Hi, I know that there are already a lot of things on the wish list for squeakmap but nevertheless I wanted to add a very little one.
What do you think about keeping a downloads counter per Package release on Squeakmap? I know that it wouldn't be accurate because of the local cache on each client. But it will give as at least a lower bound for measuring the use of a given package.
Regards, Hernán
"Hernan Tylim" htylim@yahoo.com.ar wrote:
What do you think about keeping a downloads counter per Package release on Squeakmap? I know that it wouldn't be accurate because of the local cache on each client. But it will give as at least a lower bound for measuring the use of a given package.
Yeah, it's interesting to track. SqueakSource does this, by the way. If you host your package on SqueakSource, and put the entry on Squeakmap, then you can get download counts.
Note that it's still not a lower bound. I know that I reinstall Squeak with moderate frequency and in multiple directories, and thus I have still downloaded packages multiple times even though SqueakMap has a built-in cache.
Debian has a take on this issue. :) They have a "popularity contest". Users can, at their choice, install the popularity contest package on their machine and thus participate in the contest. If you install the package, then it will periodically contact a central site with a list of the packages you have installed. Using this information, the central site can determine the *percentage* of users who have a particular package installed on their machine. You can then extrapolate from this number to however many Debian users you think there are in the world (not a trivial question itself...), to find out how many people are using any particular package.
It's not perfect. People can rig the votes if they want. A package might be installed on people's machines, but not be getting used. Users who take part in the popularity contest may not be representative of the whole set of Debian users. But, overall, it seems more accurate that simply counting downloads.
-Lex
A related idea: After installing some package via SM this info could transferred back to the SM server (controlled by a user pref). In addition this could be combined with an UUID coupled to each image. So each used image would count once, which seems fair to me (OK, people using hundreds of images for parallel processing could cheat this measure).
Greetings Stephan
lex@cc.gatech.edu wrote:
"Hernan Tylim" htylim@yahoo.com.ar wrote:
What do you think about keeping a downloads counter per Package release on Squeakmap? I know that it wouldn't be accurate because of the local cache on each client. But it will give as at least a lower bound for measuring the use of a given package.
Yeah, it's interesting to track. SqueakSource does this, by the way. If you host your package on SqueakSource, and put the entry on Squeakmap, then you can get download counts.
Note that it's still not a lower bound. I know that I reinstall Squeak with moderate frequency and in multiple directories, and thus I have still downloaded packages multiple times even though SqueakMap has a built-in cache.
Debian has a take on this issue. :) They have a "popularity contest". Users can, at their choice, install the popularity contest package on their machine and thus participate in the contest. If you install the package, then it will periodically contact a central site with a list of the packages you have installed. Using this information, the central site can determine the *percentage* of users who have a particular package installed on their machine. You can then extrapolate from this number to however many Debian users you think there are in the world (not a trivial question itself...), to find out how many people are using any particular package.
It's not perfect. People can rig the votes if they want. A package might be installed on people's machines, but not be getting used. Users who take part in the popularity contest may not be representative of the whole set of Debian users. But, overall, it seems more accurate that simply counting downloads.
-Lex
Hi Lex,
"Hernan Tylim" htylim@yahoo.com.ar wrote:
What do you think about keeping a downloads counter per Package release on Squeakmap? I know that it wouldn't be accurate
because of
the local cache on each client. But it will give as at least a lower bound for measuring the use of a given package.
Yeah, it's interesting to track. SqueakSource does this, by the way. If you host your package on SqueakSource, and put the entry on Squeakmap, then you can get download counts.
I didn't know. Thanks for pointing it out.
Note that it's still not a lower bound. I know that I reinstall Squeak with moderate frequency and in multiple directories, and thus I have still downloaded packages multiple times even though SqueakMap has a built-in cache.
Yes, of course you are right. But even though I would really like it to have it. For example, the other day I published KeyBinder and I wanted to know if there were reaction to the [ANN] on squeak-dev. With this kind of statistics I would had know if people were starting to installing it.
Debian has a take on this issue. :) They have a "popularity contest". Users can, at their choice, install the popularity contest package on their machine and thus participate in the contest. If you install the package, then it will periodically contact a central site with a list of the packages you have installed. Using this information, the central site can determine the *percentage* of users who have a particular package installed on their machine. You can then extrapolate from this number to however many Debian users you think there are in the world (not a trivial question itself...), to find out how many people are using any particular package.
It's not perfect. People can rig the votes if they want. A package might be installed on people's machines, but not be getting used. Users who take part in the popularity contest may not be representative of the whole set of Debian users. But, overall, it seems more accurate that simply counting downloads.
This seems really interesting. Maybe we can start with simple counters and later improve on it. What do you think?
Regards, Hernán
On 16/08/04 13:19, "lex@cc.gatech.edu" lex@cc.gatech.edu wrote:
Note that it's still not a lower bound. I know that I reinstall Squeak with moderate frequency and in multiple directories, and thus I have still downloaded packages multiple times even though SqueakMap has a built-in cache.
Lex:
Another useful thing could be SqueakMap check package-cache directory and rebuilds itself with this.
Now I using one directory for all images and all could use same SqueakMap and do hand installing, but no clue how actualize each image instance info.
We wait Goran returns from China...
Edgar
Hi all!
"Lic. Edgar J. De Cleene" edgardec2001@yahoo.com.ar wrote:
On 16/08/04 13:19, "lex@cc.gatech.edu" lex@cc.gatech.edu wrote:
Note that it's still not a lower bound. I know that I reinstall Squeak with moderate frequency and in multiple directories, and thus I have still downloaded packages multiple times even though SqueakMap has a built-in cache.
Lex:
Another useful thing could be SqueakMap check package-cache directory and rebuilds itself with this.
I am not sure what the above means. the "package-cache" directory is a Monticello thing and has nothing to do with SqueakMap. SqueakMap keeps all its state in the "sm" directory.
Now I using one directory for all images and all could use same SqueakMap and do hand installing, but no clue how actualize each image instance info.
Eh... if you have all images in one directory then SqueakMap uses the stuff in the "sm" directory, regardless of which image you use. So the package cache in there is shared and so is the map itself (meaning that the latest snapshot of the map itself will not be downloaded needlessly for multiple images).
We wait Goran returns from China...
Edgar
Well, I am back now! :) Though with severely limited time, email me in private if you want to be sure to get an answer.
regards, Göran
Well, I am back now! :) Though with severely limited time, email me in private if you want to be sure to get an answer.
regards, Göran
Hi Göran,
What do you think of implementing a very simple download counter for each package release on the map? If help is needed and you or someone can do the server side I can help with the client side of the implementation.
Note that I am not asking for any weird or complex statistic, just a very simply counter to be able to have an idea if a package was installed.
Regards, Hernán
Hi!
Hernan Tylim htylim@yahoo.com.ar wrote:
Hi Göran,
What do you think of implementing a very simple download counter foreach package release on the map? If help is needed and you or someone can do the server side I can help with the client side of the implementation.
Well, how would it work? I mean... normally the server isn't involved at all when installations are made. And the client cache also makes sure installations can be made even if you aren't online.
But... the first thing that crosses my mind is that this is a very simple case of the more "broad" vision of a system I was blabbering about earlier - a publish/subscribe system for "developer events".
The idea was that you set as Preferences what kind of "development events" you would like to broadcast. "Package installations" could be one of those.
Note that I am not asking for any weird or complex statistic, just avery simply counter to be able to have an idea if a package was installed.
Yeah, well... I will see what I can do. I agree it would be nice to have.
Regards, Hernán
regards, Göran
Well, how would it work? I mean... normally the server isn't involved at all when installations are made. And the client cache also makes sure installations can be made even if you aren't online.
Hi Göran,
The feature with I would be more than happy is just an HTTP request counter for the files hosted on the SM Server.
I know that this will not work as an installation counter, only donwload requests, but nevertheless I think this is a good info to have.
I also know that not all the packages are using the new SM file hosting service. But why don't you look at this as a new feature of this service. The packages which use your HTTP server will have this counter, the ones who don't, well, not.
But... the first thing that crosses my mind is that this is a very simple case of the more "broad" vision of a system I was blabbering about earlier - a publish/subscribe system for "developer events".
The idea was that you set as Preferences what kind of "development events" you would like to broadcast. "Package installations" could be one of those.
I like how this sound. It's a lot more of what I asked, but I like it. :)
Note that I am not asking for any weird or complexstatistic, just a
very simply counter to be able to have an idea if a package
was installed.
Yeah, well... I will see what I can do. I agree it would be nice to have.
Great, Thanks
Regards, Hernán
squeak-dev@lists.squeakfoundation.org
-
goran.krampe＠bluefish.se -
Hernan Tylim -
lex＠cc.gatech.edu -
Lic. Edgar J. De Cleene -
Stephan Rudlof