Approximate Query Processing over Secure Key/Value Stores

点击翻译按钮获取中文摘要

Secure key/value stores can be used to store a large amount of data in a central place while providing cell-levelaccess control to protect data privacy (i.e., each key-value pair has its own security label). This is often veryattractive to many domains (e.g., healthcare, government) that have a strong requirement for data security andprivacy. However, a key limitation that stops secure key/value stores from being widely adopted is that theycannot support analytical queries very efficiently.To overcome the limitation, in this proposal, we will collaborate with PHEMI to build an approximate queryprocessing (AQP) engine over secure key/value stores. The key insight is to allow users to run analyticalqueries on a sample data and return them approximate answers with error bars. Since the queries only need tobe executed on a sample, the query response time can be significantly improved. We will work together withPHEMI to address two challenging research problems: (1) how to efficiently create a random sample of thedata stored in key/value stores; (2) how to efficiently obtain approximate answers from a sample. To the best ofour knowledge, we are the first to explore approximate query processing over secure key/value stores.The PI has many years of successful experience in AQP system development. PHEMI is a big data warehousestartup headquartered in Canada. They use a secure key-value store called Accumulo as the main data store oftheir products. We anticipate that the collaborative research project will not only be beneficial to PHEMI'sproducts but also have a profound impact on how to architect a real-world AQP system.