that take full advantage of the available screen width, use proportional fonts, elastic tabstobs and coloured highlighting and rewrap stuff to fit an 80-column convention on a common storage format when saving the file.
I really don't see why we need to compromise anything here.
Personally, I'm just glad that this helps to clarify the species.
Java was, apparently, invented by an immature Canada Goose, as opposed the the Greylag variety that we see more of around here..
If your boss was edsger dijkstra, I'd sue.
All joking aside, you can run 32 bit binaries in the 64 bit versions of a lot of distros.
Generally speaking, though, you'll need the 32-bit broswer in order to use the 32-bit plugin, though.
Now, if only he could defragment his wife.
Yeah, corbomite maneuvre!
The rest of the Enterprise doesn't know about Macs.
But now at least scotty knows to use the keyboard.
Regardless of the nature of the software development team, the nature of competition remains the same.
Stagnation is costly - delaying a product launch drives people to the alternatives (both due to the alternatives updating faster, and due to the lack of progress seen by the outside world).
Of course, buggy software is costly in terms of reputation, as well, so the end question becomes at what point will the delaying of the release cost us more market share then the remaining bugs will.
Unfortunate from a purists eyes, but it's just the way things go.
Personally, I find it far more likely that the dolphins are referring to each other by their slashdot IDs.
My first machine was the ORIC-1
I was too young to remember ever not having a computer.. but I still have fond memories of its awkward rubber keys, the way you had to hook it up to the hi-fi to load the programs off a tape, and best of all, it's game of frogger.
Strange that in many ways it's the limitations of such machines that you fondly remember..
Dense Wavelength Division Multiplexing isn't so much a protocol as it is an improved method of encoding.
The main obstacle to adoption, as far as I'm aware, is the crosstalk incurred at the amplifiers.
Most fiber-optic connections these says make use of amplifying L.A.S.E.R.s wherein the incident EM photons induce the emission of photons of identical frequency from atoms which are in an energetic state. However, due to the finite power of the pumping source, and the finite population of the atoms used as lasing medium, there can be problems with crosstalk - Transmitting a high level on one frequency depletes the population of energised atoms in the lasing medium and causes the amplification ratio of the other frequencies to drop.
I read a while back about one type of L.A.S.E.R. amplifier where a single frequency was injected transversely to the path of the intended amplified radiation. This would make each frequency have a constant "big" competitor for the energised atoms, and thus drastically decrease the magnitude of this crosstalk.
Personally, I find the concept of mainframe development rather attractive, as I do any architecture substantially different from what I'm used to. I'd really like to get to know how to use and program these machines.
Problem is, I've no idea how to go about this. It wasn't offered as a module at university, and I don't exactly have one lying around I can play with.
I recall reading about how IBM donated a mainframe to an english university (reading? Can't remember) for tuition purposes, but I don't exactly want to take a second degree to go about this.
One thing that strikes me is that backward compatibility on mainframes is legendary (with many programs written for a system 360 still running without modification. This would suggest the use of old machines for training. Would there be any objection to companies donating their retired mainframes to academic institutions for this purpose?
Seems a trifle slow..
Coral cacheSite seems a bit slow.. Coralised link.
If memory serves, Adama actually proposes exactly that: Split the fleet into six parts. Doesn't get around to actually doing it, though.
Change the password on your luggage.
So, I note that a lot of the time, recently, when downloading music files, some of the hosts have the correct version and others the decoy.
This is dangerous, though, because it exposes the possibility of distribution of other types of file (eg application binaries) with shady bits from crackers inserted.
Peer-to-peer networks should look at this as a significant security risk and devise ways around it.. I think bittorrent could be made the most resistant, as there's generally a progenitor host that chunks could be validated from.
We have good filtering methods available now. I find bayes to be very effective, for example.
What I can't get over is the nagging feeling that this whole anti-spam effort is rather unstructured. I'd certainly like to chain together multiple filters, combine their results, and control all of this from within my client, but combining filters like this at the moment involves lashing them together with a script, but then what client plugins there are only control a single filter and probably wouldn't work with a funky scripted filter combination.
Lets define some standard CORBA interfaces for various types of mail filter, and a simple method of connecting them so that mail classifications and corrections can be properly distributed..
That's a good rule for preventing your own mobile phone usage annoying people. However, you have no way of predicting whether your intended recipient is in a similarly considerate place.
Hey, for all you know your recipient might be being held hostage by someone who's snapped and pledged to kill the next person whos damn phone rings.
How can you prevent this? Perhaps text the recipient to ask permission to call, and warn them to move to a better location? Can't see that catching on.