While poking around in the KTorrent UI I noticed that the upload slots did not appear to be distributed in accordance with the score associated with each peer. Upon further investigation I discovered that the peer list is not, in fact, sorted according to each peers' aca_score statistic, but instead appears to be randomly ordered at the time of the call to doUnchoking(). As a result, upload slots are being distributed at random rather than being allocated to the peers with the highest scores, which can only have a negative impact on the efficiency of the network, not to mention end-users' download rates.

The original code attempted sort the peer list by first calling ppl.setCompareFunc(ACACmp) and then qSort(ppl.begin(), ppl.end()). The problem with this is that the two-argument qSort() function uses the operator<() defined for the element-type of the list, in this case Peer*, and thus sorts the items rather uselessly by memory address rather than score.

The following patch against libbtcore/torrent/advancedchokealgorithm.cpp achieves the expected behavior:

Code: Select all

--- advancedchokealgorithm.cpp.orig   2010-01-24 21:50:53.599579271 -0600
+++ advancedchokealgorithm.cpp   2010-01-24 21:51:29.706242125 -0600
@@ -118,12 +118,7 @@
    
    static int ACACmp(Peer* a,Peer* b)
    {
-      if (a->getStats().aca_score < b->getStats().aca_score)
-         return 1;
-      else if (a->getStats().aca_score > b->getStats().aca_score)
-         return -1;
-      else
-         return 0;
+      return a->getStats().aca_score > b->getStats().aca_score;
    }
    
 
@@ -146,8 +141,7 @@
       }
       
       // sort list by ACA score
-      ppl.setCompareFunc(ACACmp);
-      qSort(ppl.begin(), ppl.end());
+      qSort(ppl.begin(), ppl.end(), ACACmp);
       
       doUnchoking(ppl,updateOptimisticPeer(pman,ppl));
    }

You are right, this seems to be a porting error when KT was ported from KDE3 to KDE4.

Will fix it.

Thank you nybble41 for investigating the problem and for the patch. And thanks Georges for ktorrent itself

This bug is actually quite critical. Despite a very massive upload bandwidth, I cannot download much because I keep uploading to the wrong people. This bug deserves its own release (ktorrent 3.3.4?).

I am trying to compile from source, with the patch above applied, because I can't download much until I do, and can't wait for a kubuntu package with the above fix.

goodtorrents wrote:Thank you nybble41 for investigating the problem and for the patch. And thanks Georges for ktorrent itself

This bug is actually quite critical. Despite a very massive upload bandwidth, I cannot download much because I keep uploading to the wrong people. This bug deserves its own release (ktorrent 3.3.4?).

I am trying to compile from source, with the patch above applied, because I can't download much until I do, and can't wait for a kubuntu package with the above fix.

There is probably going to be a 3.3.4 in the weekend, I'm hoping to get that out together with 4.0beta2.