IPFS100 asked the 3rd Q: IPFS what architecture
IPFS architecture is divided into eight layers from top to bottom:
Identity Layer: S / Kademlia generation, peer node identification information generation
Network Layer: any transport protocol, ICE NET & NET penetration
Routing layer: a distributed hash table loose peer location information and storing a subject in need
Exchange layer: How to block BitTorrent & Bit Swap distribution management
Object Layer: Merkel-DAG content-addressable not be altered, to the redundant link object
File level: similar to Git version control file system: blog list tree commit
Naming layers: a SFS (Self-Certified Filesystems) IPNS: DAG variable object naming
Application layer: applications running on IPFS use the nearest node to provide services, improve efficiency and reduce costs
Identity layer and routing layer can be interpreted together. Peer node identification and routing rules generated information is generated by Kademlia protocol developed, KAD essence of the agreement is to build a distributed loose Hash Table, DHT for short, everyone must join the DHT network generate their own identity information, and then to go through this identity information is responsible for storing contact information for this network of information resources and other members.
Comparison of core network layer, LibP2P can be used to support any transport layer protocol. Network NAT technology allows the devices share the same external network IP, we have experienced a home router is the principle.
Exchange layer, BT is similar to the Thunder this tool. Thunder is actually simulated P2P network, and create a central server. This approach has drawbacks, because the server is maintained by the Thunder uniform, if the failure, when downtime, download operation can not be performed.
Service center can also limit the number of download requests, people invented a smarter way to Bittorrent, so that each seed node data to be stored, the service is more stable.
IPFS the BitTorrent team has innovated called Bitswap, it increases the credit and billing system to encourage nodes to share, FileCoin great probability is based on Bitswap, increase user data will Bitswap increase in credit points, the more share The higher the credit points. If the user to only retrieve stored data without data, credit branch getting low, the other node of higher priority selected when embedding the connection of credit.
File object layer and the layer bonded to talk, which was 80% of management data structure IPFS, most of the data objects are present in MerkleDag structure, and this content addressable deduplication facilitated. File level is a new data structure, and DAG parallel, using the same data structures to support Git version snapshot.
Self-verification feature named layer: when other users access the object using the public key fingerprint verification, which verifies the authenticity of the user object's release, and joined the IPNS this clever design to make the DAG object name can be encrypted define and enhance readability.
Finally, the application layer, IPFS core value lies in applications that run on top of, we can use it similar CDN functionality, bandwidth at low cost, to get the desired data, so as to enhance the efficiency of the entire application.
Each protocol stack have responsibilities and collaborate with each other. Eventually be able to improve system efficiency and reduce system cost.
Reproduced in: https: //www.jianshu.com/p/d547efe0b768