If there is a clearly defined previous build of Freenet that works well, then we should consider having a more clearly defined distinction between production and stable.
Perhaps 5028 ? I think that's what the web-site encourages average / curious users to download. And it is the most common version I've seen in my logs... no one has suggested limiting (putting a version CEILING) on the 5xxx series - I guess that would be called limited forward-compatibility. And isn't frost (the primary existing client) best supported by that version ?
It is unfortunate that "stable" (in quotes) performance has degraded to what is being reported. I believe this to be mostly due to the unknown results of 5xxx & 6xxx interacting. I've NO EVIDENCE to back this up, it's only my intuition.
It is probably worth considering the separation of the -heavier- development activities from the mainstream curious users. It is also a good exercise, as at some point (maybe 1.0, or 0.8, or who knows?), I think I see a forced split coming. For instance, connection multiplexing, a feature down the road a little ways, will not be compatible with older versions. Sure, the code could support mixed-mode operation, but why waste the resources making it do so? I think the forced upgrade argument is fair, as active users will follow other active users to whatever version works best.
No one will deny that traffic CAN be simulated, by having developers and contributor/assistants, operate a separate network, and tests can be coordinated, scripts can hammer out the traffic... Developers should know best what is likely to adversely affect the "stable users'" network, and can jump lightly into the stable net. But things as advanced as what NGR aims to be (face it, it IS VERY different from what came before) really beg for some simulation to aid in this effort. I suspect that the simplest simulation we can create, simply for the purpose of testing some of these theories, might save a few light bulbs in the end. I know it sounds impractical for many of the finer tunings, but for basic concept proofs, it seems very advantageous. I have begun putting together such a package, just for my own education. I believe that alternate routing algorithms can be well tested on a single strong machine, simulating thousands of nodes worth of traffic. You don't actually have to HAVE the data present. Just keys and nodes and routes, and various algorithms for testing... this could potentially help to analyze some proposed attacks.
This is an extremely complex piece of software-artwork, and my hat is off to those who've invested in it over several years. I hope we see it reach the fruition most of us reading this are hoping for.
Ian is entirely deserving to say "it's all research so far." I commend him for addressing this issue, even if it never becomes a major issue again (fingers are crossed).
-Ken
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
