On Sun, 07 Oct 2001, David McNab wrote: > Hi all, > > I've been doing some heavy testing of fcpputsite on 0.4 (today's CVS), and > have hit a severe snag > > After inserting about 240 MB worth of 256k splitfiles, my node goes totally > catatonic - It just stops sending back responses to the outstanding ClientPut > messages. > That's 960 inserts. that's a lot of inserts. If it's a big file, I'd recommend upping the splitsize.
> What's strange is that it seems to fail at the same point every time (give or > take about 5MB). > If it's a bunch of smaller files, then after you restart the node, start with the next file that's not been inserted. > It works perfectly up till that point, then seizes up abruptly. > Yes, there's something in the node/java that just gives up after a period of prolonged beating. I'd like to know what it is, but in the meantime, us operators have to restart their nodes every couple days. > More, I repeated the test on a transient node on my Windows box, and all the > inserts, 1GB worth, go through fine. > I guess the windows jvm was able to handle it better (assuming it's the JVM). What JVMs did you use? > I had been inserting at htl=3, but always sooner or later hit a constant > stream of RouteNotFound messages, so I cut it to htl=0. > > It is even at htl=0 that I get the problems described above. > Inserting at HTL=0 is useless. it only puts the data on your local node (with some small chance that your node will forward the request even though it's HTL 0). The way freenet works, you should insert with HTL at least 5 or so so that it can get to where the network thinks it should be kept, because that's where the network will look for it. Files inserted at HTL=0 will have to be found literally randomly, by the requester having a really high HTL. That's my simple explanation of freenet routing. > From what I can tell, fcpputsite gets held up waiting for some kind of > response from ClientPut. > Even after hours, no response comes. Not even those beloved 'Pending' > messages. > > Also, I've tried a couple of times with a fresh datastore (created by > changing the size in freenet.ini), but the same thing keeps happening. > > I'm wondering if I should build a timeout mechanism into fcpputsite, so that > if a response from ClientPut doesn't come back within (say) one hour, the > thread gets killed and a new thread re-sends the ClientPut. > We could try to make fcpputsite work around whatever bug this is, but optimally we should fix the bug. (At least identify if it's a JVM thing or if it's a problem in our code.) > Thoughts please? > > David > My closing thought is that you were doing around 1000 inserts yesterday. I have a feeling this is directly related to the netstorm that was going on then. How many simultaneous inserts were you doing? And shouldn't the network be able to handle 1000 inserts? It can't be just the size of the network, I think we should definitely put some time into reworking routing so that it happens much faster. This will ease the strain on the network by not making early nodes in the chain have to keep the resources busy waiting for the response. Thelema -- E-mail: thelema314 at bigfoot.com If you love something, set it free. GPG 1536g/B9C5D1F7 fpr:075A A3F7 F70B 1397 345D A67E 70AA 820B A806 F95D -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 232 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20011007/1a5a0fe9/attachment.pgp>
