Workshop Program

Cloud-of-clouds storage exploits diversity of cloud storage vendors to provide fault tolerance and avoid vendor lock-ins. Its inherent diversity property also enables us to offer keyless data security via dispersal algorithms. However, the keyless security of existing dispersal algorithms relies on the embedded random information, which breaks data deduplication of the dispersed data. To simultaneously enable keyless security and deduplication, we propose a novel dispersal approach called convergent dispersal, which replaces original random information with deterministic cryptographic hash information that is derived from the original data but cannot be inferred by attackers without knowing the whole data. We develop two convergent dispersal algorithms, namely CRSSS and CAONT-RS. Our evaluation shows that CRSSS and CAONT-RS provide complementary performance advantages for different parameter settings.

Modern applications expand to fill the space available to them, exploiting local storage to improve performance by caching, prefetching and precomputing data. In virtualized settings, this behavior compromises storage elasticity owing to a rigid contract between the hypervisor and the guest OS: once space is allocated to a virtual disk and used by an application, it cannot be reclaimed by the hypervisor. In this paper, we propose a new guest filesystem called Harmonium that exploits the ephemeral or derivative nature of application data. Each file in Harmonium optionally has a motif that describes how the file can be reconstructed via computation, network accesses, or operations on other files. Harmonium expands files from their motifs when space is available, and contracts them back to their motifs when it is scarce. Given a target size, the system selects files to expand or contract based on the load on the CPU, network, and storage, as well as expected access patterns. As a result, Harmonium enables elastic cloud storage, allowing the hypervisor to dynamically balance storage across multiple VMs.

This discussion paper argues that there are five fundamental pitfalls, which can restrict academics from conducting cloud computing research at the infrastructure level, which is currently where the vast majority of academic research lies. Instead academics should be conducting higher risk research, in order to gain understanding and open up entirely new areas.

We call for a renewed mindset and argue that academic research should focus less upon physical infrastructure and embrace the abstractions provided by clouds through five opportunities: user driven research, new programming models, PaaS environments, and improved tools to support elasticity and large-scale debugging. The objective of this paper is to foster discussion, and to define a roadmap forward, which will allow academia to make longer-term impacts to the cloud computing community.

Cloud computing has not resulted in a fundamental change to the underlying operating systems. Rather, distributed applications are built over middleware that provides high-level abstractions to exploit the cloud’s scale and elasticity. This middleware conjoins many general purpose OS instances.

Others have demonstrated that a new operating system built specifically for the cloud can achieve increased efficiency, scale and functionality [11, 14]. However, this work does not take into account the way applications are being deployed in cloud environments. In particular, entire physical or virtual machines are being dedicated to run a single application, rather than concurrently supporting many users and multiple applications.

In this paper we introduce a new model for distributed applications that embraces a reduced role of the OS in the cloud. It allows for the construction of application-driven compositions of OS functionality wherein each application can employ its own customized operating system.

The approach of vertically partitioning the index has long been considered as impractical for building a distributed search engine due to its high communication cost. With the recent surge of interest in using High Performance Computing networks such as Infiniband in the data center, we argue that vertical partitioning is not only practical but also highly scalable. To demonstrate our point, we built a distributed image search engine (VertiCut) that performs multi-round approximate neighbor searches to find similar images in a large image collection.

Prior research on mobile computation offloading has mainly focused on how to offload as well as what to offload. However, the problem of whether the offloading should be done attracted much less attention. In addition, existing offloading schemes either require special compilation or modification to the applications’ source code or binary, making them difficult to be deployed in practice. In this work, we introduce POMAC, a framework to enable dynamic and transparent mobile application offloading to clouds. A prototype has been implemented on the Dalvik virtual machine and our preliminary evaluations show that POMAC can outperform existing schemes significantly and work with real-world applications seamlessly.

Correctly applying security settings of various different applications is a time-consuming and in some cases a very difficult task. Moreover, with explosion in cloud computing popularity, cloud users are able to download and run pre-packaged virtual appliances. Many users may assume that these come with correct security settings and never bother to check or update these settings. In this paper we propose an architecture that can automatically and transparently improve security settings of arbitrary network applications in a cloud computing setup. Users can deploy virtual machines with different applications, and our system will attempt to find and test better security settings tailored towards their specific setup. We call this approach “hot-hardening” since our techniques are applied to running applications.

Note: The video and audio of this presentation have been removed at the request of the authors.

Large scale cloud infrastructure requires many management applications to run concurrently to function smoothly. Traditional management applications fall short, tripped up by unexpected failures and unanticipated interference. We present a graph based data model for cloud infrastructure, enabling architects and operators to describe large scale complex infrastructure. A goal state driven framework enables quick and safe application development, sustaining infrastructure growth and maintenance at huge scale.

A revolution is beginning in communication networks with the adoption of network function virtualization, which allows network services to be run on common off-the-shelf hardware—even in virtual machines—to increase flexibility and lower cost. An exciting prospect for cloud users is that these software-based network services can be merged with compute and storage resources to flexibly integrate all of the cloud’s resources.

We are developing an application aware networking platform that can perform not only basic packet switching, but also typical functions left to compute platforms such as load balancing based on application-level state, localized data caching, and even arbitrary computation. Our prototype “memcached-aware smart switch” reduces request latency by half and increases throughput by eight fold compared to Twitter’s TwemProxy. We also describe how a Hadoop-aware switch could automatically cache data blocks near worker nodes, or perform some computation directly on the data stream. This approach enables a new breed of application designs that blur the line between the cloud’s network and its servers.

Aggregation underlies the distillation of information from big data. Many well-known basic operations including top-k matching and word count hinge on fast aggregation across large data-sets. Common frameworks including MapReduce support aggregation, but do not explicitly consider or optimize it. Optimizing aggregation however becomes yet more relevant in recent “online” approaches to expressive big data analysis which store data in main memory across nodes. This shifts the bottlenecks from disk I/O to distributed computation and network communication and significantly increases the impact of aggregation time on total job completion time.

This paper presents LOOM, a (sub)system for efficient big data aggregation for use within big data analysis frameworks. LOOM efficiently supports two-phased (sub)computations consisting in a first phase performed on individual data sub-sets (e.g., word count, top-k matching) followed by a second aggregation phase which consolidates individual results of the first phase (e.g., count sum, top-k). Using characteristics of an aggregation function, LOOM constructs a specifically configured aggregation overlay to minimize aggregation costs. We present optimality heuristics and experimentally demonstrate the benefits of thus optimizing aggregation overlays using microbenchmarks and real world examples.

High availability and performance of a web service is key, amongst other factors, to the overall user experience (which in turn directly impacts the bottom-line). Exogenic and/or endogenic factors often give rise to anomalies that make maintaining high availability and delivering high performance very challenging. Although there exists a large body of prior research in anomaly detection, existing techniques are not suitable for detecting long-term anomalies owing to a predominant underlying trend component in the time series data.

To this end, we developed a novel statistical technique to automatically detect long-term anomalies in cloud data. Specifically, the technique employs statistical learning to detect anomalies in both application as well as system metrics. Further, the technique uses robust statistical metrics, viz., median, and median absolute deviation (MAD), and piecewise approximation of the underlying long-term trend to accurately detect anomalies even in the presence of intra-day and/or weekly seasonality. We demonstrate the efficacy of the proposed technique using production data and report Precision, Recall, and F-measure measure. Multiple teams at Twitter are currently using the proposed technique on a daily basis.

Organisations are starting to publish datasets containing potentially sensitive information in the Cloud; hence it is important there is a clear audit trail to show that involved parties are respecting data sharing laws and policies.

Information Flow Control (IFC) has been proposed as a solution. However, fine-grained IFC has various deployment challenges and runtime overhead issues that have limited wide adoptation so far.

In this paper we present MrLazy, a system that practically addresses some of these issues for MapReduce. Within one trust domain, we relax the need of continuously checking policies. We instead rely on lineage (information about the origin of a piece of data) as a mechanism to retrospectively apply policies on-demand. We show that MrLazy imposes manageable temporal and spatial overheads while enabling fine-grained data regulation.

System testing is an essential part of software development. Unfortunately, comprehensive testing of large systems is often resource intensive and time-consuming. In this paper, we explore the possibility of leveraging hierarchical virtual machine (VM) fork to optimize system testing in the cloud. Testing using VM fork has the potential to save system configuration effort, obviate the need to run redundant common steps, and reduce disk and memory requirements by sharing resources across test cases. A preliminary experiment that uses VM fork to run a subset of MySQL database test suite shows that the technique reduces VM run time to complete all test cases by 60%.

Accounting and billing of cloud resources is vital for the operation of cloud service providers and their tenants. In this paper, we categorize the trust models of current industrial and academic cloud billing solutions, and discuss the problems with these models in terms of degree of trust, scalability and robustness. Based on the analysis, we propose a novel public trust model to ensure natural and intuitive verification of billable events in the cloud. Leveraging a Bitcoin-like mechanism, we design BitBill, a scalable, robust and mutually verifiable billing system for cloud computing. Our initial results show that BitBill has significantly better scalability (supporting 10x concurrent tenants using the billing service) than the state-of-the-art third-party centralized billing system.

Since energy-related costs make up an increasingly significant component of overall costs for data centers run by cloud providers, it is important that these costs be propagated to their tenants in ways that are fair and promote workload modulation that is aligned with overall cost-efficacy. We argue that there exists a big gap in how electric utilities charge data centers for their energy consumption (on the one hand) and the pricing interface exposed by cloud providers to their tenants (on the other). Whereas electric utilities employ complex features such as peak-based, time-varying, or tiered (load-dependent) pricing schemes, cloud providers charge tenants based on IT abstractions. This gap can create shortcomings such as unfairness in how tenants are charged and may also hinder overall cost-effective resource allocation. To overcome these shortcomings, we propose a novel idea of a virtual electric utility (VEU) that cloud providers should expose to individual tenants (in addition to their existing IT-based offerings). We discuss initial ideas underlying VEUs and challenges that must be addressed to turn them into a practical idea whose merits can be systematically explored.