• No results found

HiPerDNO. High Performance Computing Technologies for Smart Distribution Network Operation FP

N/A
N/A
Protected

Academic year: 2021

Share "HiPerDNO. High Performance Computing Technologies for Smart Distribution Network Operation FP"

Copied!
19
0
0

Loading.... (view fulltext now)

Full text

(1)

HiPerDNO

High Performance Computing Technologies for Smart Distribution Network Operation FP7 - 248135

Project coordinator: Dr Gareth Taylor (BU)

Consortium Members: Brunel University (BU), Electricite de France (EF), IBM Israel, University of Oxford, UK Power Networks plc (UKPN), Union Fenosa Distribution (Union Fenosa), Indra Sistemas, GTD Systems de Information (GTD), Korona Inzeniting (Korona), Elektro Gorenjska Podjetje za Distribucijo Elektricne Energije (EG) and Fraunhofer IWES

Document Title Prototype deployment of selected architecture and necessary documentation, D1.2.2

Document Identifier HiPerDNO/2011/D.1.2.2

Version Version 1.0

Work package number WP1

Sub-Work package Number WP1.2

Distribution Public

Reporting consortium member Oxford

(2)

Prototype deployment of selected architecture and necessary

documentation, D1.2.2

Executive Summary

This document presents the HiPerDNO HPC Platform prototype developed in WP1.2. The prototype covers all major components of the HiPerDNO HPC platform, and it was presented during the HiPerDNO Quarterly Progress Meeting in Clamart on 29th – 30th June 2011. The prototype can be downloaded from the FTP server hosted by Brunel University. The prototype can be installed and run following the instruction and tutorial in this document.

This document also lists the requirements for the HPC scheduler. These requirements are met by the HPC prototype, and represent our current understanding, formed through interaction with the HiPerDNO partners, of the needs of DNO applications. Obviously, as collaborations with partners progress so will the specification and deployment of their needs within the HiPerDNO HPC Platform.

(3)

Document Information

HiPerDNO

Project Number FP7 - 248135 Acronym HiPerDNO

Full Title

Prototype deployment of selected architecture and necessary documentation, D1.2.2

Project URL http://www.hiperdno.eu

Document URL N/A

Deliverable Number

D.1.2.2 Title

Prototype deployment of selected architecture and necessary documentation (M18)

N/A Work Package

Number WP1.2 Title

Investigation and evaluation of new HPC platforms and architectures.

Date of Delivery Work Plan Actual

Status Version 1.0

Nature Prototype Internal Report External Report Dissemination Dissemination Level Public Consortium

Author(s) (Partners) Stef Salvini, David Wallom

Lead Author Name Piotr Lopatka E-mail [email protected]

Partner Oxford Phone 44 1865 610607

Abstract

This document and the prototype of the HPC System constitute deliverable D1.2.2. The prototype covers all major components of the HiPerDNO HPC platform, and it was presented during the HiPerDNO Quarterly Progress Meeting in Clamart on 29th – 30th June 2011. The prototype can be downloaded from the FTP server hosted by Brunel University. Access details can be found on Basecamp. This document describes the prototype, its installation and use.

(4)

Table of Contents

1. Introduction ... 5

2. Prototype Description ... 6

Purpose of the Prototype ... 6

Functional Description of HPC Prototype ... 6

HPC Interface to the Data System ... 7

Applications scheduling ... 7

3. Prototype Deployment – Installation Tutorial ... 9

Prerequisites ... 9

Creation of VM ... 9

Virtual network setup. ... 10

Head-Node: hpc1 ... 11

Compute-Nodes: hpc1, hpc2, hpc3, hpc4 ... 12

Fast Storage node: storage ... 12

Data System nodes: hdfs1, hdfs2 ... 13

4. Testing the Prototype ... 14

DNO-like model applications ... 14

Running the example ... 14

Other key commands (HDFS, Torque, MAUI) ... 16

Editing schedules ... 17

5. From Prototype to System Deployment ... 18

(5)

1. Introduction

This document describes the prototype of the HiPerDNO HPC Platform that was described and demonstrated at the HiPerDNO Quarterly Progress Meeting in Clamart on 29th -30th June 2011. The prototype implemented the design concepts introduced in the Deliverable 1.2.1.

The chapters of this document discuss the prototype implementation, deployment and the tests that were conducted. The HPC prototype software package has been made available to all partners through the FTP server hosted by Brunel University. Installation instructions are included in this deliverable.

(6)

2. Prototype Description

Essential reading : HiPerDNO deliverables 1.2.1 [1], 3.1.1 (particularly chapters 2 and 3) [7].

The prototype consists of the HPC Engine (HPCE) and the Data Subsystem (DSS). The HPCE schedules and executes the DNO applications making optimal usage of resources while ensuring appropriate balance and priority between mission critical and non-critical applications. The Data System is a resilient and secure repository for the data necessary to execute DNO applications. A fast storage system is an essential component of the HPCE: it is a local storage system for compute nodes, where applications as well as their current operational data (scratch data) are stored.

Purpose of the Prototype

The purpose of this prototype is two-fold. In the first place, it allowed both feasibility and implementation studies of the system architecture presented in [1], as it covers all major components of the HiPerDNO HPC Platform as defined in that document. In the second place, the prototype allows the testing of the DNO application scenarios to run on the full HPC system. Test applications of appropriate scheduling requirements and characteristics were employed during the demonstration (see chapter 4). It also allows the integration and test deployment of real DNO applications, which is the subject of on-going work.

Functional Description of HPC Prototype

The HPC prototype consists of the HPCE and the DSS and it is depicted in Figure 1. One of the nodes in the HPCE is the head-node. In the prototype it plays also the role of the DMS, responsible for requesting execution of jobs in the HPC and it includes the scheduler and its required mechanisms. The HPC nodes require the presence of a common storage space (Fast Storage in figure 1).

(7)

HPC Interface to the Data System

All data are stored in the DSS. The DSS is not visible to the applications or indeed to either the DMS or the HPCE: data can only be accessed through appropriate client-server mechanisms. Both retrieval and storage of data are carried out through the client-server interface.

In the prototype, data retrieval and storage results for the model applications are greatly simplified. We expect that in the HiperDNO HPC Platform appropriate data design, including metadata, will be carried out, but this is application-dependent and details will emerge in due course through the collaborations already started.

The prototype relies on meta-data protocol that allows identification of the types of the application and of data access. Exchange of meta-data precedes exchange of content.

The client-approach also allows changing the DSS technology (currently HDFS), without affecting the rest of the system and can be security hardened.

Applications scheduling

The scheduler ensures that applications are executed according to the execution priorities set by the DNOs. Based on collaboration with partners, we (WP1.2) have laid out the preliminary scheduler requirements for HPC applications, and are further discussed in chapter 4 below, and in [7]. We would like to stress that these requirements are not final: they corresponds to our current understanding, and no doubt will be modified following the DNOs feedback and the emergence of fully defined requirements as the project progresses.

High-level requirements that must be supported by the scheduler are pre-emption and partitioning. Pre-emption allows the HPC system to stop a running task in order to make resources available for a high-priority task (kill-and-requeue). Low priority jobs are killed and re-added to the appropriate scheduler queue, without altering their job id, hence their place in the appropriate queue. Pre-empted tasks can then be restarted as soon as HPC Engine resources become available.

Partitioning allows limiting the amount of resources that can be dedicated to jobs of a specific priority. The scheduler requirements are listed in greater detail in table 1. We have employed the Maui scheduler in the prototype, sitting on the Torque resource manager. Maui is a freely available, open-source package, and it supports all the requirements in table 1, with the exception of the optional no 6. Maui does not allow medium priority jobs to be both pre-emptable by higher priority jobs and at the same time pre-emptying lower priority ones: roles of pre-emptee and pre-emptor, so to speak, are incompatible, although this might be further studied. In any case, we do not believe that this has any major impact at this stage of the HiPerDNO project. The suggested partitioning of computing resources is adequate for current needs. Notice that we had to create a patch to ensure that Maui satisfies requirements no. 5 in table 1. This patch is of course part of this deliverable and is made available together will other sources, scripts and model applications.

(8)

No.

Type

Requirement

1.

Resources

Must

The scheduler

supports execution of jobs requiring 1...n

compute nodes.

2.

Pre-emption

Must

The scheduler supports pre-emption.

3.

Pre-emption

Must

Jobs can be classified as critical and non-critical (critical jobs

always pre-empt non-critical jobs).

4.

Pre-emption

Must

Requeueing mechanism is supported, i.e. killing jobs and

putting them at the top of their queue for execution.

5.

Pre-emption

Must

Critical job can pre-empt as many low-priority jobs as

necessary (but no more) to get the required resources for

execution.

6.

Pre-emption

Nice

to

have

Another class of jobs can be defined with intermediate

priority level “between critical and non-critical”. These jobs

can empt non-critical jobs, however they can be

pre-empted by critical jobs at the same time (pre-emption

cascade).

7.

Partitioning

Must

If no. 6 is not possible, the scheduler must allow the

partitioning of resources. The amount of compute resources

allocated to a certain class of critical jobs can be limited.

8.

Scheduling

latency

Must

To satisfy mission-critical requirements, high priority jobs

must be run when they are submitted.

Pre-emption and scheduling frequency must be sufficiently

fine-grained to allow the stopping of low priority jobs and

starting of high-priority jobs quickly and effectively. (Order

of 1 second?)

(9)

3. Prototype Deployment – Installation Tutorial

The following sections discuss the installation of the HPC prototype. The prototype is to be deployed in a virtualized environment. We would like to stress that the target deployment infrastructure for the HPC system will be the real hardware. This is however the part of on-going work and not a part of this tutorial (see also Chapter 5).

A number of prerequisites must be met before starting the HPC prototype installation. The deployment consists of two stages. The first stage involves the creation of a virtual cluster (nodes and network), the installation of the operating system and the network configuration. The second stage involves the installation and configuration of software which is central to the prototype: resource manager, scheduler, fast storage and data system software.

Prerequisites

The following list presents all necessary components for the deployment of the prototype. We suggest using VirtualBox as virtualization software and Ubuntu Server 10.04 as an operating system for the nodes.

 Host

o Hardware: Desktop PC (we used: Intel Core 2 quad, 8 GB RAM) o OS: Linux (we used: Ubuntu 10.04)

 Virtual Machines (VM)

o Virtualisation software: Oracle VirtualBox (www.virtualbox.org) o Guest OS: Ubuntu Server 10.04 (www.ubuntu.com) (or Ubuntu 10.04)  Prototype software (Archive: D1.2.2.tar.gz from Basecamp), including

o Cluster Resources MAUI v.3.3.1 (patched version) o Cluster Resources Torque v. 3.0.0

o MAUI configuration file: maui.cfg o Torque configuration file: torque.cfg o demo scripts and applications o a video of running prototype

o HDFS (Hadoop Yahoo distribution 20.2, hadoop.apache.org or www.cloudera.com)

Creation of VM

Seven virtual machines should be created. The names listed below are fixed, as they are part of the prototype configuration: hpc1, hpc2, hpc3, hpc4, storage, hdfs1, hdfs2.

The OS on the VMs do not require a GUI, and therefore have modest requirements: one CPU core and 384MB of RAM per virtual machine are sufficient. VirtualBox virtualization software provides graphical interface which makes it easy to create VM’s quickly. During the creation most parameters can be left unchanged.

During the VM creation the network interfaces must be set correctly. We chose to use two network interfaces per virtual machine. The first one gives access to the Internet (security updates,

(10)

software repositories). The second interface connects the VMs together using a virtual network and makes them accessible from the host. In terms of VirtualBox configuration, two network interfaces should be enabled during the VM generation: the first one will be a NAT (Network Address Translation)for Internet access, the second is a Host-Only-Adapter for the virtual network connection with other VMs and the host.

The OS must then be installed. The installation of the Ubuntu Server 10.04 is very straightforward, and it takes only few minutes. During installation, all the default options can be accepted. When asked, create the default username “proto”, and mark the OpenSSH package for installation. It is possible to speed up the whole process by cloning the virtual hard drive of the first VM and plugging it into the other six. It is important to remember to change their hostnames. For more information on cloning, please refer to the VirtualBox documentation [5, chapter 5].

Virtual network setup.

This step assumes that all the VMs have been implemented, the OS has been installed in each of them and that the user “proto” is enabled and has rights to execute administrator’s tasks (sudo). The network configuration requires editing two configuration files. The IP addresses are static for the virtual cluster network, and dynamic for the access to the Internet. The static addresses were chosen arbitrarily and they can be changed if necessary. The following section can be appended to each VM /etc/hosts file.

// Prototype hostnames mapping 192.168.56.10 hpc1 192.168.56.11 hpc2 192.168.56.12 hpc3 192.168.56.13 hpc4 192.168.56.80 hdfs1 192.168.56.81 hdfs2 192.168.56.100 storage

The VirtualBox settings of each VM define the first network interface as NAT and the second as Host-Only-Adapter and the Linux kernel in the guest OS should assign to them eth0 and eth1 descriptors, respectively. The following section can be added to the /etc/network/interfaces file of every VM operating system if required. It may be necessary to substitute the IP address field for each VM to avoid any address conflicts.

// The primary (NAT) network interface auto eth0

iface eth0 inet dhcp

// The secondary (virtual network) interface auto eth1

(11)

After VMs are rebooted, the network should be up and running. In order to verify both interfaces, try to ping a machine on the internet and the host virtual network interface.

It is suggested to enable SSH access on all nodes for the user “proto”, so that the nodes can be accessed from the host. By default the SSH service is enabled if the OpenSSH software was marked for installation during OS installation and we suggest to enable password-less SSH access from the host to all nodes. Instructions on how to do it can be found in the last section of this chapter, which discusses the installation of Data System software.

This step completes the first stage of the installation. The Virtual cluster can be started and it is possible to connect to any VM machine from the host machine using SSH. In the next step, we are going to install the required software packages on the virtual nodes.

Head-Node: hpc1

The hpc1 node is the head-node. It requires the following software packages to be installed:  MAUI scheduler

 Torque server module

Please use the MAUI installation package provided with this deliverable and not the one available on the website above: we have applied a required patch to the package we provided.

MAUI

Copy, unpack and install the file: maui-{version}.tar.gz, using the following commands: >tar –xzvf maui-*.tar.gz

>cd maui-* >./configure >make

>make install

Copy the file maui.cfg from the installation directory to the following location on the hpc1 node: maui.cfg -> (hpc1) /var/spool/maui/maui.cfg

The MAUI installation and troubleshooting documentation can be found in [2]. Torque

Copy, unpack and install the file: torque-{version}.tar.gz, using the following commands: >tar –xzvf torque-*.tar.gz

>cd torque-* >./configure >make

>sudo make install

The installation of Torque server module creates /var/spool/torque directory where configuration files are placed. From the installation folder torque files copy the following file:

nodes -> (hpc1) /var/spool/torque/server_priv/nodes Set up Torque submission queues by copying and installing torque configuration file: torque.cfg -> (hpc1) ~/torque.cfg

(12)

>sudo qmgr < torque.cfg

Torque installation and troubleshooting documentation can be found in [3].

Compute-Nodes: hpc1, hpc2, hpc3, hpc4

Torque

The nodes hpc1, …, hpc4 are compute nodes. Notice that hpc1 node is also the head-node. The compute nodes require the following software installed:

 Torque compute node module  NFS client

The Torque module will allow compute nodes to communicate with the head-node.

To install the Torque compute node module login in the hpc1 node, enter the Torque installation director and create compute module installation packages:

>cd torque-* >make packages

It will generate a list of Torque packages. Copy and install the following packages on all four compute nodes:

>sudo torque-package-mom-linux-i686.sh --install >sudo torque-package-clients-linux-i686.sh --install To enable Torque to run as a service, execute the following commands: >sudo cp contrib/init.d/debian.pbs_mom /etc/init.d/pbs_mom >update-rc.d pbs_mom defaults

From the installation folder torque files copy the following file:

server_name -> (hpc1,…, hpc4) /var/spool/torque/server_name config -> (hpc1,…, hpc4) /var/spool/torque/mom_priv/config

NFS Client

Compute nodes require the following software:  NFS Client

To install NFS Client on Ubuntu virtual machine, follow the following tutorial [6].

https://help.ubuntu.com/community/SettingUpNFSHowTo

On each compute node: hpc1, …, hpc4 create ~/shared/demo folder for the user proto. This folder should be mapped to the exported folder on the Fast Storage System (see next section).

Fast Storage node: storage

(13)

Create shared folder:

>mkdir -p ~/storage/demo

Export the folder as a shared folder, by adding the following entry to the /etc/fstab file: >/home/proto/storage/ /export/storage none bind 0 0

Verify whether exported folder can be accessed from all compute nodes.

Data System nodes: hdfs1, hdfs2

The Data System requires installation of:  HDFS (Hadoop Distributed File System)

The HDFS package can be obtained from the Cloudera website under the following link:

https://ccp.cloudera.com/display/CDHDOC/CDH3+Installation

The HDFS has few prerequisites. It requires the JDK6 and SSH packages. The details of installing JDK6 are not addressed here as they are described in the link above. SSH setup requires password-less type of access between two machines. Setting up password-less access requires two steps: generating the key for the user, and copying this key to a remote machine for subsequent password-less access.

To generate an encryption key, execute on both machines hdfs1 and hdfs2 the following command: >ssh-keygen –t rsa

When asked, confirm default parameters. To enable password-less SSH access between hdfs1 and hdfs2 machines, run the following commands on the nodes:

On hdfs1 node:

>ssh-copy-id –I ~/.ssh/id_rsa.pub proto@hdfs2 On hdfs2 node:

>ssh-copy-id –I ~/.ssh/id_rsa.pub proto@hdfs1

Before the HDFS cluster can be used, it should be formatted. A list of useful HDFS commands, including the command to format the HDFS system is presented in the next chapter.

(14)

4. Testing the Prototype

DNO-like model applications

The prototype contains three “model” DNO applications as they are studied in the HiPerDNO project: distributed state estimation (DSE), condition-monitoring (CM), data-mining (DM). These applications are different with respect to criticality, execution pattern (periodic, aperiodic) and required compute resources. The proposed characteristics are summarized in table 2 were chosen as an example scenario. At the same time they are based on the input provided by the partners in the project.

DNO Application Scheduling characteristics

DSE mission-critical, periodic, short submission interval, parallel

CM High-priority, periodic, long submission interval, batch of serial tasks DM non-critical w.r.t timely results, aperiodic, long-execution time

Table 2. Model DNO application, execution requirements.

The DSE is a periodic job with a short submission interval (few minutes; for the purpose of this prototype shortened significantly). It is a parallel task requiring a number of compute nodes and it is a mission-critical task, thus requiring immediate execution at the moment of submission.

The CM is a periodic task with long submission intervals. The CM submission consists of a batch of small serial jobs submitted together at regular intervals. CM jobs are high priority and are executed on a limited number of compute nodes

The DM jobs are not periodic; they are serial tasks, submitted at arbitrary intervals, requiring long execution times. They are not critical jobs from the DNO point of view and can be executed when the system load due to higher priority jobs allows. Thus their execution can be postponed. The model DM applications includes checkpointing/warm restarting, thus allowing DM jobs to pre-empted and rescheduled (kill-and-requeue).

Running the example

In the previous chapter HPC hardware infrastructure and software components (Torque, MAUI, NFS, HDFS) were set up. Before the prototype can be run the DNO model applications and submission scripts must be deployed from the host, as well as basic data layout in the DSS.

Start the prototype by executing the following script on the host: >./startvms

When all virtual machines are running, deploy all scripts and DNO model applications from the host by running:

(15)

Open the terminal window and log into the hdfs1 node. Create data folders for DNO applications: >hadoop dfs –mkdir DSEIn

>hadoop dfs –mkdir DSEOut >hadoop dfs –mkdir CMIn >hadoop dfs –mkdir CMOut >hadoop dfs –mkdir DMIn >hadoop dfs –mkdir DMOut

Next, generate example input files for DNO applications by running: >./set.py DSE 10

>./set.py CM 2 >./set.py DM 500

These commands will create the input files needed by applications running in the prototype. The second parameter behind the name of the application (DSE, CM, DM) specifies the length of its input file and effectively the execution time of the application.

Run the Data Store server script: >./ds.py

The script is a part of the client-server interface, which sits on top of the Data System. Open the terminal window and log into the node hpc1. Run the following script: >./monitor.sh

Open the second terminal window and log into the node storage. Run the following script: >./filemon.sh

The two terminal windows will show scheduling and checkpoint information accordingly.

Open another terminal window and login into hpc1 node. Run the prototype demonstration with the following command:

>./demo.sh

The demo takes a couple of minutes to execute. In a terminal window with scheduling information, one can see which jobs are submitted but waiting for execution, which are executing and on which nodes. The second terminal window shows how checkpoint files are created. Both terminal windows are shown in figure 2, which presents a snapshot from the running prototype. We can see that five jobs are present in the HPC system: four DM jobs and one DSE job. Three DM jobs are not running because of the DSE job pre-emption. Only one DM job is in execution. In the right terminal window we can see that new checkpoint files are being generated for the DM job running.

(16)

Figure 2. HPC Prototype demo.

Other key commands (HDFS, Torque, MAUI)

In order to manage the prototype the following commands may prove to be useful. More complete list of commands to manage HDFS, Torque, MAUI can be found through the references [2-4].

HDFS

The basic operations on HDFs include formatting the HDFS file system, copying, listing, and removing the files. To format the HDFS , using the following command:

>sudo hadoop namenode –format

To display the list of files in the HDFS directory issue the following command: >hadoop dfs –ls URI

To remove a file from the HDFS, issue the following command: >hadoop dfs –rm URI

To display the content of the file onto a terminal, issue the following command: >hadoop dfs –cat URI

To copy a file from the local file system to HDFS, use the following command: >haddop dfs –copyFromLocal {localSrc} URI

To copy a file from HDFS to a local file system, use the following command: >hadoop dfs –copyToLocal URI <localdst>

For the complete list of HDFS commands please refer to [4]. Torque

By default, Torque starts its standard PBS scheduler when the head node boots up. The PBS scheduler is not part of the prototype and it should be stopped manually, before enabling MAUI (see the next section). To stop the PBS scheduler, type the following command on the hpc1 front node:

(17)

are running, use the following command on the hpc1 front node: >pbsnodes

In the output of this command check if for each node the state is set to free. It if is, it means the node is up and running and waiting for jobs to be submitted.

In order to check the list of submitted jobs use the following command on the head node HPC1: >qstat

The command will display the list of jobs submitted to the resource manager, together with their state. MAUI

The key configuration tool for the MAUI scheduler is its configuration file. It is not the purpose of this document to go in depth on the configuration parameters of the MAUI schedulers. For those interested, refer to [2]. The configuration file maui.cfg is provided together with this deliverable and it should be copied to appropriate location on the HPC1 head node before starting MAUI scheduler. The location for MAUI configuration file is: /var/spool/maui/maui.cfg.

Make sure that the default PBS scheduler is not running (see previous section). To start MAUI use the following command on the HPC1 head node:

>maui

Editing schedules

The prototype is shipped with an example scenario “demo.sh”. It is possible to submit CM, DM, DSE jobs without demo.sh script. In order to submit CM or DM type:

>./submit.sh DM >./submit.sh CM

To submit DSE, one can specify additionally how many nodes DSE job would require (max 3). >./submit.sh DSE 2

>./submit.sh DSE 3

To influence the runtime of the jobs modify the size of the input file stored in the Data Store (see section Data System setup in the previous chapter). For example to shorten by half the execution time of the DSE execute the following command on the hdfs1:

(18)

5. From Prototype to System Deployment

The HPC Prototype described in this document lays the ground for the work ahead in the WP1.2 and other work packages in the HiPerDNO project.

We are planning to deploy the current HPC prototype on real hardware. This deployment will be made available to partners in the HiPerDNO project. Based on initial discussions within the OeRC team we will make available an Intel-based cluster consisting of seven or eight dual-CPU nodes. It will be configured so as to allow both the development and tuning of real applications as well as study of the overall HPC System under realistic deployment conditions.

We plan to use Rocks (http://www.rocksclusters.org), to ease the deployment and maintenance of the system, and interconnects adequate to the task in hand (hopefully, 10 Gbit Ethernet). Work on the deployment will commence in August and we plan to make it available to all partners in September. We are also starting collaborations with various HiPerDNO partners (EDF, BU, Indra/UF) with the aim of porting to and integrating their algorithms on the HPC platform. More information can be obtained from the forthcoming HiPerDNO deliverable 3.1.1 ([7]).

(19)

6. References

1. S. Salvini, P. Lopatka, D. Wallom, HiPerDNO Deliverable D.1.2.1 “Report on the

architecture and performance criteria used for selection.”

2. Adaptive Computing Enterprises, “Maui Scheduler Administrator’s Guide version 3.2”,

http://www.adaptivecomputing.com/resources/docs/maui/mauiadmin.php

3. Cluster Resources, “TORQUE Admin Manual version 3.0”,

http://www.clusterresources.com/torquedocs21/

4. The Apache Software Foundation, “HDFS User Guide.”,

http://hadoop.apache.org/common/docs/r0.18.3/hdfs_user_guide.pdf

5. Oracle Corporation, “Oracle VM VirtualBox”,

http://www.virtualbox.org/manual/UserManual.html

6. Ubuntu documentation, “Setting up NFS – HowTo”,

https://help.ubuntu.com/community/SettingUpNFSHowTo

7.

HiPerDNO project deliverable, D3.1.1, “Detailed specification report on HPC architecture and platform standardisation to support the development of novel DMS functionality”

http://www.hiperdno.eu (www.virtualbox.org (www.ubuntu.com r www.cloudera.com https://help.ubuntu.com/community/SettingUpNFSHowTo https://ccp.cloudera.com/display/CDHDOC/CDH3+Installation http://www.adaptivecomputing.com/resources/docs/maui/mauiadmin.php http://www.clusterresources.com/torquedocs21/ http://hadoop.apache.org/common/docs/r0.18.3/hdfs_user_guide.pdf http://www.virtualbox.org/manual/UserManual.html

References

Related documents

The PROMs questionnaire used in the national programme, contains several elements; the EQ-5D measure, which forms the basis for all individual procedure

The threshold into the stadium is through a series of layers which delaminate from the geometry of the field to the geometry of the city and creates zones of separation,

more than four additional runs were required, they were needed for the 2 7-3 design, which is intuitive as this design has one more factor than the 2 6-2 design

Lincoln University &gt; Understanding Copyright https://ltl.lincoln.ac.nz/advice/copyright/ or. http://library.lincoln.ac.nz/About/Copyright/ Creative

[r]

Submicrometer period silicon diffraction gratings by porous etching, to be published in

ROB GALLAGHER RESEARCH DIRECTOR Topics: Analytics Big Data Billing systems Business intelligence Digital asset management Digital rights management (DRM) Information management

By comparing different groups of retirees, the paper shows that understanding the nature of retirement, and in particular the extent to which retirement is likely to be