Ronen I. Kat

Storage modeling for power estimation

Power consumption is a major issue in todays datacenters. Storage typically comprises a significant percentage of datacenter power. Thus, understanding, managing, and reducing storage power consumption is an essential aspect of any efforts that address the total power consumption of datacenters. We developed a scalable power modeling method that estimates the power consumption of storage workloads. The modeling concept is based on identifying the major workload contributors to the power consumed by the disk arrays. To estimate the power consumed by a given host workload, our method translates the workload to the primitive activities induced on the disks. In addition, we identified that I/O queues have a fundamental influence on the power consumption. Our power estimation results are highly accurate, with only 2% deviation for typical random workloads with small transfer sizes (up to 8K), and a deviation of up to 8% for workloads with large transfer sizes. We successfully integrated our modeling into a power-aware capacity planning tool to predict system power requirements and integrated it into an online storage system to provide online estimation for the power consumed.

HyperTree for self-stabilizing peer-to-peer systems

Peer-to-peer systems are prone to faults; Therefore, it is extremely important to design peer-to-peer systems that automatically regain consistency or, in other words, are self-stabilizing. In order to achieve the above, we present a deterministic structure that defines the entire (IP) pointers structure among the machines, for every n machines; i.e., defines the next hop for the insert, delete, and search procedures of the peer-to-peer system. Thus, the consistency of the system is easily defined, monitored, verified, and repaired. We present the HyperTree (distributed) structure, which supports the peer-to-peer procedures while ensuring that the out-degree and the in-degree (the number of outgoing/ incoming pointers) are b logbn where n is the actual number of machines and b is an integer parameter greater than 1. Moreover, the HyperTree ensures that the maximal number of hops involved in each procedure is bounded by logbn. A self-stabilizing peer-to- peer distributed algorithm based on the HyperTree is presented.

Efficient and agile storage management in software defined environments

The IT industry is experiencing a disruptive trend for which the entire data center infrastructure is becoming software defined and programmable. IT resources are provisioned and optimized continuously according to a declarative and expressive specification of the workload requirements. The software defined environments facilitate agile IT deployment and responsive data center configurations that enable rapid creation and optimization of value-added services for clients. However, this fundamental shift introduces new challenges to existing data center management solutions. In this paper, we focus on the storage aspect of the IT infrastructure and investigate its unique challenges as well as opportunities in the emerging software defined environments. Current state-of-the-art software defined storage (SDS) solutions are discussed, followed by our novel framework to advance the existing SDS solutions. In addition, we study the interactions among SDS, software defined compute (SDC), and software defined networking (SDN) to demonstrate the necessity of a holistic orchestration and to show that joint optimization can significantly improve the effectiveness and efficiency of the overall software defined environments.

Usage centric green performance indicators

Energy effciency of data centers is gaining importance as energy consumption and carbon footprint awareness are rising. Green Performance Indicators (GPIs) provide measurable means to assess the energy effciency of a resource or system. Most of the metrics commonly used today measure the energy effciency potential of a resource, system or application usage, rather than the energy effciency of the actual usage. In this paper, we argue that the way that the resources and systems are actually used in a given data center configuration is at least as important as the effciency potential of the raw resources or systems. Hence, for data center energy effciency, we suggest to both select energy effcient components (as done today), as well as optimize the actual usage of the components and systems in the data center. To achieve the latter, optimization of usage centric GPI metrics should be employed and targeted as a primary green goal. In this paper we identify and present usage centric metrics, which should be monitored and optimized for improving energy effciency, and hence, reduce the data center carbon footprint.

Leveraging disk drive acoustic modes for power management

Reduction of disk drive power consumption is a challenging task, particularly since the most prevalent way of achieving it, powering down idle disks, has many undesirable side-effects. Some hard disk drives support acoustic modes, meaning they can be configured to reduce the acceleration and velocity of the disk head. This reduces instantaneous power consumption but sacrifices performance. As a result, input/output (I/O) operations run longer at reduced power. This is useful for power capping since it causes significant reduction in peak power consumption of the disks. We conducted experiments on several disk drives that support acoustic management. Most of these disk drives support only two modes — quiet and normal. We ran different I/O workloads, including SPC-1 to simulate a real-world online transaction processing workload. We found that the reduction in peak power can reach up to 23% when using quiet mode. We show that for some workloads this translates into a reduction of 12.5% in overall energy consumption. In other workloads we encountered the opposite phenomenon-an increase of more than 6% in the overall energy consumption.

Self-stabilizing distributed file systems

A self-stabilizing distributed file system is presented. The system constructs and maintains a spanning tree for each file volume. The spanning tree consists of the servers that have volume replicas and caches for the specific file volume. The spanning trees are constructed and maintained by self-stabilizing distributed algorithms. File system updates use the tree to implement file read and write operations.

Setting energy efficiency goals in data centers: the GAMES approach

Energy-aware service centers take into account energy consumption of infrastructures, machines, applications, storage systems, and their distributed computing architecture. The approach to energy efficiency in data centers in the GAMES (Green Active Management of Energy in IT Service centers) project is presented: Green Performance Indicators (GPIs), i.e., properties that, continuously monitored, evidence the level of consumed energy by the centers IT resources, can be the basis of a systematic approach to increase energy efficiency. The GPIs are the basis for improving energy efficiency with adaptive actions and to achieve a higher level of green maturity, as prescribed, for instance, in the GreenGrid Data Center Maturity Model (DCMM), based on a usage-centric perspective in GPIs. The paper briefly describes monitoring of GPIs and the adaptation actions adopted to reach the green goals. Preliminary experimental results are discussed.

When consensus meets self-stabilization

This paper presents a self-stabilizing failure detector, asynchronous consensus and replicated state-machine algorithm suite, the components of which can be started in an arbitrary state and converge to act as a virtual state-machine. Self-stabilizing algorithms can cope with transient faults. Transient faults can alter the system state to an arbitrary state and hence, cause a temporary violation of the safety property of the consensus. New requirements for consensus that fit the on-going nature of self-stabilizing algorithms are presented. The wait-free consensus (and the replicated state-machine) algorithm presented is a classic combination of a failure detector and a (memory bounded) rotating coordinator consensus that satisfy both eventual safety and eventual liveness. Several new techniques and paradigms are introduced. The bounded memory failure detector abstracts away synchronization assumptions using bounded heartbeat counters combined with a balance-unbalance mechanism. The practically infinite paradigm is introduced in the scope of self-stabilization, where an execution of, say, 264 sequential steps is regarded as (practically) infinite. Finally, we present the first self-stabilizing wait-free reset mechanism that ensures eventual safety and can be used in other scopes.

ADSC: application-driven storage control for energy efficiency

While performance and quality of service are the main criteria for application data management on storage units, energy efficiency is increasingly being stated as an additional criterion for evaluation. Due to the increasing energy consumption of storage subsystems, improving their energy efficiency is an important issue. In this paper we present a novel approach to storage management whereby both mid-level (file placement) and low level (disk mode) aspects are controlled, in a tiered storage architecture. The proposed mechanism is based on policies, and it is implemented via fuzzy logic rules, in contrast to attempting to build a model of the storage subsystem. The inputs to the storage management system are high level (application), mid level (file system) and low level (disk access patterns) information. The effectiveness of our approach has been validated by means of a case study using a TPC-C benchmark modified to access file level data. Results from this simulation are presented.

Searching for a lion in the desert: optics-based acquisition algorithms for wireless sensor networks

In this paper we address the task of locating and mapping sensor nodes equipped with passive optical transmitters deployed within a wireless sensor network. The passive transmitter is in the form of a modulatable corner cube retro-reflector (CCR) and is interrogated by a base station equipped with a variable sized laser beam prior to data harvesting using optical wireless communication. We propose and discuss an efficient acquisition algorithm based on the binary search concept and evaluate its performance in locating a solitary sensor node. We demonstrate the use of an analytical tool for assessing the acquisition cost for different error probability distributions and search regimes. The common line-of-sight (LOS) requirement for both sensor location and data harvesting renders the need for individual sensor node identification redundant and thus yields a simplified search regime by comparison with previously published research on node localisation.