SHAWS A Scalable Highly Available Web Server

1
SHAWSA Scalable Highly Available Web Server

• SHAWS Technology
• Presentation
• Shai Guday
• Director Commercial Systems Technologies
• s.guday_at_opengroup.org

2
SHAWSA Scalable Highly Available Web Server

• High Availability Clusters
• SHAWS Technology Overview
• SHAWS Opportunities
• SHAWS II

3
High Availability Clusters

• Existing clusters (e.g. ServiceGuard) fail over
  system level services

4
High Availability Shortcomings

• System dependencies
• System service failover requires system
  dependencies
• Added complexity and expense to both cluster and
  platform programs
• Limited in scalability
• Difficult to scale to large numbers of nodes
• Load balancing is not a well integrated
  functionality
• Ease of use
• Difficult to install and administer
• Users resent having to become system proficient
• Configuring a large cluster is akin to
  programming in assembler

5
SHAWS TechnologyKey Properties

• Scalability
• Scales to scores of nodes
• High Availability
• RT failover
• Load Balancing
• Multiple schemes
• Platform Independence
• Supports heterogeneous clusters
• Ease of Use
• Single point of control, dynamic reconfiguration
• Transparency
• Being a client means never having to see a failure

6
Creating AService Namespace

• An extended service namespace is composed from
  all cluster services

7
SHAWS ServerLoad Balancing

• Distributed probabilistic load balancing
• Failure resilient
• Scalable
• Tunable

8
SHAWS ClientHigh Availability

• Clients receive namespace topology and load
  information
• Client-side software transparently monitors and
  controls connections

9
SHAWS Administration

• Single point of control
• From any cluster node
• Synchronized across cluster nodes
• Dynamic reconfiguration
• Nodes can be added, deleted, or remounted
• New junction maps are faulted in by clients as
  needed
• Client behavior is controlled by the individual
  user
• Aggressiveness of failure determination

10
SHAWSStatus and Availability

• All original deliverables tracked to the original
  schedule (set in June 1996), newer items are
  additional.
• August 1997
• Initial version for HP-UX 10.10
• Integration with latest stable Apache code base
• October 1997
• Initial version for NT 4.0
• Improved GIPC scalability algorithms
• January 1998
• Version 1.0 for NT 4.0, HP-UX 10.20, OSF1/AD

11
SHAWS AdoptionOpportunities

• Low cost solutions for Web service clusters
• No need for proprietary H/W
• Extending NT or UNIX clusters
• Overcoming cluster size limitations
• Platform compatibility
• Migration path from NT to UNIX or vice versa
• Clustered applications
• Applications could benefit from SAFER toolkit
• Deployment of algorithms and design in other
  products

12
SHAWS PossibleFuture Directions

• Removing the need for a client proxy - Underway
• Virtual clusters
• Use of COTS Web Servers
• Integration with ServiceGuard, HACMP, etc...
• Integration with NT Enterprise Cluster Server
• QoS guarantees

13
SHAWS Embedded SLP

• Removing the need for a client proxy
• SHAWS 1.0 required clients to install a proxy
• Both MSIE 4.0 and NN 4.0 can support SLP
  functionality via a combination of ECMAScript
  functionality and applets

14
SHAWS IIA Scalable Highly Available Web Server

• Functional Objectives
• Deliver a Virtual Cluster technology platform,
  supporting geographically dispersed systems.
• Extend the current SHAWS benefits to encompass
  the virtual cluster domain.

15
Virtual Clusters

• A virtual cluster node may consist of
• A single node
• An existing SHAWS local cluster
• Any combination of the above regardless of their
  geographic location
• Scales to scores of nodes and/or clusters
• Provides support for the needs of the truly
  global enterprise

16
Virtual Clusters

• Maintain hierarchical administrative domains
• Enable local cluster administration for sites
• Enable global administration for the enterprise
• Allow enterprises to leverage global resources
• Extended load balancing for Internet topologies
• Clients account for network delays
• Are extremely adaptable
• May leverage multiple platforms
• Can be purchased, deployed, and adminstered by
  individual enterprise business units