Towards Faster Similarity Search by Dynamic Reordering of Streamed Queries
| Authors | |
|---|---|
| Year of publication | 2018 |
| Type | Article in Proceedings |
| Conference | Transactions on Large-Scale Data- and Knowledge-Centered Systems XXXVIII |
| MU Faculty or unit | |
| Citation | |
| Doi | http://dx.doi.org/10.1007/978-3-662-58384-5_3 |
| Keywords | stream processing; similarity search |
| Description | Current era of digital data explosion calls for employment of content-based similarity search techniques, since traditional searchable metadata like annotations are not always available. In our work, we focus on a scenario where the similarity search is used in the context of stream processing, which is one of the suitable approaches to deal with huge amounts of data. Our goal is to maximize the throughput of processed queries while a slight delay is acceptable. We propose a technique that dynamically reorders the queries coming from the stream in order to use our caching mechanism in huge data spaces more effectively. We were able to achieve significantly higher throughput compared to the baseline when no reordering and no caching were used. Moreover, our proposal does not incur any additional precision loss of the similarity search, as opposed to some other caching techniques. In addition to the throughput maximization, we also study the potential of trading off the throughput for low delays (waiting times). The proposed technique allows to be parameterized by the amount of the throughput that can be sacrificed. |
| Related projects: |