Skip to Main content Skip to Navigation

Enforcing Service-Specific Replica Consistency Models and Response Time Requirements for Heterogeneous Replicated Services

Corina Ferdean 1
1 Regal - Large-Scale Distributed Systems and Applications
LIP6 - Laboratoire d'Informatique de Paris 6, Inria Paris-Rocquencourt
Abstract : Replication is largely accepted as a technique to improve
the performance of popular Internet services, especially when they are
accessed simultaneously by many wide-area dispersed clients.
In the context of Replicated Service Hosting Systems, we address the problem of
selecting the replica, that provides satisfactory response time, to
the final clients and the problem of providing the appropriate
consistency constraints for all replicas. Existing work recognizes the
response time, as the metric that correlates directly with the user perceived performance. However, they don't take into account the real resource demands of the services and the current resource utilization.
This is not appropriate when we consider that each service has its specific needs with respect to
the system and the network resources that it uses. %the variety of the service workloads.
Also, the resources are heterogeneous, with respect to their capacities and load. We propose a response time estimation approach, where the
response time is decomposed into independent components (service times
and waiting times), and each component is approximated separately, by
taking into account the capacity and the current utilization of the resources, needed by the service workload. The service supplier expresses the performance he/she wishes to be observed by the clients, by specifying a bound or by requiring the best possible response time.
The response time requirement is enforced by a replica selection protocol, configured with the criterion, defining what is the suitable replica which should perform the clients' requests. We implemented this solution and the results, obtained experimentally, showed satisfactory fulfillment of the response time requirements, at a reasonable cost.

Existing work on replica consistency management offers a limited set of guarantees, for all the replicated services. However, different consistency constraints are pertinent from a service to another. We identify three consistency dimensions: quality of observable state, scheduling and dependency control. We propose a consistency meta-model, which reifies existing consistency guarantees, by identifying the parameters characterizing each dimension.
The meta-model is attached a consistency protocol, which enforces service-specific consistency
models, specified statically by the service suppliers. A
service-specific consistency model contains the safety constraints,
needed for the service to function correctly, when it is replicated,
and the liveness constraints needed for bounding the discrepancy
between the state of a particular replica and the ideal replica
state. The protocol is built up of three independent building bricks,
corresponding to the three orthogonal consistency dimensions. The
main thesis message is that our replication approach was able to manage the heterogeneity
of the services, with respect to their operation semantics and resource demands,
when providing the appropriate replica consistency guarantees and performance requirements.
Complete list of metadatas
Contributor : Corina Ferdean <>
Submitted on : Monday, December 11, 2006 - 12:36:23 PM
Last modification on : Friday, January 8, 2021 - 5:46:03 PM
Long-term archiving on: : Wednesday, April 7, 2010 - 12:22:47 AM


  • HAL Id : tel-00119595, version 1


Corina Ferdean. Enforcing Service-Specific Replica Consistency Models and Response Time Requirements for Heterogeneous Replicated Services. Networking and Internet Architecture [cs.NI]. Université Pierre et Marie Curie - Paris VI, 2006. English. ⟨tel-00119595⟩



Record views


Files downloads