Lecture Notes for Client-Server Interfaces

18 February 2003 - Client Concurrency

Client Problems

• My iterative client is slow.

  • It makes a sequence of n server calls.

  • The server S_i request-response completes in R_i time.

  • Client execution time must be at least R₁ + R₂ + ... + R_n.

• My iterative client is unreliable.

  • Server S_i has a failure rate of 1 out of every r_i requests.

  • The client failure rate is no better than 1 out of min(r₁, r₂, ..., r_n).

• Better client structure.

  • Iterative servers tend to mush phases together.

    • Control and data flow can be left implicit.

  • Concurrent access in each phase requires clearer separation between phases.

  • Not good concurrent design is much worse than poor iterative design.

• Operational flexibility - the parts rather than the whole.

Client-Side Concurrency

• Response time is limited by the request-response model.

• Greater reliability by taking advantage of server replication.

• Structure - good for design, good for extending the request-response model

• Operational flexibility.

Uses of Client-Side Concurrency

• Fast service.

• Wide service.

• Probabilistic protocols - leader election; reliable distributed distribution.

Asyncnronous Server Interactions

• Each server interaction is syncnronous.

• Independent servers are asynchronous.

This page last modified on 25 February 2003.