aMule Forum

Please login or register.

Login with username, password and session length
Advanced search  

News:

We're back! (IN POG FORM)

Author Topic: proof of no good algorithm?  (Read 9124 times)

jackiszhp

  • Newbie
  • Karma: 0
  • Offline Offline
  • Posts: 4
proof of no good algorithm?
« on: August 24, 2010, 04:01:48 PM »

Hi,
all,

I am not good with this, sorry to take your time to look at this.
From the following diagram, can we say that the sharing algorithm is not good.
Why so many people don't have the same trunk? There are so many people, so every trunk should have the same availability.

 In fact, I found this very long time ago, but people don't believe it.


Logged

Stu Redman

  • Administrator
  • Hero Member
  • *****
  • Karma: 214
  • Offline Offline
  • Posts: 3739
  • Engines screaming
Re: proof of no good algorithm?
« Reply #1 on: August 24, 2010, 04:23:29 PM »

The screenshot shows that almost everybody has the file complete. Green is what you already have, black is what you are missing. If nobody had it as you think it would be red. So there is no problem.
Logged
The image of mother goddess, lying dormant in the eyes of the dead, the sheaf of the corn is broken, end the harvest, throw the dead on the pyre -- Iron Maiden, Isle of Avalon

HolyMan

  • Newbie
  • Karma: 1
  • Offline Offline
  • Posts: 2
Re: proof of no good algorithm?
« Reply #2 on: July 02, 2012, 04:15:24 PM »

First post on here so sorry if I say something stupid. I am just in the process of transferring from E-Mule 0.50 to aMule 2.3.1.

But if  iunderstand jackiszhp right he has  a point here. But he may be barking up the wrong tree.
Point is that I noticed that very often some chunks are very rare while others are abindant among downloaders. Of course that is to blame on the software they use (very often marked as E-mule 0.48 from China). But I noticed that E-Mule (not aMule) had a behaviour where downloads from different sources tend to follow eachother. If 1 sources starts chunk #6 for instance the other follows and starts to upload #6 too. Of course this means that download is biased toward more common chunks. And if the "follower" source has a more rare chunk available it wastes it's time helping to upload the more common chunk.

Obviously this is neither in the common good of the community nor in the private interest of me. Because this way rare chunks stay rare and if a source of a rare chunk decides to quit we all stay behind with an unfinishable file.
This was one of the reasons why I am transferring to aMule now and if I witness the same behaviour I will definitely put this in the new features forum (if it isn't already there).
With more elaborate ideas about the chunk selection algorithm.
Logged

Stu Redman

  • Administrator
  • Hero Member
  • *****
  • Karma: 214
  • Offline Offline
  • Posts: 3739
  • Engines screaming
Re: proof of no good algorithm?
« Reply #3 on: July 02, 2012, 10:40:37 PM »

The chunk selection is probably fine. It's the downloader who determines which chunk is transferred by the way.

What is happening with rare chunks is simply:
- a file is shared, people start downloading / sharing it
- after some time people start moving completed file out of share
- those who were late jumping on the wagon have no full sources available
- if they are unlucky there are some chunks left nobody of the downloaders has
- they share what they have among them; after a while they all have the same without the missing chunks
- there appear to be some full sources maybe, but if it's someone on a slow connection sharing thousands of files nobody may get a download slot

In most cases you get the file completed. It may take months but it works out most of the time.
Logged
The image of mother goddess, lying dormant in the eyes of the dead, the sheaf of the corn is broken, end the harvest, throw the dead on the pyre -- Iron Maiden, Isle of Avalon