To be updated :
sharing same files, > 700 MB, UL limit = 30kB/s, correct UL bandwidth margin
Stats :
Runtime : 2d1h54m
UL limit 30 kB/s
UL sessions : 1257
Successfull UL sessions : 1076
Failed UL sessions : 181
Average Uploaded per session : 4.70 MB
Average UL time : 24'12"
Verbose log :
% completed sessions : 44 % IC95% [39.80-48.11] (84% total UL)
avgul / completed session : 8.19 MB
Stats :
Runtime : 2d11h14m
UL limit 30 kB/s
UL sessions : 1556
Successfull UL sessions : 1355
Failed UL sessions : 201
Average Uploaded per session : 4.53 MB
Average UL time : 14'40"
Verbose log:
% completed sessions : 38 % IC95% [34.42-41.08] (85 % total UL)
avgul / completed session : 7.25 MB
- Xtreme 7.2 :
Ul slot limit 5 kB/s
Stats :
Runtime : 2d1h51m
UL limit 30 kB/s
UL sessions : 1757
Successfull UL sessions : 1653
Failed UL sessions : 104
Average Uploaded per session : 3.00 MB
Average UL time : 17'37"
Verbose log:
- MorphXT v11.3 :
no PBF, no friend slot, no PS, no slot limit, don't remove spare trickle slots, slotfocus
Stats :
Runtime : 2d1h45m
UL limit 30 kB/s
UL sessions : 1734
Successfull UL sessions : 1421
Failed UL sessions : 313
Average Uploaded per session : 3.46 MB
Average UL time : 15'17"
Verbose log:
% completed sessions : 31 % IC95% [28.65-33.05] (86 % total UL)
avgul / completed session : 8.58 MB
Stats and emule verbose log.
Graphic : % completed sessions (with IC 95%)
the numbers are
relative (maybe depending on shared files, didn't assess variance/file) but order/behavior may not change.
Why completed Ul session ? because they represente 85 to 90% of total UL meaning remaining sessions are kind of "wasted UL session" mainly used to adjust UL or simply closed for connection pb. And when you waited 12 h to get a slot you don't want your slot to be open and closed 50 s later to adjust UL... globally it creates a part of random queue (whereas basically it based on waiting time and credits). Random means 50 % longer... and 50 % quicker, but usually we always focus on longer DL, that's a human bias.
-morphXT seems to be very sensitive to UL /opens slots too quickly :
- few/no trickle slots : opens new slots (mostly quickly close) to offset an UL drop (yellow bars in graph)=>
spike then focus ... then spike because UL focused on too few slots
graph =
cycle spike/focus
In the end there are much more interrupted/failed sessions for morph. Stats are not different with no slotfocus (maxclient data rate 3 or 5kB/s).
a big spike in verbose :
-xtreme seems very steady, a long and important drop in UL for xtreme doesn't trigger any burst of UL slots
-ZZUL has a more efficient Ul : it keeps trickle slot open longer, morph quickly close them => clients "in stand by" have more chance to complete their session
-emule 049.c : baseline 8 slots vs 10-11 for Xtreme -> some small "UL slots burst" i can't see in Xtreme + longer UL session (8.19 vs 4.69 MB), but i bet main difference comes from bursts and not from session lenght.
This post has been edited by Anoxie: 30 April 2009 - 12:59 PM