Online deep learning (ODL) has become an important methodology for modeling time-based performance of computer systems. An open problem is the intelligent selection of features from raw workload traces of computer systems. The best methods are overly sensitive to noisy data, causing frequent feature changes and re-training. Using all available features inflates training time and introduces model artifacts if some features should have been dropped. We present WinnowML, a method for automatically determining the most relevant feature subset for a predictive time-series model. WinnowML combines existing feature ranking algorithms and a history of each feature’s ranking to iteratively rank a feature set to lower prediction error and maximize long term relevance. From this ranked feature set, the most relevant and stable subset is selected to train a model. Experimentally, we show how WinnowML can lower a model’s mean absolute relative error up to 42% on average compared to the closest performing approach. Additionally, we lower the fluctuation in feature ranking and selection up to 65%. We also demonstrate how to combine WinnowML and a model search tool to provide improvements in performance of up to 14.5% when compared to using all the feature available.

Efficient provenance storage is an essential step towards the adoption of provenance. In this paper, we analyze the provenance collected from multiple workloads with a view towards efficient storage. Based on our analysis, we characterize the properties of provenance with respect to long term storage. We then propose a hybrid scheme that takes advantage of the graph structure of provenance data and the inherent duplication in provenance data. Our evaluation indicates that our hybrid scheme, a combination of web graph compression (adapted for provenance) and dictionary encoding, provides the best tradeoff in terms of compression ratio, compression time and query performance when compared to other compression schemes.

We present the Deep Store archival storage architecture, a large-scale storage system that stores immutable data efficiently and reliably for long periods of time. Archived data is stored across a cluster of nodes and recorded to hard disk. The design differentiates itself from traditional file systems by eliminating redundancy within and across files, distributing content for scalability, associating rich metadata with content, and using variable levels of replication based on the importance or degree of dependency of each piece of stored data. We evaluate the foundations of our design, including PRESIDIO, a virtual content-addressable storage framework with multiple methods for interfile and intra-file compression that effectively addresses the data-dependent variability of data compression. We measure content and metadata storage efficiency, demonstrate the need for a variable-degree replication model, and provide preliminary results for storage performance.

In-place reconstruction of differenced data allows information on devices with limited storage capacity to be updated efficiently over low-bandwidth channels. Differencing encodes a version of data compactly as a set of changes from a previous version. Transmitting updates to data as a version difference saves both time and bandwidth. In-place reconstruction rebuilds the new version of the data in the storage or memory the current version occupies-no scratch space is needed for a second version. By combining these technologies, we support highly mobile applications on space-constrained hardware. We present an algorithm that modifies a differentially encoded version to be in-place reconstructible. The algorithm trades a small amount of compression to achieve this property. Our treatment includes experimental results that show our implementation to be efficient in space and time and verify that compression losses are small. Also, we give results on the computational complexity of performing this modification while minimizing lost compression.

A family of communication protocols, called quorum multicasts, is presented that provides efficient communication services for widely replicated data. Quorum multicasts are similar to ordinary multicasts, which deliver a message to a set of destinations. The protocols extend this model by allowing delivery to a subset of the destinations, selected according to distance or expected data currency. These protocols provide well-defined failure semantics, and can distinguish between communication failure and replica failure with high probability. The authors have evaluated their performance, taking measurements of communication latency and failure in the Internet. A simulation study of quorum multicasts showed that they provide low latency and require few messages. A second study that measured a test application running at several sites confirmed these results.>

Voting protocols ensure the consistency of replicated objects by requiring all read and write requests to collect an appropriate quorum of replicas. It is proposed to replace some of these replicas with volatile witnesses that have no data and require no stable storage, and to regenerate them instead of waiting for recovery. The small size of volatile witnesses allows them to be regenerated much easier than full replicas. Regeneration attempts are also much more likely to succeed since volatile witnesses can be stored on diskless sites. It is shown that under standard Markovian assumptions two full replicas and one regenerable volatile witness managed by a two-tier dynamic voting protocol provide a higher data availability than three full replicas managed by majority consensus voting or optimistic dynamic voting provided site failures can be detected significantly faster than they can be repaired.>

The read and write availabilities of replicated data managed by the regeneration algorithm, a replica control protocol based on file regeneration, are evaluated, and two regeneration protocols are presented that overcome some of its limitations. The first protocol combines regeneration and the available copy approach to improve availability of replicated data. The second combines regeneration and the dynamic voting approach to guarantee data consistency in the presence of network partitions while maintaining a high availability. Expressions for the availabilities of replicated data managed by both protocols are derived and found to improve significantly on the availability achieved using extant consistency protocols.>

Two novel dynamic voting algorithms are proposed. One, called optimistic dynamic voting, operates on possibly out-of-date information, which greatly increases the efficiency of the algorithm and simplifies its implementation. The other, called topological dynamic voting, explicitly takes into account the topology of the network on which the copies reside to increase the availability of the replicated data. The authors compare availabilities of replicated data managed by both algorithms with those of data managed by existing voting protocols using a simulation model with realistic parameters. Optimistic dynamic voting is found to perform as well as the best existing voting algorithms while topological dynamic voting performs much better than all other voting algorithms when two or more copies reside in the same nonpartitionable group.>