Server Disk Research Articles

Storage-Saving Scheduling Policies for Clusters Running Containers

Container technology plays an important role in the virtualization landscape today. A container uses its own file system consisting of a stack of layers, which are stored on the execution server's disk. Containers running on the same server share the layers they have in common, and this sharing results in valuable savings in server storage space. Many containers can run on a single server, but when their resource demands grow enough, they are distributed across a cluster of nodes/servers by orchestration systems, such as Kubernetes. In this work, we found that for the same amount of containers to run, the storage required is higher for clusters consisting of a larger number of nodes. The severity of this storage overhead depends on the scheduling policy used to select the nodes that run the containers. By comparing different storage-saving scheduling policies that differ from each other in the depth of storage knowledge they leverage to make decisions, our analysis reveals that only deep, layer-level knowledge can effectively counter the growth in storage demand as the cluster size increases. Policies with coarser-grained knowledge achieve limited benefit because they achieve performance that is nearly equal to that of a random, zero-knowledge policy.

Evaluating CephFS Performance vs. Cost on High-Density Commodity Disk Servers

CephFS is a network filesystem built upon the Reliable Autonomic Distributed Object Store (RADOS). At CERN we have demonstrated its reliability and elasticity while operating several 100-to-1000TB clusters which provide NFS-like storage to infrastructure applications and services. At the same time, our lab developed EOS to offer high performance 100PB-scale storage for the LHC at extremely low costs while also supporting the complete set of security and functional APIs required by the particle-physics user community. This work seeks to evaluate the performance of CephFS on this cost-optimized hardware when it is combined with EOS to support the missing functionalities. To this end, we have setup a proof-of-concept Ceph Octopus cluster on high-density JBOD servers (840 TB each) with 100Gig-E networking. The system uses EOS to provide an overlayed namespace and protocol gateways for HTTP(S) and XROOTD, and uses CephFS as an erasure-coded object storage backend. The solution also enables operators to aggregate several CephFS instances and adds features, such as third-party-copy, SciTokens, and high-level user and quota management. Using simple benchmarks we measure the cost/performance tradeoffs of different erasure-coding layouts, as well as the network overheads of these coding schemes. We demonstrate some relevant limitations of the CephFS metadata server and offer improved tunings which can be generally applicable. To conclude, we reflect on the advantages and drawbacks related to this architecture, such as RADOS-level free space requirements and double-network penalties, and offer ideas for improvements in the future.

Open Source network boot server for low-cost computer network learning

Universities in developing countries must have innovative ways of educating students to anticipate higher education costs. Computer network courses require server and client computers that are not cheap in terms of hardware and software. The most vulnerable computer hardware is the Hard Disk Drive (HDD), while the very expensive operating system license cost often forces universities to keep using outdated operating systems. This study aims to build open source network boot server by using PXE, DNSMASQ, TFTP, Casper and NFS. The PXE, DNSMASQ, TFTP, Casper and NFS will work together to become Open Source Network Boot Server (OSNBS). We conducted a simulation to measure the client boot speed from the network by using five clients and two versions of the uBuntu operating system image that have the capability of being a tool for computer networking courses. In addition, the processor server and hard disk server will be further monitored and analyzed. The simulation were carried out using virtualization on a desktop computer. We found that all clients successfully booted from the network. The latest uBuntu operating system image can efficiently improve the use of server resources for the processor and hard drive. In addition, the use of open source operating systems and existing hardware have also significantly reduced IT expenditure. This finding reinforces that OSNBS is suitable for low-cost computer network learning.

Clustering in Hypergraphs to Minimize Average Edge Service Time

We study the problem of clustering the vertices of a weighted hypergraph such that on average the vertices of each edge can be covered by a small number of clusters. This problem has many applications, such as for designing medical tests, clustering files on disk servers, and placing network services on servers. The edges of the hypergraph model groups of items that are likely to be needed together, and the optimization criteria that we use can be interpreted as the average delay (or cost) to serve the items of a typical edge. We describe and analyze algorithms for this problem for the case in which the clusters have to be disjoint and for the case where clusters can overlap. The analysis is often subtle and reveals interesting structure and invariants that one can utilize.

Software and hardware for integrated ACS of casting quality

The implementation of the “Automated system for the operational control of castings production (AS OCCP)” is the first stage and the basis of the integrated automated production control system. It performs three main tasks: control and accounting (the course of production, products, materials, etc.), improvement of the casting quality and the operational control of technological processes. The second stage is based on mathematical processing and analysis of data coming from AS OCCP, which allows the optimal ranges of parameters of technological processes – “Automated system for optimization and analysis of production progress (AS OAPP)” to be determined. AS OAPP consists of two subsystems: quality analysis (QAS) and technology Control (TCS). The first one solves the problems of data analysis and modeling, the second one – real-time calculation of optimal process parameters and forecasting. The tasks of the first and second stages compete for access to different hardware resources. The most critical parameter for the AS OCCP is the performance of the server disk arrays, for the AS OAPP is the processor performance. In both cases, the scaling of the system is effectively solved by parallelizing operations on different servers that make up the cluster, and on different processors (cores) on the same server.

SOFTWARE AND HARDWARE AUTOMATED SYSTEM OF CASTS DEFECTS NON-DESTRUCTIVE MONITORING

Introduction of the “Automated system for operational control of casts production (OCCP AS)” makes the basis of an integrated automated production control system (APCS). It performs three main tasks: control and recording (production, products, materials, etc.), improving quality of casts and operational management of technological processes. Solution of these tasks was accomplished through automating data collection in real time for all production operations, recording material flows, creating operational communication channels, as well as centralized collection, processing and representation of data by the process information server. The next step in building an effective automated control system is to stabilize product quality in changing external conditions, for example, quality of materials, and to optimize production (technology change in order to reduce costs for constant or higher product quality). The second stage is based on mathematical processing and analysis of data coming from OCCP AS, it allows to determine optimal ranges of parameters of technological processes – “Automated system for optimization and analysis of production progress (OAPP AS)”. OAPP AS consists of two subsystems: quality analysis and technology management. The first solves the problem of data analysis and modeling, the second – calculation of real-time optimal process parameters and real time prediction. The stages tasks compete for access to different hardware resources. The most critical parameter for OCCP AS is performance of server disk arrays, for OAPP AS it is processor performance. In either case, system scaling is effectively solved by parallelizing operations across different servers, forming a cluster, and across different processors (cores) on the same server. To process defect images and to obtain cause-and-effect characteristics, you can use OpenCV software package, which is an open source computer vision library. In course of processing, Sobel operator, Gauss filter and binarization were used. They are based on processing pixels using matrices. Operations on pixels are independent and can be performed in parallel. The task of clustering is reduced to definition of an expert method or using various mathematical algorithms for defects belonging to a specific cluster (data block) through a set of values of dependent factors. Thus, data blocks are formed by the criterion of the defect cause. Calculation of a data block to which a product defect belongs can be very resource-intensive operation. To increase efficiency of image recognition systems and parallelization ofsearch operations, it makes sense to place data clusters on different servers. As a result, there is a need for a distributed database. This is a special class of DBMS, which requires appropriate software. Generation of OAPPAS based on a multi-node cluster with ApacheCassandra DBMS installed and using Nvidia video cards supporting CUDA technology on each node will be the cheapest and most effective solution. Video card is selected based on required number of graphics processors on the node.

A milestone for DPM (Disk Pool Manager)

The DPM (Disk Pool Manager) system is a multiprotocol scalable technology for Grid storage that supports about 130 sites for a total of about 90 Petabytes online. The system has recently completed the development phase that had been announced in the past years, which consolidates its core component (DOME: Disk Operations Management Engine) as a full-featured high performance engine that can also be operated with standardWeb clients and uses a fully documented REST-based protocol. Together with a general improvement on performance and with a comprehensive administration command-line interface, this milestone also brings back features like the automatic disk server status detection and the volatile pools for deploying experimental disk caches. In this contribution we also discuss the end of support for the historical DPM components (that also include a dependency on the Globus toolkit), whose deployment is now only linked to the usage of the SRM protocols, hence can be uninstalled when these are not needed any more by the site.

IMPLEMENTASI FTP SERVER DENGAN METODE TRANSFER LAYER SECURITY UNTUK KEAMANAN TRANSFER DATA MENGGUNAKAN CENTOS 5.8

Information needed in life, to get information and exchange information needed is easy to do anytime and anywhere. File and data information, computer network is one of the important things for transfer information. One of the media that can be used is FTP (File Transfer Protocol), an application to exchange data between clients and servers on a network, by means of clients uploading files stored on the server hard disk and accessible through clients connected to the internet network. Through the development of current technology, important files that are uploaded and downloaded are very vulnerable to the world of the internet such as the wireshark tool that can be read username, password, and files that are uploaded and downloaded by the client. Therefore an FTP server was created with a data transfer security system. FTP servers that have activated the TLS (Transfer Layer Security) feature can provide user name information, passwords, and files that are uploaded and downloaded by the client that cannot be read by the Wireshark tool. In full capacity, use the hard disk server, then use the Quota Disk used to provide the client quota in uploading. Thus, FTP servers that have not used TLS are user names, passwords and files can be read by the Wireshark tool. While FTP servers that have used TLS are safer because of user names, passwords and files not read by the Wireshark tool Keywords :Disk Quota, File Transfer Protocol, (TLS) Transfer Layer Security DOI 10.5281/zenodo.3373760

A block-level caching optimization method for mobile transparent computing

In mobile transparent computing, a large number of concurrent data requests from heterogeneous clients via the network need to be processed in a timely fashion, and servers have to repeatedly fetch (search and read) the data from storage, which may cause numerous I/O costs. Generally, disk access speed is more limited than memory; therefore, massive I/O operations at servers may become a bottleneck for the system, and transport delay of the total network caused by the limitation of wireless bandwidth and stability may lead to poorer user experience. Hence, caching method plays a significant role in performance improvement of transparent computing systems. In this paper, we propose a block-level caching optimization method for the server and client by analyzing the system bottleneck in mobile transparent computing. We first analyze the storage format of the data file and the three-layer structure in the server according to the characteristics of requesting data from the client to the server and propose a block-level cache based on the access time and access frequency for the server. Second, considering the restriction of bandwidth and stability of the wireless network, we analyze network boot processes from the client’s startup and propose a client block-level cache optimization combined with local storage access technology. Finally, experimental results demonstrate that the server block-level cache optimization can effectively reduce the amount of server disk I/O, improving the concurrent ability of the server. In addition, the client block-level cache can significantly increase startup speed of the client, reduce network traffic and improve user experience.

Design of a Fault Tolerant Architecture for Private Cloud Computing Infrastructure

Objectives: Cloud computing has emerged as a cost-effective technology providing services to business-enterprise and IT industry. In this paper, we discuss a novel fault-tolerant scheme from an administration point of view that corresponds to implementation of Redundant Array of Independent Nodes (RAIN) model based architecture for ensuring fault-tolerance in a private cloud infrastructure setup. Methods: After exploring various combinations of Redundant Array of Independent Disks (RAID) and Logical Volume Manager (LVM) techniques, we propose a novel idea of having a fault-tolerant architecture for cloud infrastructure. Findings: As an alternative to the existing cloud fault-tolerant techniques which use expensive and specialized shared storage system, local server disk pool will be used for storing VM images. This architecture provides a robust infrastructure built over Internet Small Computer System Interface (iSCSI) targets and software RAID. Application: This RAIN based model can be deployed over the existing cloud infrastructure. This storage model can be used by cloud toolkits to provide fault tolerant backend for storage.

Ceph-based storage services for Run2 and beyond

In 2013, CERN IT evaluated then deployed a petabyte-scale Ceph cluster to support OpenStack use-cases in production. With now more than a year of smooth operations, we will present our experience and tuning best-practices. Beyond the cloud storage use-cases, we have been exploring Ceph-based services to satisfy the growing storage requirements during and after Run2. First, we have developed a Ceph back-end for CASTOR, allowing this service to deploy thin disk server nodes which act as gateways to Ceph; this feature marries the strong data archival and cataloging features of CASTOR with the resilient and high performance Ceph subsystem for disk. Second, we have developed RADOSFS, a lightweight storage API which builds a POSIX-like filesystem on top of the Ceph object layer. When combined with Xrootd, RADOSFS can offer a scalable object interface compatible with our HEP data processing applications. Lastly the same object layer is being used to build a scalable and inexpensive NFS service for several user communities.

Monitoring Evolution at CERN

Over the past two years, the operation of the CERN Data Centres went through significant changes with the introduction of new mechanisms for hardware procurement, new services for cloud provisioning and configuration management, among other improvements. These changes resulted in an increase of resources being operated in a more dynamic environment. Today, the CERN Data Centres provide over 11000 multi-core processor servers, 130 PB disk servers, 100 PB tape robots, and 150 high performance tape drives. To cope with these developments, an evolution of the data centre monitoring tools was also required. This modernisation was based on a number of guiding rules: sustain the increase of resources, adapt to the new dynamic nature of the data centres, make monitoring data easier to share, give more flexibility to Service Managers on how they publish and consume monitoring metrics and logs, establish a common repository of monitoring data, optimise the handling of monitoring notifications, and replace the previous toolset by new open source technologies with large adoption and community support. This contribution describes how these improvements were delivered, present the architecture and technologies of the new monitoring tools, and review the experience of its production deployment.

A constant-factor approximation for multi-covering with disks

We consider variants of the following multi-covering problem with disks. We are given two point sets $Y$ (servers) and $X$ (clients) in the plane, a coverage function $\kappa :X \rightarrow \mathbb{N}$, and a constant $\alpha \geq 1$. Centered at each server is a single disk whose radius we are free to set. The requirement is that each client $x \in X$ be covered by at least $\kappa(x)$ of the server disks. The objective function we wish to minimize is the sum of the $\alpha$-th powers of the disk radii. We present a polynomial time algorithm for this problem achieving an $O(1)$ approximation.

ЗАСТОСУВАННЯ НЕЙРО-НЕЧІТКОГО ПІДХОДУ ДЛЯ ПІДВИЩЕННЯ НАДІЙНОСТІ І ОПТИМАЛЬНОЇ РОБОТИ КОМП'ЮТЕРНОЇ МЕРЕЖІ

The paper presents analysis of the existing traditional methods of finding the best route for routing in networks. As an alternative to the traditional method, was proposed a neuro-fuzzy approach for the optimization of the routing process with the prediction of failure of the server's hard disk drive. For the prediction we used the package FuzzyTech environment Matlab. The program that is based on the unit neuro-fuzzy logic was written. The initial data are taken data corporation Google, which was published at a conference in Toronto. Based on these data, the predicted probability of failure of the hard disk server, resulting in changes in the architecture of a corporate network, and consequently bring changes in the routing process. Specific advantages of the neuro-fuzzy method over the traditional, namely, accounting expert opinion, the ability to self-learning and the ability to work with non-linear functions.

A new hybrid heuristic for replica placement and request distribution in content distribution networks

In Content Distribution Networks (CDN), in order to better serve clients, it is necessary to replicate contents at surrogate servers and distribute requests using such servers. The decisions of where to place replicated contents and how to distribute clients can be modeled as an optimization problem known as Replica Placement and Request Distribution Problem (RPRDP). In this paper we use a model that regards several realistic details that are not treated simultaneously in the literature, such as: constraints in server disk space and bandwidth, QoS requirements of requests and changes in the network conditions. Also, a new hybrid method, that uses exact and heuristic concepts simultaneously, is presented to solve the RPRDP. We compared the results obtained by the proposed algorithm with: a bound obtained by an exact offline approach, a solution of a real CDN provider and other hybrid heuristics. Results show that the proposed method outperforms the solution used in real CDNs and also in all other studied methods.

System level traffic shaping in disk servers with heterogeneous protocols

Disk access and tape migrations compete for network bandwidth in CASTORs disk servers, over various protocols: RFIO, Xroot, root and GridFTP. As there are a limited number of tape drives, it is important to keep them busy all the time, at their nominal speed. With potentially 100s of user read streams per server, the bandwidth for the tape migrations has to be guaranteed to a controlled level, and not the fair share the system gives by default. Xroot provides a prioritization mechanism, but using it implies moving exclusively to the Xroot protocol, which is not possible in short to mid-term time frame, as users are equally using all protocols. The greatest commonality of all those protocols is not more than the usage of TCP/IP. We investigated the Linux kernel traffic shaper to control TCP/ IP bandwidth. The performance and limitations of the traffic shaper have been understood in test environment, and satisfactory working point has been found for production. Notably, TCP offload engines' negative impact on traffic shaping, and the limitations of the length of the traffic shaping rules were discovered and measured. A suitable working point has been found and the traffic shaping is now successfully deployed in the CASTOR production systems at CERN. This system level approach could be transposed easily to other environments.

Analysis and improvement of data-set level file distribution in Disk Pool Manager

Of the three most widely used implementations of the WLCG Storage Element specification, Disk Pool Manager[1, 2] (DPM) has the simplest implementation of file placement balancing (StoRM doesn't attempt this, leaving it up to the underlying filesystem, which can be very sophisticated in itself). DPM uses a round-robin algorithm (with optional filesystem weighting), for placing files across filesystems and servers. This does a reasonable job of evenly distributing files across the storage array provided to it. However, it does not offer any guarantees of the evenness of distribution of that subset of files associated with a given "dataset" (which often maps onto a "directory" in the DPM namespace (DPNS)). It is useful to consider a concept of "balance", where an optimally balanced set of files indicates that the files are distributed evenly across all of the pool nodes. The best case performance of the round robin algorithm is to maintain balance, it has no mechanism to improve balance.In the past year or more, larger DPM sites have noticed load spikes on individual disk servers, and suspected that these were exacerbated by excesses of files from popular datasets on those servers. We present here a software tool which analyses file distribution for all datasets in a DPM SE, providing a measure of the poorness of file location in this context. Further, the tool provides a list of file movement actions which will improve dataset-level file distribution, and can action those file movements itself. We present results of such an analysis on the UKI-SCOTGRID-GLASGOW Production DPM.

Design and implementation of a reliable and cost-effective cloud computing infrastructure: the INFN Napoli experience

Over the last few years we have seen an increasing number of services and applications needed to manage and maintain cloud computing facilities. This is particularly true for computing in high energy physics, which often requires complex configurations and distributed infrastructures. In this scenario a cost effective rationalization and consolidation strategy is the key to success in terms of scalability and reliability. In this work we describe an IaaS (Infrastructure as a Service) cloud computing system, with high availability and redundancy features, which is currently in production at INFN-Naples and ATLAS Tier-2 data centre. The main goal we intended to achieve was a simplified method to manage our computing resources and deliver reliable user services, reusing existing hardware without incurring heavy costs. A combined usage of virtualization and clustering technologies allowed us to consolidate our services on a small number of physical machines, reducing electric power costs. As a result of our efforts we developed a complete solution for data and computing centres that can be easily replicated using commodity hardware.Our architecture consists of 2 main subsystems: a clustered storage solution, built on top of disk servers running GlusterFS file system, and a virtual machines execution environment. GlusterFS is a network file system able to perform parallel writes on multiple disk servers, providing this way live replication of data.High availability is also achieved via a network configuration using redundant switches and multiple paths between hypervisor hosts and disk servers. We also developed a set of management scripts to easily perform basic system administration tasks such as automatic deployment of new virtual machines, adaptive scheduling of virtual machines on hypervisor hosts, live migration and automated restart in case of hypervisor failures.

Performance Analysis and Improvement of Storage Virtualization in an OS-Online System

An OS-online system called TransCom is based on a virtual storage system that supports heterogeneous services of the operating system and applications online. In TransCom, OS and software which run in the client are stored on the centralized servers, while computing tasks are carried out by the clients, so the server is the bottleneck of the system performance. This paper firstly analyzes the characteristics of its real usage workload and builds a queuing model to locate the system bottlenecks. According to the study, the disk is the primary bottleneck, so an optimal two-level cache arrangement policy is developed on both the server and the client, which aims to avoid most of the server disk accesses. LRU algorithm is used in the client-side cache. A cache management algorithm called Frequency-based Multi-Priority Queues (FMPQ) proposed in this paper is used in the server-side cache. Experiment results show that the appropriate cache arrangement can significantly improve the capability of the TransCom server.

Exabyte Scale Storage at CERN

The future of data management for LHC at CERN brings new requirements for scalability and a change of scheduling and data handling compared to the HSM mass storage system in use today. A forecast for disk based storage volume at CERN in 2015 is on the Exabyte scale with hundreds of millions of files.A new CERN storage architecture is represented as a storage cluster with an analysis, archive and tape pool with container based data movements and decoupled namespaces.Main assets of a new system is high-availability and life cycle management for large storage installations. Today this is one of the major issues at the CERN computer centre with more than 1,000 disk servers and continuous hardware replacement. Another key point is distributed meta data handling with in-memory caching and persistent key-value stores to reduce latencies and operational complexity.Focus of this paper will be on the analysis pool implementation providing low-latency, nonsequential file access and a hierarchical namespace. A summary of performance indicators and first operational experiences will be reported.