Object storage

Object storage (also known as object-based storage) is a computer data storage that manages data as objects, as opposed to other storage architectures like which manages data as a file hierarchy, and block storage which manages data as blocks within sectors and tracks. Each object typically includes the data itself, a variable amount of metadata, and a globally unique identifier. Object storage can be implemented at multiple levels, including the device level (object-storage device), the system level, and the interface level. In each case, object storage seeks to enable capabilities not addressed by other storage architectures, like interfaces that are directly programmable by the application, a namespace that can span multiple instances of physical hardware, and data-management functions like data replication and data distribution at object-level granularity. Object storage systems allow retention of massive amounts of unstructured data in which data is written once and read once (o

Adaptive video delivery using semantics

Olivier Steiger

The diffusion of network appliances such as cellular phones, personal digital assistants and hand-held computers has created the need to personalize the way media content is delivered to the end user. Moreover, recent devices, such as digital radio receivers with graphics displays, and new applications, such as intelligent visual surveillance, require novel forms of video analysis for content adaptation and summarization. To cope with these challenges, we propose an automatic method for the extraction of semantics from video, and we present a framework that exploits these semantics in order to provide adaptive video delivery. First, an algorithm that relies on motion information to extract multiple semantic video objects is proposed. The algorithm operates in two stages. In the first stage, a statistical change detector produces the segmentation of moving objects from the background. This process is robust with regard to camera noise and does not need manual tuning along a sequence or for different sequences. In the second stage, feedbacks between an object partition and a region partition are used to track individual objects along the frames. These interactions allow us to cope with multiple, deformable objects, occlusions, splitting, appearance and disappearance of objects, and complex motion. Subsequently, semantics are used to prioritize visual data in order to improve the performance of adaptive video delivery. The idea behind this approach is to organize the content so that a particular network or device does not inhibit the main content message. Specifically, we propose two new video adaptation strategies. The first strategy combines semantic analysis with a traditional frame-based video encoder. Background simplifications resulting from this approach do not penalize overall quality at low bitrates. The second strategy uses metadata to efficiently encode the main content message. The metadata-based representation of object's shape and motion suffices to convey the meaning and action of a scene when the objects are familiar. The impact of different video adaptation strategies is then quantified with subjective experiments. We ask a panel of human observers to rate the quality of adapted video sequences on a normalized scale. From these results, we further derive an objective quality metric, the semantic peak signal-to-noise ratio (SPSNR), that accounts for different image areas and for their relevance to the observer in order to reflect the focus of attention of the human visual system. At last, we determine the adaptation strategy that provides maximum value for the end user by maximizing the SPSNR for given client resources at the time of delivery. By combining semantic video analysis and adaptive delivery, the solution presented in this dissertation permits the distribution of video in complex media environments and supports a large variety of content-based applications.

EPFL2005

Pixels: An Efficient Column Store for Cloud Data Lakes

Anastasia Ailamaki, Haoqiong Bian

To benefit from the cloud’s higher elasticity and price-efficiency, most modern data-lake engines support S3-like cloud object storage (COS) services as their optional or preferred underlying storage. Meanwhile, the widespread column stores, such as Parquet, are applied in these data lakes to improve analytical performance. However, as these column stores were designed for on-premise HDFS, they often suffer from the high latency of COS and deliver sub-optimal query performance. We observe that by optimizing the storage layout and data access pattern, we can effectively hide and mitigate the high latency. In this paper, we present Pixels, a column store optimized for the cloud that solves the problem by (1) the workload-driven storage layout optimization within and across the row group boundaries; (2) the I/O scheduling concerning the optimized storage layout and the performance characteristics of COS. They collectively improve the analytical performance in a transparent way that does not affect data ingestion and query execution in data lakes. Evaluations show that Pixels outperforms the state-of- the-art column store on COS by more than one order of magnitude on real-world workload and by 1.93x on TPC-H. Moreover, the performance of Pixels is also portable to HDFS.

IEEE2022

Adaptive video delivery using semantics

Olivier Steiger

EPFL2005

Adaptive video delivery using semantics

Progressive geometrical compression of arbitrary shaped video objects

Pixels: An Efficient Column Store for Cloud Data Lakes

Adaptive video delivery using semantics

Progressive geometrical compression of arbitrary shaped video objects

Pixels: An Efficient Column Store for Cloud Data Lakes