Design and implementation of vsStor: A large-scale video surveillance storage system
Guoliang Liu, Jianjun Xu, Yuanjing Feng, Guoquan Lu · 2018
An ever increasing number of installed surveillance cameras, higher network bandwidth requirements and larger storage space consumption are major considerations when designing a video surveillance storage system. Traditional filesystems are not tailored for receiving and storage of large scale video streams. This paper presents vsStor, a PB-scale, network-based video surveillance storage system which is used for storing thousands of streams. We use Linux's epoll facility to handle large concurrent connections, and employ a two-level metadata approach to manage large amount of metadata: the video's coarse timestamps (one-hour time span) are recorded in database, and the fine timestamps (seconds to tens of seconds) are stored on HDDs. Another major characteristics of vsStor is the scale-out architecture design which allows the performance to grow by adding more hardware. Performance tests demonstrate that the system delivers an aggregate bandwidth of 800 MB/s for writes and 900 MB/s for reads in a highly concurrent environment.