E-Locus - Institutional Repository of the University of Crete

Home Search

Results - Details

Search command : Author="Πρατικάκης" And Author="Πολύβιος"

Current Record: 15 of 82

[Add to Basket]

Identifier

000456799

Title

Utility-driven performance management over a shared resource pool

Alternative Title

Διαχείριση απόδοσης με βάση συναρτήσεις χρησιμότητας πάνω από κοινόχρηστους πόρους

Author

Κουρού, Γαρυφαλιά Λ.

Thesis advisor

Μαγκούτης, Κωνσταντίνος

Reviewer

Μπίλας, Άγγελος
Πρατικάκης, Πολύβιος

Abstract

Autonomic computing systems aim for self-adaptation and self-management, typically using a decision-making process to comply with user-defined service goals. In this thesis, we propose a scheme to effectively provide utility-driven fair sharing of resources with explicit quality-of-service (QoS) targets. We describe Sprinkler, a two-level architecture for achieving performance QoS guarantees for latencysensitive clients over shared resource pools. Sprinkler utilizes a higher-level controller to achieve fair sharing of aggregate system throughput while a lower-level scheduler simultaneously achieves explicit per-client latency targets in concurrent access to a shared pool of resources. The lower-level scheduler isolates each client from others that may exceed their load specifications (average throughput and/or maximum burst size) as decided by the upper-level controller. A distinctive feature of Sprinkler is that controller allocations are advisory, namely a client may decide to exceed them hoping to leverage excess capacity from reduced demand of other clients. Such speculative, non-complying clients can benefit from spare capacity, if it is indeed available, but only hurt themselves if not, as Sprinkler protects compliant clients from violating their latency targets under overload conditions. Sprinkler makes clients aware of their latency metrics at any point in time, so that they can decide if they can afford risking latency violations. We implement a prototype of Sprinkler and demonstrate its effectiveness in fairly regulating per-client throughput while providing latency guarantees in experiments with synthetically generated workloads under workload variations.

Language

English

Subject

Differentiated services

Fairness

Isolation

Quality of service

Απομόνωση

Διαφοροποίηση υπηρεσιών

Δικαιοσύνη

Ποιότητα υπηρεσίας

Issue date