search

dat-ecosystem / comm-comm

Community Communication
https://comm-comm.datproject.org/
MIT License
89 stars 8 forks source link

20201118 - meeting #178

Closed jwerle closed 3 years ago

jwerle commented 3 years ago

:alarm_clock: Time: 18 Nov 2020, 14:00 America/New_York

Just an hour hanging out and talking about DAT stuff.

:point_right: Post a comment in this github issue with any topics you'd like to cover this week

:link: https://zoom.us/j/181036238 (Native)
:link: https://zoom.us/wc/join/181036238?pwd= (Web)

Timezone Date/Time
Buenos Aires 18 Nov, 4 PM
Toronto 18 Nov, 2 PM
New York 18 Nov, 2 PM
Los Angeles 18 Nov, 11 AM
Berlin 18 Nov, 8 PM

Or in your local time: https://www.timeanddate.com/worldclock/fixedtime.html?msg=Dat+Comm-Comm+Meeting+2020-11-18&iso=20201118T19&p1=1440&ah=1

RangerMauve commented 3 years ago

Notes: https://hackmd.io/a6vn_I5WRjCumBsrdPnuUA

urbien commented 3 years ago

guys, pls share your discord handles if you are on hypercore or DatEcosystem servers

DougAnderson444 commented 3 years ago

@urbien

ipfs-hyper

urbien commented 3 years ago

@DougAnderson444 thanks for this. Let's go deeper. See if you can help me extend this

Feature IPFS Hypercore
Addressing scheme Content-based addressing for individual blocks of data Public-key-based addressing for a group of objects
Addressing stability Low. Dynamic address, changing as block changes (supports deduplication property). IPNS provides data structure address stability but requires constant refreshes High. Static stable address (supports discovery and mutability properties)
Addressing granularity Each block has unique address (block's hash) Address (pub key) corresponds to a collection of objects / files, each consisting of many blocks
Filesystem Low. Files are composed of lists of linked blocks, no notion of file directories, metadata, Great. Files are managed as full-blown filesystem, with API, REST API and Posix-compliant API variants
Storage size efficiency Great. Same block can be reused even between files (Deduplication). But when block changes, old block remains in storage. Low. No block-level dedup. Change in one byte, creates a new version of the file. File-level dedup can be achieved with additional management level, called corestore.
Mutability Low. Originally developed for static content. IPNS attempts to fix that by providing stable hash-based address for the tip of the list of blocks. IPLD?? Great. All data structures, hypercore (log), hypertirie (key-value store) and DB (hyperbee) are built to efficiently edit and propagate changes to peers
Human-friendly naming No. URL contains hash No. URL contains hash
Discovery TBD Dynamic discovery of data structures or peers by topic. Topic is a name in a self-organizing gossiping protocol, derived from Bittorrent (DHT)
Databases TBD. OrbitDB on top of IPFS. Two variants: Key-value store (Hypertrie) and LevelUP compatible (Hyperbee). Provides unique streaming capability to greatly improve storage size and startup, while not requiring DB server. Community-produced replicated DB on top, KappaDB and multi-hyperbee
Direct P2P connections Good. NAT traversal for direct P2P connections (Libp2p library) Good, does not use central nodes, instead leverages DHT (component is called Hyperswarm)
mafintosh commented 3 years ago

@urbien great table!

urbien commented 3 years ago

@mafintosh thanks! I added it to the FAQ and expanded it. Still a lot more work on comparison with IFPS is needed. The FAQ has outlined the areas of comparison that are lacking. For example, they have now 3 databases OrbitDB, AvionDB and ThreadDB and it would be great to review them and compare to what we have with Hyperbee.

@DougAnderson444 - can you help on any of the outstanding points vs IPFS in the FAQ?