An Architecture for a Secure Service Discovery Service

Czerwinski, et al

service discovery pervasive computing

@inproceedings{czerwinski:mobicom-1999,
  title={An Architecture for a Secure Service Discovery Service},
  author={Czerwinski, S.E. and Zhao, B.Y. and Hodes, T.D. and
          Joseph, A.D. and Katz, R.H.},
  booktitle={{ACM}/{IEEE} International Conference on Mobile
             Computing and Networking ({Mobicom})},
  pages={24--35},
  year={1999},
  organization={{ACM}}
}

Attach:Czerwinski-Mobicom1999.pdf

How to locate a service out of a wide area network of hundreds of thousands of nodes

Services described in ad hoc XML descriptions

Relatively straightforward XML template matching for querying

Differentiates between already running services, and those available on demand

Both push and pull based models

Must manage:

• Network partitions
• Component failures---meantime to failure with so many nodes is low
• Bandwidth limits
• Load balancing among the registries

All soft state

Registries organized into hierarchical structure

• Servers divide into network extents defined by IP address ranges
  • Parents may start new servers if overloaded, restart children if they fail

Significant focus on security, authenticating servers, services, and clients

• Interesting aspect is private services: Only matched to queries from authorized clients

Registries discover each other and clients discover them over global multicast channel

• Advertisement includes local multicast to use to contact that registry

Does not really address how to develop the hierarchical structure

• Mentions it might be done by manual configuration, or external information such as location
• Discusses maintaining multiple hierarchies in parallel to address different top level topic areas, do some gross load balancing

Descriptions are hashed into Bloom filters which are then propagated up the hierarchy, aggregated in the natural way

• Deletions handled by counting, or periodically wiping the table

Other systems:

• Jini matching based on Java object matching, which is not very flexible or efficient, and very prone to mismatches, e.g., minor version mismatches
• DNS gets its efficiency from the hierarchical, unique names structure
• SLP doesn't really address how to scale beyond the LAN, doesn't have a real structure to it