On Wed, 29 Jun 2005, Bill Cox wrote: > To summarize some of the issues identified so far: > - Need to deal with high tracker overhead > - Need to download just headers efficiently > - Need to deal with small file updates efficiently > - Need to be optimized for lots of small files instead of one big file > - Need to be aware of electrical distance between peers to find good peers In order for swarming to be efficient, clients also need to keep files around long after they have finished using them, and clients need to stay connected long after they have completed their downloads. So, BT-yum can no longer delete rpms after installing them, and BT-yum needs to hang around connected to the tracker for some time (hours? days?) after completing updates. In theory one could have an "always-on" swarming updater running 24/7 in the background (this would have other benefits such as instant updates) but lots of people might not like the idea of their machine eating up bandwidth 24/7. To me a BT-yum looks less and less attractive, and a yum with parallelized downloads looks much better. -Dan
