Files
Munteanu_31731200_2017.pdf
Open access - Adobe PDF
- 1.15 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Each day the content generated is exceedingly bigger and the quality of service much more demanding. To cope with this shortcoming, stronger consistency guarantees are bartered for higher availability and lower latency. In this thesis we describe HoneyPot, a protocol that improves the overall performance of the target system at the expense of slight data staleness. Honeypot is a lightweight caching-layer protocol with transactional causal+ consistency semantics. This transactional model offers the strongest consistency guarantees under availability and convergence and insures causal ordering of transaction execution. As a result, HoneyPot improves throughput and reduces latency by adapting data locality regardless to the access pattern and workload. HoneyPot carries out the incoming transactions on locally replicated keys, on behalf of their owner, reducing network usage and preventing the formation of hot-spots in the cluster. The results against the target system show an improvement of throughput and latency proportional to the cache-hit ratio and a small overhead in the worst case scenario.