Michael Rogers wrote: > Jano wrote: >> Here are these results. I'd like to hear your comments, since the clear >> LIFO advantage is curious (mind you, in some cases it more than doubles >> the other techniques) and maybe it's a simulator artifact. > > Very interesting... even accounting for the large amount of variation > between runs, LIFO seems to far exceed the throughput at which FIFO > collapses (with no flow control in either case - there's no clear > advantage to using LIFO if you already have flow control). I've > experimented with LIFO instead of FIFO in some simulations of a flooding > protocol and it seems to work well there too.
See my new simulation below; I think this time I've managed to really break something in the simulator. >> Do we know what's >> the typical route length in these simulations? I expected the multi-hop >> thing to ruin the intuitively good performance of the single-hop case. > > I haven't measured the route length I'm afraid; in fact I'm not sure > it's well-defined for all searches - what's the route length for an > insert that leaves data at multiple branches of the search tree? Ummm. Each branch is counted as a success? All of them must succeed? Just the longer one? Is this nonsense? See this new run. Some points about it: * I'm dropping messages at the tail when queues reach 50.000 messages queued (for search and transfer queues). I implemented this in the hope of getting rid of OOMs. I'm getting them anyway, so I've screwed something in the process. I could only simulate up to 30 with lifo queues and this change; see the graph. I don't think it's correct. Have we some idea on what is the theoretical maximum throughput for the simulated network, as currently defined? * I'm counting just remote successes. If we are measuring the load balancing performance, I don't think the local hits are of any interest and could mask the remote ones. * I'm not computing failures anymore since messages dropped by far exceed successes. Half an hour of simulation would produce near 1GB of logs, given the amount of msgs dropped! This is something I'd like to come back later, if I resolve the current apparent misbehavior. * I'm not seeing the clear win in throughput of backoff over the others in the fifo case when going to 50, as appeared in your last graph. Can it be attributed to the averaging you did? It's possible we're not using the same revision? I'm on 11135. -------------- next part -------------- A non-text attachment was scrubbed... Name: lifo-fifo-all-1-50-total.png Type: image/png Size: 7435 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/tech/attachments/20061212/b1659431/attachment.png>
