planning www caching strategies'

1
planning www caching strategies.

• NATO workshop, Kaunas 21.05.98
• Ingrid Melve, UNINETT

2
overview.

• web cache systems
• software
• location of servers
• policy
• documentation

3
web proxy cache.

• local copies of web objects
• special servers
• reduce traffic
• byte savings
• number of connections
• speed up retrieval
• most documents are local of far away

4
truths.

• networks carry a lot of web traffic
• bandwidth is limited
• web caching is good

5
actions

• provide web cache services
• use cooperative web cache systems
• ensure fault tolerance
• network topology and routing policies are
  important

6
software criteria.

• several servers co-operate
• scalable by at least 100
• fall back gracefully

7
squid.

• freeware
• cooperative servers
• lots of functionality
• ongoing development
• support for ICP
• available for UNIX only (today)

8
other proxies.

• NetCache
• UNIX and WindowsNT
• appliance box
• Cisco (transparent caching)
• Mirror Image
• Border Manager FastCache, Novell
• CacheFlow
• Netscape, Microsoft, IBM, Sun

9
clients.

• configure clients to use cache
• Netscape Automatic Proxy Configuration
• Netscape
• Internet Explorer
• central configuration
• (transparent proxying)

10
cache location.

• close to traffic flow
• close to users (first level)
• your side of bottlenecks
• where more networks join together
• close to your Internet connection
• network topology is important
• routing policy is important

11
mesh.

• access control
• ask before you use a proxy
• cost of being sibling is small (?)
• keep track of siblings and parents
• contact your Internet Service Provider

12
policy.

• do not violate routing policy
• observe Acceptable Use Policy
• keep logs unavailable
• register your server
• contact your ISP if you have outside siblings
• peering agreements with other providers

13
top level hardware.

• dedicated proxy servers
• UNIX-based
• special hardware (expensive)
• farm of several servers

14
local hardware.

• insufficient may worsen situation
• PC with UNIX
• memory is important for speed
• enough disk
• good I/O

15
statistics.

• keep logs unavailable (unless anonymized)
• log files grow large, process often
• beware of privacy laws
• decide what information you need
• bandwidth, bytes saved
• connections, hit rate
• time, latency savings

16
ongoing work.

• most academic networks have a web proxy service
• Squid development, NLANR
• TF-CACHE, TERENA
• DESIRE

17
DESIRE, web cache.

• caching requirements and alternatives
• demonstrate for caching schemes
• cost/benefit analysis report
• building web cache mesh

18
savings on bandwidth.

• small first level cache 3 Gbyte/month
• estimate best case (256 kbps) 4.5 Gbyte/month
• large first level cache 7 Gbyte/month
• estimate best case 35 Gbyte/month
• top level cache (12S 3L) 30 Gbyte/month
• estimate best case 90 Gbyte/month

19
TF-CACHE objectives.

• deployment of caching hierarchies in Europe
• link these with initiatives in the US and Europe
• help those starting a caching service with
• documentation
• information
• assistance

20
documentation

• http//wwwcache.ja.net/
• http//squid.nlanr.net/
• http//ircache.nlanr.net/
• http//www.uninett.no/prosjekt/desire/arneberg/