I have been using Debian since 1999 and I gradually changed from stable to testing to unstable very time after a release.
Since last year, I keep a local mirror of Debian unstable, because I have many systems to update.
The main conclusion I get from using unstable, is that it is not really unstable. The only recurring problem I have had in the past year is that every couple of months the postfix configuration broke upon automatic install, so I had to redo it manually (most of my systems are updated daily through cron).
I am busy investigating selling barebones PC's with a pre-install of Debian, and I must say that the update speed is very large. However, Debian only follows the release cycles of the individual packages.
I think that you need to decide for yourself what features you need, make a snapshot of a local mirror and start testing for feature completeness and to remove bugs.
Updates to this snapshot should be done by manually importing new packages, or by fixing the Debian sources and adding the new package in the snapshot.
This way it should be possible to create Debian based releases in a much shorter timeframe than Debian normally needs.
I think however, if you need to do this, then you need a formal organisation with paid people which can be dedicated to testing and releasing.
The main problem with Debian is that there does not seem to be a dedicated team (paid or voluntary), that has the time and inclination to take this massive snapshot (what, 12000, 14000 packages) and test them all for one release.
Maybe they should only test some subsets which can be released on two or three CD's, release these and provide the rest of the repositories as additional features.
Philips worldwide uses Lotus Notes, despite the fact that they are a premium client of Microsoft.
The reason is that everybodies mail is encrypted.
The decision to change to Lotus Notes was made after it was discovered that the sysadmins could read all mail, also from upper management. With Lotus Notes that is not possible.
Although I already program for five years in Perl, I also really can't stand the noise, especially on datastructures.
However, Perl is not the only language with this deficiency.
Since most languages like Perl, Python and Ruby use hashes as substitutes for records, I also find something like
var = record['fieldname']['field2']
too noisy.
Granted, you can declare a class and add members which are accessible using the '.' notation, but I find it cumbersome.
I work much with large datastructures, and I like it much better the way they are handled in COBOL, ie. predeclare them to have one single point of truth, and all members on all levels accessible by just concatenating them with dots.
I had the rebuild problems also with my Promise SX6000 controller, but strangely I had it more in the beginning, and when I used ReiserFS on my RAID-5 system (running Red Hat 9).
The same goes for upgrading the controllers firmware.
It seems that both companies make the most money of a market based around small companies who want to give themselves the air of professionality by implementing their own storage solutions.
It is called good coding practices and it consists of design, implementation, testing and feedback.
A good testing program is easy to find, it is called Nessus.
When you test, you should be able to feedback information about the testing to the developers through a change management mechanism, call it problem reporting or trouble tickets, it does not matter.
The status and severity of these problem reports should be followed closely, so as not to have outstanding critical issues anymore upon release time.
So it seems that there are many places at MS where the feedback loop goes wrong :
Bad or non-existing testing practices
Problem reporting feedback
Follow-up on outstanding problems
You make the mistake of thinking that a team working on the network stack has to comb through all 300 million lines of code. I think that even at Microsoft there is some partitioning between teams and OS features.
The fact that this mistake is repeated is a sure sign that there is not enough attention at MS about the way they write software, ie. the processes.
Gary Kildall was a mathematics teacher, but he worked very closely with the Intel engineers from the 4004 and 8008 teams.
His first OS attempt was something for a 4004 based board.
Re:Pentagon Wars -vs- Inmates Running the Asylum
on
QA != Testing
·
· Score: 1
I also think that feature creep originates in the fact that someone somehere does not have enough knowledge of the problem domain and/or program, to see where a given feature fits, ie. can it be done with reusing existing features or does it open a new category which has to be carefully added to the original ?
One of the first things that Hitler did when he came to power, was to ban the possession of weapons by citizens.
This is something that pisses me of when certain members of political parties want to ban gun ownership here (in Belgium).
I am on a middle ground for gun ownership. I had to do my military service, so I have a used a real weapon (though for target practice only, I haven't even had assault training).
I think that everyone who wants to own a gun, should be compelled to serve some time in the military (be it some weekends or some months), and that there should be a administration setup for this purpose.
It has to do with safety and cost. These are most easily met by this setup.
Safety = human safety and equipment safety.
If you want to run 220V (110V) into expensive equipment, you have to design for it, which makes the equipment more expensive, and should something happen, your expensive equipment can get badly (completely) damaged. Ever wondered why PC power supplies are built in sturdy inox casings ?
Human safety is reached by providing an isolating transformer, and by stepping down the voltage to the equipment, you can design the equipment cheaper.
It is possible to shrink the transformer, but then you have to step up the AC frequency. For that you will need a complex circuit using thyristors (or MOSFETS) and an extra (small) transformer to power the thyristor (MOSFET) controlling circuit. This will bring up the cost again.
I haven't seen it mentioned in the whole discussion, but a low voltage normally provides larger current.
One of the examples above mentioned 5W, 1A. If your cable is not thick enough or too long, you will have a serious voltage drop at the end of the line, unless you can sense the voltage there and feed it back to the regulator.
That is why 48V is used. It is not extremely dangerous (DC is more dangerous than AC because of elektrolytic effects on the blood) and the current through the wire will be less, which means a lower voltage drop, which means you can reach out farther.
I've had Firefox come in at 700Mb, after a couple of weeks of continuus usage.
Seriously, these browsers seem to have big gaping memory leak. I do not mind using them, because I only use Linux, but I don't like such behaviour.
OT : I see the same problem with X. I noticed on my portable, after logging in and out a couple of times, that every time the memory usage of X has increased by about 100Mb.
I also have a PC configured as X workstation, and there is also a gradually increasing memory usage, I have to reboot it every ten days.
Slashdot Mirror
I have been using Debian since 1999 and I gradually changed from stable to testing to unstable very time after a release.
Since last year, I keep a local mirror of Debian unstable, because I have many systems to update.
The main conclusion I get from using unstable, is that it is not really unstable. The only recurring problem I have had in the past year is that every couple of months the postfix configuration broke upon automatic install, so I had to redo it manually (most of my systems are updated daily through cron).
I am busy investigating selling barebones PC's with a pre-install of Debian, and I must say that the update speed is very large. However, Debian only follows the release cycles of the individual packages.
I think that you need to decide for yourself what features you need, make a snapshot of a local mirror and start testing for feature completeness and to remove bugs.
Updates to this snapshot should be done by manually importing new packages, or by fixing the Debian sources and adding the new package in the snapshot.
This way it should be possible to create Debian based releases in a much shorter timeframe than Debian normally needs.
I think however, if you need to do this, then you need a formal organisation with paid people which can be dedicated to testing and releasing.
The main problem with Debian is that there does not seem to be a dedicated team (paid or voluntary), that has the time and inclination to take this massive snapshot (what, 12000, 14000 packages) and test them all for one release.
Maybe they should only test some subsets which can be released on two or three CD's, release these and provide the rest of the repositories as additional features.
Even with less habitants. When I was looking for a new laptop last year (September), there was already much choice in laptops with AMD64 processors.
I have an AMD64 laptop, and with cpufreqd installed I can get 2 hours of run time.
Mind you, my screen is always at its brightest, and maybe the power management could be better, and I also installed 512 Mb RAM extra (up to 1 Gb).
And the other 2/3 are probably spam...
Philips worldwide uses Lotus Notes, despite the fact that they are a premium client of Microsoft.
The reason is that everybodies mail is encrypted.
The decision to change to Lotus Notes was made after it was discovered that the sysadmins could read all mail, also from upper management. With Lotus Notes that is not possible.
Although I already program for five years in Perl, I also really can't stand the noise, especially on datastructures.
However, Perl is not the only language with this deficiency.
Since most languages like Perl, Python and Ruby use hashes as substitutes for records, I also find something like
var = record['fieldname']['field2']
too noisy.
Granted, you can declare a class and add members which are accessible using the '.' notation, but I find it cumbersome.
I work much with large datastructures, and I like it much better the way they are handled in COBOL, ie. predeclare them to have one single point of truth, and all members on all levels accessible by just concatenating them with dots.
Just some of the things I am doing with my system (which has BTW, also 2 Gig RAM).
I had the rebuild problems also with my Promise SX6000 controller, but strangely I had it more in the beginning, and when I used ReiserFS on my RAID-5 system (running Red Hat 9).
The same goes for upgrading the controllers firmware.
It seems that both companies make the most money of a market based around small companies who want to give themselves the air of professionality by implementing their own storage solutions.
It is called good coding practices and it consists of design, implementation, testing and feedback.
A good testing program is easy to find, it is called Nessus.
When you test, you should be able to feedback information about the testing to the developers through a change management mechanism, call it problem reporting or trouble tickets, it does not matter.
The status and severity of these problem reports should be followed closely, so as not to have outstanding critical issues anymore upon release time.
So it seems that there are many places at MS where the feedback loop goes wrong :
You make the mistake of thinking that a team working on the network stack has to comb through all 300 million lines of code. I think that even at Microsoft there is some partitioning between teams and OS features.
The fact that this mistake is repeated is a sure sign that there is not enough attention at MS about the way they write software, ie. the processes.
Do you mean AI, the movie.
There is an older SF story, unfortunately I do not remember the title, nor the author, about a computerised teddy bear.
Gary Kildall was a mathematics teacher, but he worked very closely with the Intel engineers from the 4004 and 8008 teams.
His first OS attempt was something for a 4004 based board.
I also think that feature creep originates in the fact that someone somehere does not have enough knowledge of the problem domain and/or program, to see where a given feature fits, ie. can it be done with reusing existing features or does it open a new category which has to be carefully added to the original ?
One of the first things that Hitler did when he came to power, was to ban the possession of weapons by citizens.
This is something that pisses me of when certain members of political parties want to ban gun ownership here (in Belgium).
I am on a middle ground for gun ownership. I had to do my military service, so I have a used a real weapon (though for target practice only, I haven't even had assault training).
I think that everyone who wants to own a gun, should be compelled to serve some time in the military (be it some weekends or some months), and that there should be a administration setup for this purpose.
You have have guns and rifles which are very bad at being weapons : compressed air guns with tiny lead bullets.
It has to do with safety and cost. These are most easily met by this setup.
Safety = human safety and equipment safety.
If you want to run 220V (110V) into expensive equipment, you have to design for it, which makes the equipment more expensive, and should something happen, your expensive equipment can get badly (completely) damaged. Ever wondered why PC power supplies are built in sturdy inox casings ?
Human safety is reached by providing an isolating transformer, and by stepping down the voltage to the equipment, you can design the equipment cheaper.
It is possible to shrink the transformer, but then you have to step up the AC frequency. For that you will need a complex circuit using thyristors (or MOSFETS) and an extra (small) transformer to power the thyristor (MOSFET) controlling circuit. This will bring up the cost again.
Yes, and there is a reason why they use 48V.
I haven't seen it mentioned in the whole discussion, but a low voltage normally provides larger current.
One of the examples above mentioned 5W, 1A. If your cable is not thick enough or too long, you will have a serious voltage drop at the end of the line, unless you can sense the voltage there and feed it back to the regulator.
That is why 48V is used. It is not extremely dangerous (DC is more dangerous than AC because of elektrolytic effects on the blood) and the current through the wire will be less, which means a lower voltage drop, which means you can reach out farther.
Mod this one Funny, :-)
Yeah, and he should not forget the fire extinguisher as he tries to build such a thing.
I have been in a financial environment where they used Oracle on HP/UX (and IBM DB/2, and even Bull).
Too late! I use them already at home!
This certainly shows a difference between the New World and the Old World.
It seems that here in Europe almost everybody uses manual transmissions, while in the USA almost everybody uses automatic transmissions.
I have never heard any woman who knows how to drive, complain about manual transmissions.
apt-get install alsa_utils -y
alsa_setup
Works a charm...
I did it Saturday evening on my laptop.
I've had Firefox come in at 700Mb, after a couple of weeks of continuus usage.
Seriously, these browsers seem to have big gaping memory leak. I do not mind using them, because I only use Linux, but I don't like such behaviour.
OT : I see the same problem with X. I noticed on my portable, after logging in and out a couple of times, that every time the memory usage of X has increased by about 100Mb.
I also have a PC configured as X workstation, and there is also a gradually increasing memory usage, I have to reboot it every ten days.
There is also cattle and pigs.