Cachet Technologies

1
Cachet Technologies

• Technology Overview
• February 1998

2
Outline

• Our Technology Focus
• First Application Caching at the Edge of the
  Internet
• Comparison with Competing Products
• Some Details of the Technology
• A new hashing protocol
• Other Potential Applications
• Current Plans

3
Technology Focus

• General
• Development of applications and infrastructure
  for distributed networks such as the Internet.
• Current Emphasis
• A solution to the Resource Location Problem in
  distributed networks.

4
Resource Location Problem

• Where to put it and how to find it in a
  rapidly changing environment where there is no
  centralized control and information is imperfect.

5
First Application

• Construction of large distributed caches at the
  edges of the Internet
• Keeps commonly accessed items close to users
• Results in improved performance (e.g. latency)
  for users and decreased cost (e.g. bandwidth
  utilization) for service providers
• Can be sold to service providers and/or user
  groups (e.g. companies) with 10-10M members

6
Example
CNN
MIT
Users
ISP
Internet
IBM
7
Example
CNN
MIT
Users
ISP
Internet
IBM
Cache
8
Different Approaches

• Proxy Caches (Microsoft, Cacheflow)
• Monolithic Caches (Cisco, Inktomi)
• Large caches made up of clusters of machines
  under centralized control and placed at a
  specific point in the network
• Distributed Caches (NetCache)
• Large cache made up of loosely coupled machines
  that are spread throughout the network

9
Monolithic vs. Distributed

• Need faster response time for users
• Caches should be close to the users at the edges
  of the network (Distributed)
• Need reduced traffic out of network
• Cache should have large capacity (Monolithic?)
• Cachets Vision Distributed caches that
  cooperate to form a cache with large aggregate
  capacity.

10
Necessary Components

• Resource Location
• Where do I put it and how do I find it?
• Resource Discovery
• Who else is out there that will cooperate with
  me?
• Preventing Hot Spots
• How do I cooperate without being overloaded by
  requests?

11
Our Key Technologies

• Invention of Consistent Hashing to solve resource
  location problem
• Discovery of protocols that enable caches to
  learn about the network
• Development of a novel replication strategy to
  prevent hot spots
• i.e. when to make more copies?

12
Our Advantages

• No special purpose hardware
• No centralized control
• Very loosely coupled network of caches
• Perfect information on status of caches is not
  needed
• Fully scaleable and fault tolerant
• Automatic load balancing
• Prevents swamping and hot spots

13
Some of the Difficulties
1
Where do I put file X?
2
5
X
3
4

• Standard Way Hashing h(x) axb mod n

14
Standard Hashing The Good

• Balanced load
• Fast to compute
• Local control

15
Standard Hashing The Bad

• Perfect information is needed

axb mod 4 2
16
Standard Hashing The Bad

• Perfect information is needed

17
Standard Hashing The Bad

• Perfect information is needed

18
Standard Hashing The Bad

• Perfect information is needed

19
Standard Hashing The Ugly

• Small changes in the system cause drastic changes
  in the cache

Caches
C
B
A
Pages
1
2
3
4
5
6
7
8
9
10
11
20
Standard Hashing The Ugly

• Small changes in the system cause drastic changes
  in the cache

Caches
D
C
B
A
Pages
1
2
3
4
5
6
7
8
9
10
11
21
Standard Hashing Summary

• Useless for large and/or evolving distributed
  networks

22
Consistent Hashing

• Consistency under changing ranges

Caches
C
B
A
Pages
1
2
3
4
5
6
7
8
9
10
11
23
Consistent Hashing

• Consistency under changing ranges

Caches
D
C
B
A
Pages
1
2
3
4
5
6
7
8
9
10
11
24
Consistent Hashing Features

• Continuous balanced load
• Fast to compute
• Local control
• Scales without any catastrophic updating
• Users need only partially correct view of the
  network. Enables lazy resource discovery.
• Enables resource location to take into account
  load balancing and geographic information.

25
Related Applications

• Eliminating hot spots on the web (same technology
  on larger scale)
• Distributed file servers
• Swamping protection as a service
• Distributed e-mail servers
• On-line movies

26
Current Plans

• Implement Demo 1.0 at LCS
• Software package installed by user
• Program integrates automatically into
  loosly-coupled distributed cache
• Distributed cache performs as if a large virtual
  cache existed between LCS and the Internet
• Development of market strategy and business plan
• Secure VC funding and identify CEO for company