Namn: Mehdi Amirijoo
E-mail: g-meham@ida.liu.se
Avdelning: SaS/RTSLab

Datum: Måndagen, 30:e september.
Tid: 13-15 (13.15-15.00)
Lokal: Donald Knuth

20 poängsexjobb

Språk: Engelska

Titel:
Algorithms for Managing QoS for Real-Time Data Services using Imprecise 
Computation

Sammanfattning:
Lately the demand for real-time data services has increased. Applications used 
in manufacturing, web-servers, e-commerce etc. are becoming increasingly 
sophisticated in their data needs. In these applications it is desirable to 
process user requests within their deadlines using fresh data. However, the 
workload of the databases cannot be precisely predicted and, hence, the 
databases can become overloaded. As a result, many deadline misses and freshness 
violations may occur. To address this problem we propose a QoS-sensitive 
approach to guarantee a set of requirements on the behavior of the database, 
even in the presence of unpredictable workloads.  Our approach is based on 
imprecise computation, where it is possible to trade off resource needs for 
quality of requested service. Imprecise computation is applied on both data and 
transactions. A QoS specification is given in terms of data and transactions 
quality. We propose two dynamic balancing schemes to balance the workload and 
the quality of the data and transactions, so that a given QoS specification can 
be satisfied. Further, we apply feedback control scheduling to provide 
robustness against unpredictable workload variations. We have carried out a set 
of experiments to evaluate the performance of our algorithms. In our simulation 
studies we have applied a wide range of workload and run-time estimates to model 
potential unpredictabilities. Our approach shows an improvement in performance 
compared to a set of baselines. Using our algorithms, the behavior of a RTDB, in 
terms of transaction and data quality, can be controlled. Further, it has been 
established that our approach is robust against varying workload and inaccurate 
run-time estimates.