Scholarly article on topic 'JINR Cloud Infrastructure'

JINR Cloud Infrastructure Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
Keywords
{"cloud technologies" / virtualization / OpenNebula / OpenVZ / KVM / LizardFS}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — V.V. Korenkov, N.A. Kutovskiy, N.A. Balashov, A.V. Baranov, R.N. Semenov

Abstract Cloud technologies are already wide spread among IT industry and start to gain popularity in academic field. To provide a modern computing facility for JINR users participating in different national and international scientific projects as well as to satisfy the needs of developers (development, testing and debugging various apps in various environments) and system administrators (testing and studying specifics of installation and operation of new apps or verifying updates) a cloud infrastructure was deployed at the Laboratory of Information Technologies of the Joint Institute for Nuclear Research (LIT JINR). Initially having simple setup with single front-end cloud node and few cluster nodes only it is now transformed into robust powerful cloud platform with front-end core services in high availability configuration distributed over two servers including MySQL data base replicas, and cloud data store based on distributed network file system. Users authentication in cloud web GUI via Kerberos, a custom driver for OpenVZ virtualization support as well as a driver for integration with other private clouds on infrastructure level and visualization of statistics on cloud resources usage have been developed in-house for our own needs. The JINR cloud is used not only by JINR users, developers and system administrators to fulfill JINR commitments in different national and international projects but also as a computing facility in BES-III experiment.

Academic research paper on topic "JINR Cloud Infrastructure"

Procedia Computer Science

Volume 66, 2015, Pages 574-583

YSC 2015. 4th International Young Scientists Conference on Computational Science

JINR cloud infrastructure

V.V. Korenkov1,2, N.A. Kutovskiy1,2, N.A. Balashov2, A.V. Baranov2,

R.N. Semenov1,2

1Plekhanov Russian University of Economics, Moscow, Russia. Laboratory ofInformation Technologies, Joint Institute for Nuclear Research, Dubna, Russia korenkov@jinr.ru, kut@jinr.ru, balashov@jinr.ru, baranov@jinr.ru, roman@jinr.ru

Abstract

Cloud technologies are already wide spread among IT industry and start to gain popularity in academic field. To provide a modern computing facility for JINR users participating in different national and international scientific projects as well as to satisfy the needs of developers (development, testing and debugging various apps in various environments) and system administrators (testing and studying specifics of installation and operation of new apps or verifying updates) a cloud infrastructure was deployed at the Laboratory of Information Technologies of the Joint Institute for Nuclear Research (LIT JINR). Initially having simple setup with single front-end cloud node and few cluster nodes only it is now transformed into robust powerful cloud platform with front-end core services in high availability configuration distributed over two servers including MySQL data base replicas, and cloud data store based on distributed network file system. Users authentication in cloud web GUI via Kerberos, a custom driver for OpenVZ virtualization support as well as a driver for integration with other private clouds on infrastructure level and visualization of statistics on cloud resources usage have been developed in-house for our own needs. The JINR cloud is used not only by JINR users, developers and system administrators to fulfill JINR commitments in different national and international projects but also as a computing facility in BES-III experiment.

Keywords: cloud technologies, virtualization, OpenNebula, OpenVZ, KVM, LizardFS

1 Introduction

As it is known a cloud computing is a "model for enabling ubiquitous network access to a shared pool of configurable computing resources" [1]. To provide a cutting-edge computing facility for JINR users participating in different national and international scientific projects as well as to satisfy the needs of JINR developers (development, testing and debugging various apps in various environments) and system administrators (testing and studying specifics of installation and operation of new apps or testing updates) a cloud infrastructure (hereinafter referred to as "JINR cloud service", "cloud service", "cloud") was deployed at the LIT JINR. Another motivation for building JINR cloud was to increase

574 Selection and peer-review under responsibility of the Scientific Programme Committee of YSC 2015

© The Authors. Published by Elsevier B.V.

doi: 10.1016/j.procs.2015.11.065

an efficiency of the overall LIT IT infrastructure functioning (more efficient servers and services management, better hardware utilization, higher services and storage systems reliability) with help of virtualization what cloud technologies are relied on.

JINR cloud service is built upon an Infrastructure as a Service (IaaS) model. Such model provides network access to computational, software and information resources (networks, servers, storage devices, services and application software), allowing to allocate resources on-demand according to dynamically changing requirements: cloud users can obtain, configure and deploy VMs themselves with the minimal assistance of the IT specialists. The cloud service is expected to reduce the total cost ownership of the computing infrastructure and also to reduce its support complexity. To tune cloud service for our users', administrators' and managers' needs some in-house developments were done. All this aspects are described below.

2 Service implementation

First attempts to evaluate cloud technologies to be applied for solving LIT and JINR tasks had been performed in 2011. To that moment a distributed training and testing grid infrastructure [1] (t-infrastructure for short) deployed at LIT had been running already for a several years. All services of the t-infrastructure were deployed inside OpenVZ [2] containers. Since apart from the motivation mentioned above it was planned to migrate all of them into the cloud then one of the key requirements for the cloud software was the OpenVZ support or at least a simple way to develop and implement a driver for it. Other criteria were the following:

• stable, feature-rich and actively developing software with Linux compatibility;

• freeware with modular and extensible architecture as well as open source code;

• robust and helpful community with ability to get qualified technical support from developers in a reasonable amount of time;

• clear and complete documentation for administrators, developers and users.

Taking all these criteria into account as well as basing on cloud software reviews the OpenNebula [3] software was chosen as a platform for JINR cloud.

JINR cloud infrastructure consists of the several major components:

• cloud core service,

• cloud scheduler,

• MySQL database back-end,

• user and API interfaces,

• cluster nodes (CNs) where virtual machines (VMs) are running.

All these components can be deployed in various combinations including hosting all of them on a single server (virtual or physical one) or separating over a set of nodes.

Each JINR cloud CN has enabled either OS-level virtualization based on OpenVZ (for fast and lightweight Linux containers) or full virtualization based on KVM [4] (for the cases when the access to kernel is required from the inside VM or non-Linux OS needs to be run).

There are two types of cluster nodes:

• servers with two hard disks configured as RAID1 ("mirror" mode) for highly reliable VMs (e.g. production instances of IT services);

• servers with single hard disk for regular VMs (tests, training, research, etc).

At the moment JINR cloud consists of 40 servers with various hardware characteristics in terms of CPU, RAM and disk size. Total number of cores is 202 and total RAM capacity is 412 GB. The number of registered users is 74 having 81 running VMs.

2.1 High-availability cloud setup

Initially the JINR cloud had simple setup: OpenNebula core, scheduler and interfaces were deployed on single server which was a virtual machine itself hosted on one of the CN [5].

High demand in cloud resources and thus a necessity to extend its capacity as well as a need to reduce downtime of core OpenNebula services and to increase JINR cloud availability and reliability

JJ^JR private subnets

JINR gateway

Figure 1: A schema of JINR cloud transformation from simple setup into high available configuration with storage based on distributed network file system

for users motivated to re-organize JINR cloud into high-available OpenNebula cluster with more computational and storage resources. A schema of JINR cloud transformation from simple setup into HA configuration is shown on Figure 1.

Two cloud front-end nodes with identical set of such components as cloud core, scheduler, MySQL database server, graphical web-interface "Sunstone" (accessible at URL http://cloud.jinr.ru) and rOCCI-server are running on two VMs (FN1 and FN2) hosted on two separated physical servers. All listed above services are running at the same time only on a single FN (the active one) excepting MySQL DB instances which are running on both active and passive FNs in a master-master mode to keep the information in DBs of both FNs consistent. When the active node fails the passive one takes control. All running cloud services and interfaces stay accessible at the same IP address and URLs. Thus a change of active FN is totally transparent for cloud users.

All cloud servers are connected to the JINR private subnet (it is for the whole internal cloud traffic like VM images transfers, storage nodes synchronization, etc) and to the JINR public subnet (excepting storage nodes) which is intended for users interactions with cloud web-GUI as well as for the whole VMs public traffic.

2.2 Distributed network file system

System and image datastores of HA-cloud are hosted on the distributed network file system (DNFS) based on LizardFS [6] and deployed on storage nodes (SNs). Total amount of its disk space is 16 TB but because of enabled redundancy = 2 the effective size is about 8 TB. In initial JINR cloud setup all datastores were deployed on FN local disk drive with 500 GB size and it was impossible to extend storage capacity dynamically by adding more disk drives without stopping the server with FN deployed on it. LizardFS provides a possibility to add more disks and servers "on the fly".

Because of not fast enough private network speed (1 Gbit/s only) all VMs on CNs are running from local copy of images which are copied from LizardFS-based storage nodes (SNs) to CNs and back. The usage of DNFS as a storage backend for OpenNebula datastores allows to extend a disk space dynamically as well as to increase a reliability of VM images storage because of configurable redundancy parameter in LizardFS.

A schema of LizardFS-based storage with each server roles is shown on the Figure 2.

JINR private subnet.

master, chunk- metalogger and shadow and

server and web- chunk-server chunk-server

based admin GUI

Figure 2: A schema of LizardFS-based storage deployed at JINR

2.3 Cloud and VM access

Two user interfaces are available to access the JINR cloud:

• command line interface (CLI);

• web-based graphical user interface "Sunstone" (either simplified or full-featured one depending on the user group he belongs to).

The service can be used by authenticated users only. Currently Sunstone is configured to authenticate users registered in JINR user database (JINR user DB for short). Since Kerberos is widely used at JINR for user authentication in various IT services it was highly desirable to have such type of user authentication in JINR cloud web-GUI enabled too. So such feature was developed in-house since Sunstone doesn't support Kerberos authentication out of the box.

To secure an information exchange between users' web-browser and JINR cloud web-interface a SSL encryption was enabled with use of SSL certificate on the web-server.

Initially user has zero quotas on cloud resources and hence (s)he can't create VMs. To get some resource the user has to send a corresponding request via in-house developed and integrated into Sunstone a web-form available in the main menu (the screenshot with such web-form for cloud resources request is shown on the Figure 3).

The user has to fill in all fields such as his first and last names, own email address, supervisor's full name and his email address, choose a lab which user represents, a topic from topical plan for JINR research and international cooperation, number of total cores, RAM and storage as well as number of VMs, operating system type (Linux or Windows) and a purpose these VMs will be used for. By pressing "Send" button all that info is submitted to JINR cloud manager who is responsible to dispose

of cloud resources. The user gets a notification via email if his request was accepted and what quotas he receives or if his request was rejected with a reason specified.

All VM images available in JINR cloud for users are customized with Kerberos authentication enabled by default. It means that user can log into his running VM using own Kerberos login and password. Then to gain a superuser privileges user needs to execute 'sudo su -' command in the VM shell.

Another option for users to get root access inside own running VM is to supply rsa/dsa-key in VM template before instantiating VM and log in that VM via ssh protocol as a root user.

OpenNebula Qf Request resources

Dashboard fe Virtual Resources fîi Infrastructure "!■ Marketpiace ¿§3 OneFlow

Qf Request resources

I,III Statistics

0 General Information

\ Required Resources

Manager's full name

Manager's e-mail

Laboratory: vblhep

Topic number: No topic Details on the listed topics see here

I CPUs (cores), total per all VMs RAM (GB), total per all VMs Storage (GB), total per all VMs

Number of virtual machines

OS type: Linux

Comment (purpose of the requested resources or reason for quotas change)

Figure 3: A screenshot of web-form for JINR cloud resources request

2.4 Cloud infrastructure monitoring

To monitor cloud servers and most critical cloud components a dedicated monitoring service [7] based on Nagios monitoring software [8] was deployed. In case of any problem with monitored objects a cloud administrators get notifications via SMS and email.

The state of virtual machines is monitored currently by OpenNebula, some their parameters are tracked by built-in monitoring system and its information is available on the Sunstone dashboard.

3 Cloud bursting

Private clouds usually have much smaller computing capacities than commercial public ones. With growing demands there is a chance that a small private cloud would eventually run out of its internal

resources and wouldn't be able to deal with peak load at some point. One of the ways to solve the problem is to outsource the workload to an external cloud provider: it could be a commercial cloud service, such as Amazon EC2 and Rackspace, or a partner infrastructure providing application programming interfaces (APIs). This concept is called "cloud bursting". When a private cloud grants the ability to cloud burst to a public cloud it becomes hybrid.

JINR cloud service quickly became overprovisioned, not able to meet fast-growing demands. The decision was made to cooperate with some clouds of partner organizations having plenty of spare resources and to try to integrate with such clouds using cloud bursting concept.

OpenNebula platform, which JINR cloud is based on, has a built-in cloud bursting mechanism with a set of drivers for some popular platforms (Amazon EC2, IBM SoftLayer and Microsoft Azure) and being open-source it allows to develop any platform-specific driver. JINR cloud team developed cloud bursting driver to integrate JINR cloud with partner clouds either OpenNebula-based one (and in that case it is possible to enable real time external cloud resources monitoring) or any other cloud platform which supports Open Cloud Computing Interface (OCCI) [9] but then statically defined values of external cloud resources are possible only.

The driver is based on ruby implementation of OCCI (rOCCI) and OpenNebula XML-RPC interface. The reason why both interfaces are used at the same time is that OCCI allows to better reuse code since it is provided by other popular cloud platforms like OpenStack while OpenNebula XML-RPC is OpenNebula-specific and can't be used to integrate with other platforms. At the same time OCCI doesn't provide monitoring operations while OpenNebula XML-RPC does. It was decided to use rOCCI for management operations and XML-RPC for monitoring.

Developed at JINR cloud bursting driver for OpenNebula was successfully tested and made possible to link two OpenNebula testbeds deployed at JINR cloud. The next step is to use that driver to connect with partner clouds (more on that is in "Future plans" section).

4 Service usage

4.1 Deployed components

Currently the JINR cloud usage is developed in three directions:

• test, educational and research tasks as part of participation in various projects using cloud and grid technologies;

• systems and services deployment with high reliability and availability requirements;

• extension of computing capacities of the grid-infrastructures.

Services and testbeds currently deployed in JINR cloud are listed below (see Figure 4 also).

1) EMI-based testbed (it is used for trainings, testing, development and research tasks related to grid technologies as well as for performing JINR obligations in local, national and international grid projects such as e.g. WLCG [10]);

2) PanDA testbed which is used for PanDA software validation and extensions development for ATLAS [11] and COMPASS [12] experiments;

3) DIRAC-based testbed (it is used for monitoring tools development for BES-III experiment [13] distributed computing infrastructure as well as its computing facility);

4) a set of VMs of NOvA [14] users for analysis and software development;

5) NICA [15] testbed for grid middleware evaluation for NICA computing model development;

6) HybriLIT [16] services:

• Indico - a service for event organization, archival and collaboration;

• FreeIPA - an integrated identity and authentication service used for user accounts management in HybriLIT;

• Git - local Git repository (well-known distributed revision control system) for hybriLIT team;

Helpdesk (a web application for the day to day operations of an IT environment including a user technical support of JINR IT services);

DesktopGrid testbed based on BOINC software (to estimate the volunteer' computing technology for possible use in solving JINR users' tasks);

web-sites

Test GitLab helpdesk development

JDS OpenNebula

development testbed l

HybriLIT services

NICA testbed

• ч/т т э'з'а'э' - 1 ' • ù-1

testbed

OpenNebula development testbed N

PanDA testbed

ÛRU-JINR

„ ШШ «Í.BD.I

J i' tJ

IFC Eta WMS • LU

ÛÛÛ

CREAM ECGI e>:

HEPweb

PC BONC «ли » ЗО-ГлО^е

JINR cioud

едите

DesktopGrid

Figure 4: A schema of services and testbeds deployed in JINR cloud

9) web-service HepWep [17] (provides a possibility to use different tools for Monte-Carlo simulation in high-energy physics);

10) test instances of the JINR document server (JDS) [18] and JINR Project Management Service (JPMS) [19];

11) VM for web-sites development including new JINR web-portal;

12) JINR GitLab - local GitLab installation [20] for all JINR users;

13) A set of users' VMs which are used for their own needs.

The JINR cloud is used not only by JINR users, developers and system administrators to fulfill JINR commitments in different national and international projects but also as a computing facility in BES-III experiment. In order to let its computing infrastructure services to interact with JINR cloud and run computing jobs there the rOCCI-server was deployed on FN and configured to use our cloud on behalf of dedicated bes3 user within some computing resources share.

Moreover a number of OpenNebula testbeds are deployed in the JINR cloud to develop and debug for current and new OpenNebula software releases all in-house implemented components such as OpenVZ and cloud bursting drivers, resources accounting and Kerberos authentication plugins. Each

of such testbeds consists of 1 OpenVZ container as FN and 2-3 KVM VMs as CNs with OpenVZ kernel installed inside them.

4.2 Training on cloud technologies

Since the beginning of 2014 when JINR cloud had been launched a several training courses on OpenNebula cloud usage and administration have been conducted at JINR upon users' or organizations from JINR member states requests (see Table 1)._

Organization Organization Training dates Number of Training type

location trainees

Institute of Experimental Prague, Czech 07-10.07.2015 2 usage

and Applied Physics, Czech Republic

Technical University

Egyptian scientific Egypt 05-09.06.2015 3 usage

organizations

JINR Dubna, Russia 26-27.01.2015 11 usage

Gdansk university of Gdansk, Poland 06.10-12.12.2014 1 usage and

technologies administration

National Scientific and Minsk, Belarus 22-29.09.2014 3 usage and

Educational Centre of administration

Particle and High Energy Physics of the Belarusian

State University_

Table 1: Information on conducted trainings on cloud technologies

4.3 Resources utilization statistics

To keep track on cloud resources utilization a one more in-house add-on for OpenNebula was developed. It provides a possibility to view such statistical information as CPU and RAM distribution over cloud clusters, resources utilization by users, departments or projects, and so on (see screenshot on Figure 5). Such statistical information is available in Sunstone menu but for cloud users with special rights only.

5 Future plans

Some work on JINR cloud deployment and development was partially supported by Russian Foundation for Basic Research (RFBR) grant #14-07-90405 "Cloud computing technologies development and integration into BITP and LIT JINR Tier-2 grid sites for ALICE experiment data processing". So to meet JINR and Bogolyubov Institute for Theoretical Physics of the National Academy of Sciences of Ukraine - BITP (Kiev, Ukraine) commitments scheduled on 2015 year it is planned to integrate JINR cloud with BITP one as well as with EGI Federated cloud [21]. Apart from that a similar integration with Institute of Physics of Azerbaijan National Academy of Sciences - IP (Baku, Azerbaijan) and Plekhanov Russian University of Economics - PRUE (Moscow, Russia) clouds is going to be implemented. Some part of such joint cloud resources is planned to be used for ALICE [22] and other experiments (e.g. BES-III) needs.

6 Conclusion

The JINR cloud service made possible to provide a modern computing facility for various groups of users and allowed to better utilize hardware resources. It also significantly simplified the job of system administrators by automating many virtual machines management tasks and by giving the users the ability to create and manage VMs by themselves within the limit of the granted quotas.

The service is actively used to cover users' demands as well as to carry out JINR commitments in various local, national and international projects.

Migration to HA-cloud setup led to increasing the service reliability and uptime.

Figure 5: A screenshot of Sunstone menu with graphics of CPU and memory distribution over JINR cloud

clusters

Open source code of OpenNebula platform as well as its flexibility and extensibility allowed to fine-tune JINR cloud service for local environment by developing in-house add-ons and drivers.

In the nearest future it is planned to make the integration with private clouds of partner organizations and international cloud infrastructure.

7 Acknowledgement

Some work on JINR cloud deployment and development was partially supported by Russian Foundation for Basic Research (RFBR) grant #14-07-90405 "Cloud computing technologies development and integration into BITP and LIT JINR Tier-2 grid sites for ALICE experiment data processing".

References

[1] R. Hristova, N. Kutovskiy, V. Dimitrov, V. Korenkov "Distributed training and testing grid infrastructure evolution" in 8th International Conference «Information Systems & GRID Technologies», Sofia, Bulgaria, 2014.

[2] "OpenVZ project" [Online]. Available: http://openvz.org/Main_Page.

[3] "Web-portal of OpenNebula project" [Online]. Available: http://opennebula.org.

[4] "KVM project" [Online]. Available: http://www.linux-kvm.org/page/Main_Page.

[5] A.V. Baranov, N.A. Balashov, N.A. Kutovskiy, R. N. Semenov "Cloud Infrastructure at JINR" Computer Research and Modeling, vol. 4, no. 3, pp. 463-467, 2015.

[6] "LizardFS project" [Online]. Available: http://lizardfs.com.

[7] "Nagios-based monitoring service for JINR cloud servers and services" [Online]. Available: http://cloud-mon.jinr.ru/nagios.

[8] "Web-portal of Nagios monitoring system" [Online]. Available: https://www.nagios.org.

[9] "OCCI working group web-site" [Online]. Available: http://occi-wg.org.

[10] "Web-portal of Worldwide LHC Computing Grid" [Online]. Available: http://wlcg.web.cern.ch.

[11] "Web-portal of ATLAS experiment" [Online]. Available: http://atlas.ch/.

[12] "Web-portal of COMPASS experiment" [Online]. Available: http://wwwcompass.cern.ch/.

[13] "Web-portal of BES-III experiment" [Online]. Available: http://bes3.ihep.ac.cn.

[14] "Web-portal of NOvA experiment" [Online]. Available: http://www-nova.fnal.gov/.

[15] "Web-portal of NICA facility" [Online]. Available: http://nica.jinr.ru.

[16] "Web-portal of HybriLIT computation facility" [Online]. Available: http://hybrilit.jinr.ru.

[17] "HepWeb service" [Online]. Available: http://hepweb.jinr.ru.

[18] "JINR document server" [Online]. Available: http://jds.jinr.ru.

[19] "JINR project management service" [Online]. Available: http://pm.jinr.ru.

[20] "JINR GitLab installation" [Online]. Available: http://git.jinr.ru.

[21] "Web-portal of EGI Federated cloud" [Online]. Available: https://www.egi.eu/infrastructure/cloud.

[22] "ALICE experiment web-portal" [Online]. Available: http://aliceinfo.cern.ch.