I got this in my logs: vendor Swapper changed # of open data bytes to 4 The number of "open data bytes" is expected to be 1 (old BS) or 2. LW uses 4 but they...
Unfortunately my I had a massive harddrive failure last week and lost all my old email, so I cannot verify the what I am about to say, but from variable names...
7368
Raphael Manfredi
Raphael_Manfredi@...
Jun 1, 2002 8:25 am
Here is what I get in my logs: GET /get/12/foo.bar HTTP/1.0 User-Agent: Gnut 0.4.26 Referrer: Gnutella file sharing network Connection: Keep-Alive User-Agent:...
... and ... provides ... an ... set ... But the data that you put into the DHT can just be the locations of content -- not the content itself. That was the...
... Actually, the suggestion made by a number of Chord-like systems is very simple and intuitively very credible: the base "node" is considered to be a...
... still ... we'd ... SHA1 values do have the quality you wanted: any subrange of the bits are just as random as the whole thing. It's not too hard to...
Raphael, ... How do you process these extended blocks though? I'm asking because I am wondering about the exact issue here. -- Mike ... Outgoing mail is...
Oso, Having browsed through the Kademlia paper, there is no doubt that XOR metrics and the symmetry that follow have merit in the peer-to-peer context....
... My sense is that the first iteration of most DHT designs does not address the issue of keyword searches -- that is layered on later. In the first pass,...
... Yep, that sounds like an only feasible way of implementing this in practice. Or, at least, I've never seen any other solution that would seem to be...
... ranges ... and for ... MB, 4 MB, ... I disagree; choosing a smaller base size for the tree costs very little, and leaves open more potential applications. ...
First I would again like to thank everyone very much for their input. I have pretty much reached my level of understanding as far as putting the protocol...
... I disagree to your disagreement. The size of the hashed chunks should be proportional to the size of the entire file, and quantized (by powers of two). In...
... I disagree yet again! :) ... Why? Using one small blocksize ensures all possible applications -- even those squeezing small chunks of giant files into...
... of ... What I'm saying is that in a real-world implementation, one must tailor the algorithm to tune it to the data. If you are dealing with files over a...
Note that the ideal minimum block size is not 1KB ! This is too small, and non optimal in performance. The best minimum is the system page size or a multiple...
... Isn't it the same cost: to compute 1 million hashes of 1 kilobyte of data to compute 1 thousand hashes of 1 megabyte of data ? I can see why the block size...
You're missing the point. Paging or drive sector sizes are irrelevant. No one has advocated (for Gnutella) 1KB as the unit size for transferring data. Nor have...
... scale ... operate ... it ... curious ... In the Gnutella case its not the operations, its the storage cost for the partial hashes. A file should have X...
... There is some good practical sense in that. ... This is where you and Gordon disagree. He claims that the hashing tree should be built based on 1K base ...
... for ... constant. ... Admittedly, I'm not particularly knowledgable about the Tiger hash or the details for building the hash tree. As for the hashing tree...
... You get this effect naturally if you keep the topmost X values at the root of the tree, since your "leaves" will then be the aggregated hashing of many 1K...
I just thought that there's a big increase of performance between computing 1 milllion hashes based on 1KB blocks, and 256K hashes based on 4KB blocks. Most...
... computing ... blocks. Excellent, this illustrates my point regarding block size exactly. If you are computing a million hashes for a single file, something...
So you would create immediately a 2MB file once you select it for download ? What do you do with parallel download attempts of different files, with similar...
... The 1kb base block size is just an internal thing of the TigerTree algorithm. These small sub-hashes are quickly combined into higher levels, and the ...
... download ? Yeah I guess. I was thinking out loud :) In any event, I'm not a big fan of partial sharing (as I mentioned numerous times). However, *if* one...
... In this area we agree: the common expectations for how much internal tree "proof" information is shared should be a proportion of file size. (I wouldn't...
