I have seen a recent suggestion to implement 'take a number' style cues for downloading .. ie first in first served.

I think this would be bad, for two reasons,

1) modem users usually have time-limits on their connections. Many other users don't (cable, DSL, T1) and I think they would block up the download pipe for popular servers.

These other users also have nice fast connections already .. so people with super duper connections already would be further benefitted while those with already poor connections would find it EVEN HARDER to download anything of note.

2) old clients or anyway those which choose not to support the feature would effectively be blocked by those that do. Which sux something chronic.

Vote for democracy - vote against cues!

Anyone know if this is already implemented? It was in Napster, wasn't it? My only experience of Napster was on cable, so it would be interesting to hear feedback.

You make an interesting point. The use of upload queues would favor those with full time connections. There would be no reason for me not to queue up scores, possibly hundreds, of downloads, with the full knowledge that eventually I'd get them all (particularly if the client queues were persistant across executions of the client). As a dial-up user you'd be at a clear disadvantage, hoping you can stay on long enough so you get to the front of the line. The current system of whoever gets to the open slot first wins, while not exactly orderly, favors no particular connection type.

Perhaps the clients should reserve at least one upload slot for "random" connections.

These problems could be reduced by defining some simple rulesets for the sharing hosts, like:

- don't queue up more than twice as many uploads than the number of upload slots you're providing
- don't queue up anybody if your current uploads will probably last for more than an hour
- don't let a single user dominate your upload queue

Please remember that queueing has some advantages and might help to solve some of Gnutellas current problems. Especially the well-known 'hammering problem' (well-known among those people who share a lot of iles) might be reduced: If a client receives a 'Host Busy' response to a download request, it might try again within the next 30-60 seconds. But if it receives a 'Queue is full' message, it knows that there is no use trying again within the next 15 minutes.

Additionally, queueing might help to make downloading from firewalled hosts a bit more reliable.

Guido