Difference between revisions of "2017 Winter Project Week/IPFS NoSQL Combination"

Revision as of 22:02, 12 January 2017

Home < 2017 Winter Project Week < IPFS NoSQL Combination

Objective	Approach and Plan	Progress and Next Steps
Evaluate IPFS / NoSQL combination for MIC databases	Evaluate IPFS' PSK feature for "private clouds" Build prototype for scanning images put images / files into IPFS put metadata into NoSQL database (ElasticSearch is what we used at FME, CouchDB is what Steve used in Chronicle) Build prototype for browsing / showing images should update live when images appear in the DB should fetch image data from IPFS	IPFS stability and status 0.4.4 provided as binaries (recommended installation, very simple) 0.4.4 does not play well with up-to-date FUSE for macOS yet performance when listing directories not available on current machine is not good (several seconds delay) Performed several experiments that showed disappointing performance (maybe MIT wifi related?), but eventually successful transfers QmPyXW927iBPHVk3hfyzXAPGDpup26WGEh4LYK6da2xMhA is a TCGA subdirectory transferred to several project week participants' computers Discussed interesting deduplication idea: Can original DICOM, anonymized DICOM and e.g. Nifty files share data blocks? Answer: Yes, there's a `--chunker` argument to `ipfs add` which defines the chunking algorithm. Apparently, the rabin chunker should already perform well, without any particular knowledge about the file formats, but it does not seem to be the default algorithm. (Valid arguments include: 'rabin' 'rabin-[avg]' or 'rabin-[min]-[avg]-[max]' with integer parameters.)

IPFS homepage https://ipfs.io
"Private Networks - First Implementation" issue: https://github.com/ipfs/go-ipfs/issues/3313
Chronicle: https://github.com/pieper/Chronicle
Also reconsider if BitTorrent or similar technologies can help here
- http://academictorrents.com/
- https://webtorrent.io/
PIP package: pip install ipfsapi

@@ Line 35: / Line 35: @@
 * Performed several experiments that showed disappointing performance (maybe MIT wifi related?), but eventually successful transfers
 * QmPyXW927iBPHVk3hfyzXAPGDpup26WGEh4LYK6da2xMhA is a TCGA subdirectory transferred to several project week participants' computers
+* Discussed interesting deduplication idea: Can original DICOM, anonymized DICOM and e.g. Nifty files share data blocks?
+** Answer: Yes, there's a <code>--chunker</code> argument to <code>ipfs add</code> which [https://github.com/ipfs/faq/issues/214 defines the chunking algorithm]. Apparently, the rabin chunker should already perform well, without any particular knowledge about the file formats, but it does not seem to be the default algorithm. (Valid arguments include: 'rabin' 'rabin-[avg]' or 'rabin-[min]-[avg]-[max]' with integer parameters.)
 |}