Also, it is important for 0.8 that we have some concrete answers from Vive on the 3 issues assigned to him at the Freenet Conference: 1. Selective randomization. 2. Targeted swapping. 3. Opennet nodes and swapping.
IMHO the correct order here is 3, 1, 2: we need to be sure that we can turn off opennet swapping and not have a massive impact on a hybrid network; this would probably solve most of the churn-related data persistence/reachability problems we have now, yielding a massive performance gain. On Tuesday 22 July 2008 21:15, Matthew Toseland wrote: > After discussing with Ian and Nextgens, these are the priorities for the near > future: > > DB4O: > This is mostly working ... but it's been mostly working for quite a while now, > and the bits that break are often quite serious. One problem is we tend to > get constant garbage collection for long periods, and even OOMs. And that's > just for requests. Inserts are harder, because the current insert code sucks; > I may end up rewriting that. Inserts also mostly work. Freesite inserts > haven't been tested; and there are a small number of vital features still to > implement (e.g. deleting stuff when it's finished with!). Most work at the > moment is on optimisation and on debugging the catastrophic results of > optimisations. :) There are some "heavy" optimisations (i.e. optimisations > which will take a lot of work but may be major improvements), which I have > mentioned on the list recently, but hopefully they won't be necessary at > least not in the near future. All in all, it should be feasible. If it is > feasible, it should be merge quality within at most a couple of weeks. > > AUTOMATED TESTING: > Before we enable FOAF routing, or even before we merge the branches, we should > have some regular testing on the real-world network going on. This would mean > running a node-based tester every day, and recording the results (probably > reporting them to a mailing list). The 3 tests currently proposed: > - Attempt to connect to all the current seednodes. Report which are online, > along with ping times. Nextgens has done most of the work for this. > - Bootstrap a new node. Measure the time taken to reach some number of peers. > - Insert random data to a node. Bootstrap a new node, measure the time taken > to fetch the data from the new node. > > FOAF ROUTING: > Friend-of-a-friend routing should dramatically improve performance: in theory > it should give us a gain on the order of halving average request path > lengths. However, we need to be able to quantify this in practice, so the > code (which nextgens has written) cannot be enabled until we have regular > testing going on. Also, some major security issues have been raised: because > an attacker can spoof any number of bogus locations, and because he knows our > peers' locations and can find out our peers' peers' locations, he can divert > either all traffic to himself, or most local traffic plus some remote > traffic. Several possible solutions have been proposed, I will elaborate them > on the other thread. > > The above, plus merging sdiz's new datastore, plus randomized splitfile > encryption, sorting out the current seednodes problems, some more work on > streamlining installation and bandwidth limiting, plus a hell of a lot of > debugging, should put us in a good position for 0.8.0 late this year or early > next year. > -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 189 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20080722/4ea3a7fe/attachment.pgp>
