WCDP: A protocol for web cache consistency

1
WCDP A protocol for web cache consistency

• Renu Tewari
• IBM Almaden Research
• tewarir_at_us.ibm.com
• Thirumale Niranjan
• IBM Software Group
• niranjan_at_us.ibm.com

2
WCDP Overview

• More that just cache invalidations.
• Updates via refresh
• Different consistency levels
• Scalability
• Object groups
• Distribution hierarchies

origin
Intermediaries
Proxy Surrogate
Single/different admin domain
CDNs, enterprise n/w, ISP
3
WCDP Architecture
WCDP Server
Mirror
Mirror
Strong consistency
HTTP/TCP
Delta consistency
WCDP Gateway
(optional protocol translation)
Distribution hierarchy (optional)
WCDP clients
4
WCDP Design Features

• Deployment cases
• Work with CDNs, ISP proxy caches,content mgmt
  engine, server-side software, publishing
  systems.
• Delivery architectures
• Point-to-point, application multicast, gateway
  across protocols and different servers
• Invalidates and updates
• Refresh directives, explicit expiration (default)
• Atomic invalidates
• Identify arbitrary groups of resources for
  consistency
• using invalidation ids and group ids

5
WCDP Design Features

• Scalability
• Group objects, messages and message distribution
  hierarchies
• Different degrees of consistency
• Strong, delta and explicit
• Failure handling
• Server and intermediary failure, network
  partition
• Security
• Plug-in based, default rely on SSL.
• Subscription support
• Explicit content group based or implicit based on
  external SLA

6
WCDP Protocol Example
HTTP request GET www.cnnfn.com
Server cnnfn
HTTP response 200 OK Obj_inv_id
abc123 Content_grp_id financial
cnnfn.com
Register request content_grp_id financial
Register response
Invalidation request Obj_inv_id abc123
Invalidate response
WCDP client
7
Message Types

• Invalidate request/response
• Invalidation action refresh directive, delayed
  refresh, invalidate
• consistency level require commit
• invalidation type atomic, individual
• Register request/response
• Used of subscription
• Join request/response
• Used after failure and recovery
• Commit request/response
• Used for strong consistency
• Heartbeat from server to client
• Used for strong and delta consistency

8
WCDP Message Example

• Invalidation request
• lt!DOCTYPE Invalidate request wcdp.dtdgt
• ltinvalidation sequence_num 100gt
• ltidentifiergt
• ltobj_invalidation_id 12345/gt
• ltobj_inv_group_id 9876 /gt
• lt/identifiergt
• ltactiongt
• ltinvalidate immediate /gt
• ltrefresh yes /gt
• ltdelay 60 sec /gt
• ltforce no /gt
• lt/actiongt
• lttype atomic no /gt
• ltconsistency require_commit no /gt
• lt/invalidationgt

• Invalidation response
• lt!DOCTYPE Invalidate response wcdp.dtdgt
• ltinvalidation sequence_num 100gt
• ltidentifiergt
• ltobj_invalidation_id 12345/gt
• ltstatus OK /gt
• lt/identifiergt
• ltidentifiergt
• ltobj_inv_group_id 9876 /gt
• ltstatus OK /gt
• lt/identifiergt
• lt/invalidationgt

9
Object Invalidation Identity

• Invalidation can address
• Full-pages, fragments, groups
• HTTP response include invalidation identifier
• obj_invalidation_id
• Opaque value with no semantic meaning
• Same obj_invalidation_id for different variants
  of object
• Cache maintains internal mapping
• Obj_invalidation_id ? obj_cache_id
• Obj_cache_id same as external URL for static
  pages
• Includes the HTTP header tags for fragments

10
Object Grouping

• Objects in a group addressed as a unit for
  consistency maintenance or message scalability
• HTTP response includes group identifier
• obj_group_invalidation_id
• E.g., all objects in a sub-directory
• Atomic invalidates
• Specify lock semantics to invalidate all objects
  in a group atomically
• Implements mutual consistency
• Individual invalidates
• Each object in group treated separately

11
Content Groups and Subscription

• Content group
• Large aggregation of objects (by topics)
• Intermediary can subscribe to a content group
• Implicit subscription
• External admin step
• Explicit subscription
• Use a register message
• Discovering content groups
• External LDAP directory
• In the HTTP response

12
Consistency Support

• Explicit
• Default
• similar to HTTP cache-control, expires tag
• Strong
• Two phase commit using invalidation acks and
  commit messages
• Requires heartbeat
• Delta
• Invalidation with ack
• Bounded message delay
• Requires heartbeat
• Mutual
• Atomic invalidate of a group of objects

13
Strong Consistency Example
WCDP Server
Invalidation request Obj_inv_id abc123 Immediate
Refreshyes Consistency require_commit yes
HTTP request GET www.cnnfn.com
HTTP response 200 OK Obj_inv_id
abc123 Content_grp_id financial
Invalidate response Statuswaiting_for_commit
Commit request obj_inv_id abc123
Commit response Status OK
WCDP client
14
Failure and Recovery

• Missing heartbeat from server (server failure)
• Client reverts to explicit consistency
• Send join message to re-subscribe
• Missing invalidation response from client (client
  failure)
• Remove from subscription group after timeout
• No further heartbeats sent
• Network partition
• Client behaves as if server has failed
• Server behaves as if client has failed
• Reliable transport is assumed

15
Status and Future Work

• Based partly on the content distribution
  framework of the Websphere Edge Server version
  2.0
• SOAP based implementation in future
• Security enhancements to be added