In anwser to Hal, >> Unless you think for some reason that in a mature Freenet, a large >> fraction of the nodes at any given time will be new, node discovery >> should pretty much take care of itself.
Ian Clarke wrote: >I am not sure that the current node discovery machanisms will be sufficient >without the crutch of the inform.php mechanism. Someone a while back did >mention that they noticed a significant reduction in activity once their >node had fallen off the end of the inform.php list - and it continued to >decrease. This would suggest an over-reliance on inform.php. Let my add my 2 cents of personal observation: I've been running a node 24/24 on a fixed IP for about ten days, and studied a little how my node was "seen" by others, and how many of them it could see. This by checking my log of incoming / outgoing requests. My own node is _not_ registered with inform.php, but got its initial "other nodes" list by reading it. I tried to make my node initially discovered by many others by performing a large number of dummy "inserts" with a high HTL. Some tens of foo/bar inserts ;-) The 3 or 4 bytes "bar" files that you may see on your node are probably mine ;-) I made some inserts at an HTL up to 99, and they almost always worked, which made me think that the active network was indeed large, despite the fact that inform.php is hardly showing more than about 10 active nodes at a given time. After performing those inserts, my node started to be contacted by others from the network. Then I almost stopped performing inserts, and took a look at what was happening. (Although I made some inserts again from time to time...) The following observations were made purely manually on a little spare time, and have no pretention to precision. However, they give an interesting insight on the way the network behaves. During the 10 past days, my node seems to have directly chatted with 22 other nodes. The nodes that my node talks with seem to be more or less always the same, with very little variation. From time to time, but rarely, a new one appears. Many have disappeared after initial contacts, and never came back. Among these 22 nodes, only 7 of them have a DNS entry that makes them look like permanent servers with fixed IPs. About 5 of them have addresses that look semi-permanent to me such as ADSL, cable or the like (in France, the IP addresses given by such connections are usually changed once a week). These permanent or semi-permanent machines are the ones with which my node talks the most. The 10 remaining machines seem to have pure dialup connections, with DNS entries giving the feeling that they are taken from a large pool, and will vary from one connection to the next one. These expected dialup nodes seems to have connections that suddenly end and never come back, which I would expect from dialup machines. So I would say, among 22 machines, 7 being permanent, 5 semi-permanent, and 10 dialup transient nodes. That makes a high percentage of transient nodes, and it has to be correctly dealt with when thinking of a node discovery system. Also, by studying manually inform.php at several different times, and adding this to the nodes known from my own node, I would make totals looking like: 16 machines having permanent DNS entries; 9 machines with no reverse DNS entry (which I assume to be permanent nodes); 10 machines with semi-permanent looking DNS entries; 48 machines looking like temporary dialups; For a total of 83 studied machines Which, if I assume machines with no reverse DNS to be permanent ones (ya know what bad DNS is ;-) would make 35 permanent or semi-permanent nodes for 48 transient ones... Surely this has to be taken into account for Freenet's architecture. However, the figures may be biased, as I expect that there may be a number of "discrete" permanent servers that don't log themselves in inform.php and thus may be less visible than transient nodes operated by people who don't care (or don't know) being shown in inform.php. But, answering to "Hal", there really looks like a good proportion of nodes are "new" ones, that don't stay long... Now the way the system behaves... At least my own node: I don't feel that requests to my node suffer from a reduction in activity over time. Activity is quite low, but stable over time. I observe some temporary bursts of activity, with periods of several hours without a single request. Then activity starts again, and so on. My node gets VERY RARE inserts. Suprisingly rare. About all the new files that come into my node are propagation of files that I was asked for and cached when getting them from another node. My node gets mostly data requests, and about 90% of them fail with timeout (usually returned by the following host to which my node forwarded the query). I was really surprised with this high number of failures. It makes me think that: - Either the people sending the requests don't understand Freenet's principle, and type filenames (even wildcards?) in the blind, resulting in this high proportion of not founds; - Or the system doesn't work so good? ...Well, almost all the requests l personnally made for data listed in one of the "indexes" succeeded, so it looks like it works (at least for me ;-) so I assume the first possibility being the good one... Having a look at the indexes that don't grow much, I feel people do not insert enough data into freenet to have a real perception of the efficiency of the insertion and search mechanism... And, seeing the high number of transient nodes out there make me fear that performing insertions with short HTLs might result in having your insert sent to possibly only transient nodes, thus making your insert mostly unavailable to the "true" Freenet network. Does this supposition make sense? Well, all this rises more questions that it gives answers, but I basically wanted to share these observations with you so it may give you matter for reflexion. Regards, and keep on with this wonderful project. michel at bouissou.net _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
