Optimally reduceing library size
 

On my box, sharing out too many files causes LimeWire 3.2.1 to bog down and eventually die of what looks like CPU starvation. It's definately fine with 6500 files and decidedly otherwise with 7500, so I try to keep it around 7000.

Now, being what I hope is a good gnut, I take care to share files that others have kindly allowed me to download (unless that is they're of very poor quality or fraudlent), and so my shared file collection gradually grows over time.

I've adopted the strategy of removing the most popular downloads in the collection that have been there for more than two weeks on the basis that they're likely to reside elsewhere in the mesh by then. After pruning, the box still uploads at 100% usage with a constantly full queue, but the uploads tend to be spread over a slightly larger proportion of the remaining files. Ergo, more variety but less capacity for 'heavy demand files'.

Can anyone with a better understanding of gnutella than myself can critique and comment. Is this a sensible strategy or is the network better served by culling rare, low-demand files to provide greater capacity for popular ones ?

Hmm...Not quite sure. I'll comment tho. I would say that you should take out, not heavily demanded files, but popular files. For example, anything by a pop artist. Chances are, the mesh has anything that is supremely popular (I would say top 40, but since no has the license to that one, I can't say it). However, some files that may be heavily requested to are heavily requested because chances are the file is not well established in the network. I would share those files, and ditch the popular ones. Chances are, if people download File A (that's rare) 100 times, only a very very small percentage of those 100 people will reliably share the file again back into the network. I'm not saying everyone is a leecher, but I am saying that not everyone is hooked into the network 24/7 like some of us, so chances are that you haven't shared any of your rare (but often dled) files enough to reach a threshold where they can be easily accessed.

For instance...one of my most popular dls is some music made by a long deceased Argentinean artist. No one cares about this guy, or knows of him. However, I easily upload them at least a couple times a day. And I've been reliably uploading them for months. However, I still get huge numbers of hits for them, so I don't think that my uploads have established the file into the gnutella mesh, for if I had, then I would be getting fewer hits.

Anyone have any comments on this?