Hi folks, I'm trying to setup some infrastructure to verify whether certain Freenet builds actually improve fetch performance over time (or not). Luckily Freenet already provides some methods to generate statistics about fetching performance. This should give some substance to the claims of some vocal users that "build N was much better than build N+1".
By comparing samples from different versions and doing some statistics we can actually determine whether the samples from a specific Freenet version are actually worse or better than a previous version. Having more people contribute data will allow us to know _faster_ whether a new version is an improvement or not. How you can help: 1) daily (preferably at the same time) run the class "freenet/node/simulator/LongTermManySingleBlocksTest". It should be included in the freenet jar I think. This will generate a CSV file with current and previous fetch results and inserts. 2) You can guarantee that tampering with the fetch results is impossible by removing the actual CHK keys. You can use sed to do this [1]. I prefer to receive results which don't contain the actual full CHK keys. 3) Upload the resulting CSV file to Freenet or make it available at some 'regular' URL where I can pull it from. Nextgens (IRC) made his simple bash script available to insert it automatically into Freenet [2] 4) Tell me the link or Freenet USK key such that it can be combined with other samples. Please provide the time (GMT+0) at which you start each daily test (max 2). 5) 'Profit' by having your data included at: USK at sCzfyxndzgtfzoPRUQExz4Y6wNmLDzOpW04umOoIBAk,~X5tIffdcfNWPIKZ2tHgSGPqIk9au31oJ301qB8kTSw,AQACAAE/fetchpull/14 (yes the layout sucks) Please spread the word on FMS/Freetalk/Sone/Flip/... as you see fit. Thanks in advance, Thomas [1] sed "s/CHK@[^\\!]*/CHK/g" [2] http://code.bulix.org/r7llb5-78863?raw
