Sigcomm'95 Workshop on Middleware

Session 5: Services for Remote Collaboration

Hans-Peter Dommel ( peter@cse.ucsc.edu), UCSC

What does middleware for remote collaboration entail?

Clarification

• Services to orchestrate communication and cooperation across network links with heterogeneous media in stable and failure modes

• Distributed Systems Issues: Replication, Availability, Reliability
• Session control/membership management
• Floor control/concurrency control
• Access control
• Replication/consistency control
• Security (authentication, authorization, encryption)
• Media Synchronization

• Approximation of face-to-face meeting quality (acceptance)
• Fine-grained sharing (media compatibility)
• Collaboration-awareness plus transparency
• Session resilience and adaptation to varying site and network conditions

Should scaling issues be offloaded to the user?

Issues

• Applications and groupware are purpose-specific and scale-dependent
• Sessions needs to allow for recursive partitioning and hierarchical subgrouping
• Problem reflects core issue of middleware: what does application vs. site vs. network take on? (complexity vs. efficiency)
• Large-scale sessions can cause cognitive overload
• Groupware is primarily "userware", i.e. users need to be informed about state
• Scales vary strongly with time, esp. in future "mobile collaboration"

• Make all system scaling issues transparent that do not affect the users' work
• But session scale must be visible for users as feedback on collaboration dynamics and service delays
• Reduce information overload in large sessions via "smart" GUI
• Make protocols adaptable and learning about network behavior

Are reliability and scalability incompatible goals?

• Analogy: how reliable is the (large-scaled) Internet?
• Reliability in terms of distributed systems: the expected time of continuous accessibility of some data
• Scalability can be defined in terms of geographic distance or number of participants or variety of collaboration tools in use...
• Basic tradeoff: reliability vs. complexity
• Complexity increases with scale, and so does resiliency cost
• Large networks are heterogeneous and unreliability has many origins
• Scalability presupposes a certain degree of reliability
• Conclusion: this is primarily a matter of computational and implementational cost

What does "real-time" mean for collaborative services?

• Groupware is interactive and employs time-critical data streams (video, audio)
• Tradeoff: Losslessness and lossiness vs. (non)real-time delivery
• Real-time is a different issue for wide-area collaboration, heterogeneous media, and dynamic membership
• Schedulability constraints are different from non-interactive scheduling
• Keeping control protocols simple and fast allows for minimization of control message overhead
• Quality-of-Service for groupware must include notion of real-time service