Julien Sopena

Towards QoS-Oriented SLA Guarantees for Online Cloud Services

Cloud Computing provides a convenient means of remote on-demand and pay-per-use access to computing resources. However, its ad hoc management of quality-of-service and SLA poses significant challenges to the performance, dependability and costs of online cloud services. The paper precisely addresses this issue and makes a threefold contribution. First, it introduces a new cloud model, the SLAaaS (SLA aware Service) model. SLAaaS enables a systematic integration of QoS levels and SLA into the cloud. It is orthogonal to other cloud models such as SaaS or PaaS, and may apply to any of them. Second, the paper introduces CSLA, a novel language to describe QoS-oriented SLA associated with cloud services. Third, the paper presents a control theoretic approach to provide performance, dependability and cost guarantees for online cloud services, with time-varying workloads. The proposed approach is validated through case studies and extensive experiments with online services hosted in clouds such as Amazon EC2. The case studies illustrate SLA guarantees for various services such as a MapReduce service, a cluster-based multi-tier e-commerce service, and a low-level locking service.

A study of the scalability of stop-the-world garbage collectors on multicores

Large-scale multicore architectures create new challenges for garbage collectors (GCs). In particular, throughput-oriented stop-the-world algorithms demonstrate good performance with a small number of cores, but have been shown to degrade badly beyond approximately 8 cores on a 48-core with OpenJDK 7. This negative result raises the question whether the stop-the-world design has intrinsic limitations that would require a radically different approach. Our study suggests that the answer is no, and that there is no compelling scalability reason to discard the existing highly-optimised throughput-oriented GC code on contemporary hardware. This paper studies the default throughput-oriented garbage collector of OpenJDK 7, called Parallel Scavenge. We identify its bottlenecks, and show how to eliminate them using well-established parallel programming techniques. On the SPECjbb2005, SPECjvm2008 and DaCapo 9.12 benchmarks, the improved GC matches the performance of Parallel Scavenge at low core count, but scales well, up to 48~cores.

SLA guarantees for cloud services

Quality-of-service and SLA guarantees are among the major challenges of cloud-based services. In this paper we first present a new cloud model called SLAaaS - SLA?aware Service. SLAaaS considers QoS levels and SLA as first class citizens of cloud-based services. This model is orthogonal to other SaaS, PaaS, and IaaS cloud models, and may apply to any of them. More specifically we make three contributions: (i) we provide a novel domain specific language that allows to describe QoS-oriented SLA associated with cloud services; (ii) we present a general control-theoretic approach for managing cloud service SLA; (iii) we apply the proposed language and control approach to guarantee SLA in various case studies, ranging from cloud-based MapReduce service, to locking service, and higher-level e-commerce service; these case studies successfully illustrate SLA management with different QoS aspects of cloud services such as performance, dependability, financial energetic costs. We provide a domain specific language that allows to describe SLA in cloud services.We present a general control-theoretic approach for managing cloud service SLA.We apply our approach on MapReduce, locking, and e-commerce services.

NumaGiC: a Garbage Collector for Big Data on Big NUMA Machines

On contemporary cache-coherent Non-Uniform Memory Access (ccNUMA) architectures, applications with a large memory footprint suffer from the cost of the garbage collector (GC), because, as the GC scans the reference graph, it makes many remote memory accesses, saturating the interconnect between memory nodes. We address this problem with NumaGiC, a GC with a mostly-distributed design. In order to maximise memory access locality during collection, a GC thread avoids accessing a different memory node, instead notifying a remote GC thread with a message; nonetheless, NumaGiC avoids the drawbacks of a pure distributed design, which tends to decrease parallelism. We compare NumaGiC with Parallel Scavenge and NAPS on two different ccNUMA architectures running on the Hotspot Java Virtual Machine of OpenJDK 7. On Spark and Neo4j, two industry-strength analytics applications, with heap sizes ranging from 160GB to 350GB, and on SPECjbb2013 and SPECjbb2005, ourgc improves overall performance by up to 45% over NAPS (up to 94% over Parallel Scavenge), and increases the performance of the collector itself by up to 3.6x over NAPS (up to 5.4x over Parallel Scavenge).

Assessing the scalability of garbage collectors on many cores

Managed Runtime Environments (MRE) are increasingly used for application servers that use large multi-core hardware. We find that the garbage collector is critical for overall performance in this setting. We explore the costs and scalability of the garbage collectors on a contemporary 48-core multiprocessor machine. We present experimental evaluation of the parallel and concurrent garbage collectors present in OpenJDK, a widely-used Java virtual machine. We show that garbage collection represents a substantial amount of an applications execution time, and does not scale well as the number of cores increases. We attempt to identify some critical scalability bottlenecks for garbage collectors.

Service Level Agreement for Distributed Mutual Exclusion in Cloud Computing

In Cloud Computing, Service Level Agreement (SLA) is a contract that defines a level and a type of QoS between a cloud provider and a client. Since applications in a Cloud share resources, we propose two tree-based distributed mutual exclusion algorithms that support the SLA concept. The first one is a modified version of the priority-based Kanrar-Chaki algorithm [1] while the second one is a novel algorithm, based on Raymond algorithm [2], where a deadline is associated with every request. In both cases, our aim is to improve Critical Section execution rate and to reduce the number of SLA violations, which, for the first algorithm represents the number of priority inversions (i.e. a higher priority request is satisfied after a lower one) and for the second one, the number of requests whose deadline is not respected. Performance evaluation results show that our solutions significantly reduce SLA violations avoiding message overhead.

A fault-tolerant token-based mutual exclusion algorithm using a dynamic tree

This article presents a fault tolerant extension for the Naimi-Trehel token-based mutual exclusion algorithm. Contrary to the extension proposed by Naimi-Trehel, our approach minimizes the use of broadcast support by exploiting the distributed queue of token requests kept by the original algorithm. It also provides good fairness since, during failure recovery, it tries to preserve the order in which token requests would have been satisfied had the failure not occurred.

A Composition Approach to Mutual Exclusion Algorithms for Grid Applications

We propose a new composition approach to mutual exclusion algorithms for applications spread over a grid which is composed of a federation of clusters. Taking into account the heterogeneity of communication latency, our hierarchical architecture combines intra and inter cluster algorithms. We focus on token-based algorithms and study different compositions of algorithms. Performance evaluation tests have been conducted on a national grid testbed whose results show that our approach is scalable and that the choice of the most suitable inter cluster algorithm depends on the behavior of the application.

Fair Comparison of Gossip Algorithms over Large-Scale Random Topologies

We present a thorough performance comparison of three widely used probabilistic gossip algorithms over well-known random graphs. These graphs represent some large-scale network topologies: Bernoulli (or Erdos-Rényi) graph, random geometric graph, and scale-free graph. In order to conduct such a fair comparison, particularly in terms of reliability, we propose a new parameter, called effectual fan out. For a given topology and gossip algorithm, the effectual fan out characterizes the mean dissemination power of infected sites. For large-scale networks, the effectual fan out has thus a strong linear correlation with message complexity. It enables to make an accurate analysis of the behavior of a gossip algorithm over a topology. Furthermore, it simplifies the theoretical comparison of different gossip algorithms on the topology. Based on extensive experiments on top of OMNet++ simulator, which make use of the effectual fan out, we discuss the impact of topologies and gossip algorithms on performance, and how to combine them to have the best gain in terms of reliability.

2W-FD: A Failure Detector Algorithm with QoS

Failure detection plays a central role in the engineering of distributed systems. Furthermore, many applications have timing constraints and require failure detectors that provide quality of service (QoS) with some quantitative timeliness guarantees. Therefore, they need failure detectors that are fast and accurate. We introduce the Two-Windows Failure Detector (2W-FD), an algorithm able to react to sudden changes in network conditions, property that currently existing algorithms do not satisfy. We ran tests on real traces and compared the 2W-FD to state-of-art algorithms. Our results show that our algorithm presents the best performance in terms of speed and accuracy in unstable scenarios.