Lecture 7-8: P2P | Notion

💽 Overview

DHTs are a technique for looking up stuff in P2P systems. Ex of DHTs: Napster, Chord, everything else in this lecture

Regular Hash Table	Distributed Hash Table
inserts, deletes, lookups in `O(1)` with key	inserts, deletes, lookups in `O(1)` (not really but they try to be as close as possible) with key
object being inserted is like a `key, value` pair	object being inserted is typically a `key, a whole file`

Challenges

Fault-tolerance, Efficiency, Load balancing
Locality (ex: simply splitting data evenly across servers is not enough because the most popular lookups could sit on the same node → unbalanced traffic)

Introducers + DNS

Necessary if no centralized server because a newly joining node needs someone to contact and be integrated into the network
DNS used to resolve clients to IP addresses → introducers use this information to populate their membership lists

📊 P2P Systems Analysis

	Memory	Lookup Latency	# Messages for lookup
Napster	`O(1)` @ client
`O(N)` @ server	`O(1)`	`O(1)`
Gnutella	`O(N)`	`O(N)`	`O(N)`
**Chord
Pastry**	`O(logN)`	`O(logN)`	`O(logN)`
Kelips	`O(sqrtN)`	`O(1)`

Chord vs. Pastry vs. Kelips
- Chord & Pastry need more messages but less memory, Kelips needs more memory but gives constant lookup time
- With Chord’s consistent hashing, we move around logN files but with Kelips we don’t need to more any files around