The UK (Academic) National Web Cache
Neil Smith, HENSA Unix, University of Kent at Canterbury
This presentation describes a JISC funded service provided for the
benefit of the Higher Education community in the United Kingdom.
Introduction
- The UK (Academic) National Web Cache
- History
- Software
- Hardware
- Networks
- Users
- Associated Issues
- Proxy resilience
- Cache co-operation
- Networks for Caches
- HTTP developments
- Politics and The Law
History
- November 1993
- Introduced experimental Web Caching service
- Summer 1995
-
Recognised as the UK National Service
About 320,000 requests per day
To about 9,600 distinct sites
From about 2,700 client machines
- 11th March 1996
-
New trans-Atlantic link
A total of 1,650,000 requests per day
To about 20,000 distinct sites
From about 8,000 client machines
Software
- Lagoon - CERN
- First Generation Caches
- Netscape - Harvest
- Second generation. Offering different facilities.
- The Next Generation
- Netscape 2
Netcache/Harvest 2
Hardware
- CPU Evolution
- From Uniprocessor Sparc 10
To Five Challenge S's + One Multiprocessor DM
- Disk
- Disk is always the bottleneck
- RAM
- Netscape Proxy RAM hungry - but depends on bandwidth
- Network
- Distribution across multiple sites
Increases available bandwidth
Improves resilience
Networks
- Back then
- 4Mbps to the US
4Mbps to Europe
- Real Soon Now
- 17Mbps to the US
34Mbps European Triangle
- The Future
- Dedicated bandwidth promised ...
... not yet delivered.
Users
- The Web is a bubble surrounding an unstable house of cards
- A proxy stacks those cards higher
- Users are increasingly unaware of what can go wrong
- The technology is new - Even premature
- In general, systems do not degrade gracefully
- If something does go wrong, of course, it must have been the proxy
You can please some of the people all of the time.
And you can please all of the people some of the time.
But you cannot please all of the people all of the time.
Proxy resilience
- A Client Issue
- Caches have to work around poor browser implementation
Fall-back or round-robin used in FTP and telnet
- Netscape Auto-config
- Introduces resilience and configurability
But depends on Netscape Navigator Version 2
- Cache clusters
- Dumb clients limit capabilities
Round-robin DNS with short TTLs
Need `buddies' to monitor and pick up failed machines
Cache co-operation
- Introduced in Harvest
- Hand crafted co-operatives
Neighbo(u)r and parent caches
Communication through light-weight UDP protocol
- But...
- Need a unifying and complete protocol
Need to recognise clusters of machines
Broadcast queries do not scale
Hand-built co-operatives cumbersome
Networks for Caches
- The Problem
- Caching is crucial
Hits give 0.05 to 0.2 second responses
Misses depend on bandwidth
- The Solution
- Dedicated bandwidth improves miss performance
- The Benefits
- Encourages use of the cache
Potential savings are enormous
Does it warrant blocking port 80?
HTTP developments
- Log exchange
- Log information passed to servers
Encourages cache friendly pages
Privacy issues
- User configuration
- User configures speed vs staleness
Need cache maintainer overrides
Need server overrides
- Protocol efficiencies
- HTTP-NG
HTTP 1.1
Politics and The Law
- Politics
-
- The Law
- Are caches publishers or distributors?
Censorship or promotion?
Copyright?
Lots of untested issues