Linearizability: a correctness condition for concurrent objects
ACM Transactions on Programming Languages and Systems (TOPLAS)
Designing algorithms for distributed systems with partially synchronized clocks
PODC '93 Proceedings of the twelfth annual ACM symposium on Principles of distributed computing
Sequential consistency versus linearizability
ACM Transactions on Computer Systems (TOCS)
A theory of clock synchronization (extended abstract)
STOC '94 Proceedings of the twenty-sixth annual ACM symposium on Theory of computing
Making operations of concurrent data types fast
PODC '94 Proceedings of the thirteenth annual ACM symposium on Principles of distributed computing
Observable clock synchronization extended abstract
PODC '94 Proceedings of the thirteenth annual ACM symposium on Principles of distributed computing
Optimal Clock Synchronization Under Different Delay Assumptions
SIAM Journal on Computing
Efficient, Strongly Consistent Implementations of Shared Memory (Extended Abstract)
WDAG '92 Proceedings of the 6th International Workshop on Distributed Algorithms
Trade-Off Results for Connection Management
FCT '97 Proceedings of the 11th International Symposium on Fundamentals of Computation Theory
Trade-offs between message delivery and quiesce times in connection management protocols
ISTCS '95 Proceedings of the 3rd Israel Symposium on the Theory of Computing Systems (ISTCS'95)
Implementing hybrid consistency with high-level synchronization operations
Distributed Computing
Hi-index | 0.00 |
The cost of using message-passing to implement linearizable read/write objects for shared memory multiprocessors with drifting clocks is studied. We take as cost measures the response times for performing read and write operations in distributed implementations of virtual shared memory consisting of such objects. A collection of necessary conditions on these response times are presented for a large family of assumptions on the network delays. The assumptions include the common one of lower and upper bounds on delays, and bounds on the difference between delays in opposite directions. In addition, we consider broadcast networks, where each message sent from one node arrives at all other nodes at approximately the same time. The necessary conditions are stated in the form of "gaps" on the values that the response times may attain in any arbitrary execution of the system; the ends of the gap intervals depend solely on the delays in a particular execution, and on certain fixed parameters of the system that express each specific delay assumptions. The proofs of these necessary conditions are comprehensive and modular; they consist of two major components. The first component is independent of any particular type of delay assumptions; it constructs a "counter-example" execution, which respects the delay assumptions only if it is not linearizable. The second component must be tailored for each specific delay assumption; it derives necessary conditions for any linearizable implementation by requiring that the "counter-example" execution does not respect the specific delay assumptions. Our results highlight inherent limitations on the best possible cost for each specific execution of a linearizable implementation. Moreover, our results imply lower bounds on the worst possible such costs as well; interestingly, for the last two assumptions on mesage delays, these worst-case lower bounds are products of the drifting factor of the clocks and the delay uncertainty inherent for the specific assumption.