=Paper=
{{Paper
|id=Vol-3041/117-121-paper-21
|storemode=property
|title=Research Cloud Computing Ecosystem in Armenia
|pdfUrl=https://ceur-ws.org/Vol-3041/117-121-paper-21.pdf
|volume=Vol-3041
|authors=Hrachya Astsatryan,Wahi Narsisian,Artashes Mirzoyan,Vladimir Sahakyan
}}
==Research Cloud Computing Ecosystem in Armenia==
https://ceur-ws.org/Vol-3041/117-121-paper-21.pdf
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
RESEARCH CLOUD COMPUTING ECOSYSTEM IN
ARMENIA
H. Astsatryana, W. Narsisian, A. Mirzoyan, V. Sahakyan
Institute for Informatics and Automation Problems of the National Academy of Sciences
of the Republic of Armenia, 1, Paruyr Sevak str., Yerevan 0014, Armenia
E-mail: a hrach@sci.am
Growing needs for computational resources, data storage within higher-educational institutions and the
requirement for a lot of investment and financial resources the idea or the concept of “National
Research Cloud Platform (NRCP)” is crucial to provide necessary IT support for educational, research
and development activities, which allow access to advanced IT infrastructure, data centers, and
applications and protect sensitive information. In this article we will illustrate the concept of NRCP,
background, deployment stages and architecture and finally some use cases.
Keywords: IaaS, NRCP, Openstack, ArmCloud, ArmCluster, ArmGrid, Earth science, Life Science,
VM
Hrachya Astsatryan, Wahi Narsisian, Artashes Mirzoyan, Vladimir Sahakyan
Copyright © 2021 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
117
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
1. Introduction
Virtualization transforms the IT industry landscape providing capabilities to run various
virtual machines (VM) in the same hardware capacities, enhancing resource sharing and improving
performance [1]. Low overhead costs for implementing this technology, high and constantly growing
demand for computing resources and the need to provide more flexible services have led to the
transition from the use of bare-metal servers and towards providing of virtualized resources (virtual
machines, storage and even network infrastructures) that are easier to scale and provide a sufficient
level of reliability. On the other hand, the Cloud computing environment has proven to be the base of
these changes, which increased the Cloud services and computing resources requirement throughout
scientific institutions and universities [2]. The term started use by Amazon Company in 2008. Later,
this novel technology was developed and provided as a service by GAFA (Google, Apple, Facebook,
and Amazon) and other public cloud providers [3]. The main approach of Public cloud providers is to
deliver on-demand services through the Internet to anyone who registers and pays for the services.
Instead of public clouds, the private cloud infrastructures built-in for a couple of institutions or
companies to host the facilities on their side [4]. For instance, national research cloud platforms
provide cloud services to the academic and research community on the top of the research and
education networks. It is possible to combine public and private cloud deployment models to create a
synergy, calling hybrid cloud. Usually, the resources of Public clouds are supported the elasticity of
computational resources in case of the need.
Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service
(SaaS) are the leading cloud computing service layers [5]. IaaS provides the infrastructure such as
VMs and other resources like VM disk image library, block and file-based storage, firewalls, load
balancers, IP addresses, or virtual local area networks. PaaS or platform as a service model delivers
computing platforms typically including an OS, programming language execution environment,
database, and web server. SaaS provides a new service delivery model to access application services
over the web without worrying about installing, maintaining, or coding the software. The SaaS
provider manages the software maintenance or setup. Therefore, the software is available to access and
operate without downloading or installing any piece of design or OS.
In addition to the leading cloud computing service layers, a wide range of services can be
provided by Cloud, with an extra layer of flexibility and scalability, such as provisioning high-demand
virtual HPC resources [6]. The critical challenge of deploying such cloud services is the complexity
and the cost to purchase and maintain the computing resources needing a lot of human efforts to keep
all the services up to date and reliable. The cost is a significant limitation for developing countries,
like in the case of Armenia. In 2018, the Institute for Informatics and Automation Problems of the
National Academy of Sciences of the Republic of Armenia (IIAP) started to realize the “National
Research Cloud Platform (NRCP)” initiative. NRCP aims to deliver on-demand cost-effective cloud
computing resources and services to the local institutions and research communities. The market
analyzes with scientific communities and stakeholders aimed to find out the demands and complexity
of scientific problems facing to solve, and gather the information related to the communities’ tools and
packages. As a result, IIAP deployed user-oriented Cloud services to fulfill almost all types of
demands in Armenia, ranging from general to domain-specific services. Section 2 represents the
architecture and design of NRCP, while the conclusions and lessons learned follow in Section 3.
2. National research cloud platform
In a first stage, a federated cloud infrastructure in the Black Sea Region has been deployed,
enabling user communities from participant countries (Armenia, Georgia, Moldova, and Romania) to
join the local virtualized resources providing them with VMs, networks, and storages [7]. The
federated infrastructure offers user communities to use local or remote resources and makes user
communities’ regional collaboration easier. The federated cloud platform based on OpenNebula
middleware address the regional problems that require large amounts of computational resources, even
when the actual simulations happen not in the zone where the data is stored. In the next stage, the
118
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
virtualization is widely implemented for the core services of the Armenian National Grid (ArmGrid)
infrastructure providing on-demand access to a sustainable computing platform [8]. The ArmGrid
infrastructure consists of seven Grid sites located in the leading research centers and universities of
Armenia. The total number of processors at the Grid sites was approximately 450 CPU cores. Instead
of a single system ArmCluster (Armenian Cluster), ArmGrid is an autonomous decentralized system
with distributed job management and scheduling opportunities [9]. Finally, a hybrid research
computing platform has been deployed combing HPC with Grid and Cloud Computing based on
ArmCluster HPC cluster, resource sharing ArmGrid Grid, and on-demand service provisioning
federated cloud infrastructures. Each infrastructure identifies rules for making up the resources and
executing applications, like resource ownership and sizing, application portability, or resource
allocation policy. Based, on these experimental infrastructures, a NRCP is suggested in Armenia aims
to have better hardware utilization, increase the storage systems reliability and services management,
and offer higher services with the virtualization support. The infrastructure provides VMs and
networking services, consists of a cloud core service and a scheduler, APIs, databases, and nodes
where VMs are running. The full virtualization on the KVM hypervisor for each computational node
has been implemented. NRCP consists of three different Zones of Cloud resources, GPU resources,
and a data lake [fig. 1]. Combining these three solutions under a single umbrella provides domains
specific services with high-availability and scalable services. The NRCP is a critical element of the
Armenian e-infrastructure [10], a complex national IT infrastructure consisting of both communication
and distributed computing infrastructures. Most importantly, all the input and output data reside on the
NRCP side to reduce the time to process the data and possibly is shared data between different
scientific groups.
Figure 1. National research cloud platform
NRCP architecture design is mainly built on multiple Cloud controllers dedicated to different
scientific communities to split the Cloud resources and the Cloud storage based on several scientific
domains. The technical information of computational resources is generalized in Table 1.
Table 1. NRCP technical specification
Server parameters Total
Server type Q-ty CPU/GPU model
CPU/GPUs Cores RAM (GB) cores
Thin 4 Intel Xeon E5-2630 v4 2 20 256 80
Fat 2 Intel Xeon Gold 6138 4 80 512 160
2 Intel Core i9-10900KF 1 10 128 20
Intel Xeon E5-2680 v3 2 24 48
2 128
Accelerated Intel Xeon Phi 7120P 2 122 244
Intel Xeon Gold 5218 2 32 64
2 192
Nvidia V100 32GB 2 10240 20480
Total (cores) 21096
119
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
Therefore, NRCP provide compute services consist of 616 physical and 20480 GPU cores,
about 3 terabytes of memory and 1620 terabytes for data storage. For instance, all Earth science
production groups are consolidated under a single Zone with the same storage node, enabling them to
share data if needed very easily and opens the door to a better collaboration possibility.
The OpenStack has been deployed and customized for NRCP providing orchestration needed
to virtualize servers, storage, and networking. The Controller, compute, network, and storage
components have been used for the deployment. Various flavors of VMs ranging from small instances
such as 2 CPU and 2 GB RAM with 40 GB HDD, to a very huge instance with 128 CPU cores, 256
GB. The Data Lake provides a scalable and secure platform that allows all users to upload and
download their data with a high-speed, process the data in real-time, use the data for different
simulations, share the data between groups. For instance, in the domain of astrophysics, the
infrastructure’s core is the Armenian VO repository providing an advanced experimental platform for
data archiving, extraction, acquisition, correlation, reduction, and use [11]. Another example is the
Armenian Data Cube [12], a complete and up-to-date EO (Earth Observation) archive data (e.g.,
Landsat, Sentinel). EO, using precise and reliable data, is a critical element to address different
environmental challenges, like water, soil, or plants.
For the last three years NRCP serves multiple scientific projects and communities. For
instance, several domain specific services have been developed for earth science user community to
address critical societal challenges, such as weather prediction, air quality monitoring and prediction,
water quality and quantity monitoring, or earth observation [13-14].
3. Conclusion and lessons learned
The article summarizes the experiences gained so far, and highlighted a few scientific use
cases, where the community is intensively using NRCP resources. Throughout the deployment and the
implementation phases of NRCP deployment for diverse scientific communities with specific domain-
oriented approaches, a list of recommendation has been collected:
To consider the complete infrastructure and its capabilities before the deployment, boosting to
choose the best possible options and tools satisfying the needs;
To conduct benchmarks and experiments before putting the solution into production; to
confirm the systems' reliability by handling different scenarios, even if the deployment of
some packages needs to be done several times;
To prepare a well-documented tutorial with exact details of all services and solutions,
considering that not every user has an IT background when using the system;
To conduct a training campaign with potential communities and explain the opportunities and
challenges. It will help to understand the benefits of such solutions by boosting scientific
experiments and simulations;
To minimize the manual deployment as much as possible. For instance, it is planned to
implement multiple bash scripts and Puppet automation;
To maximize the overall resource usage in computing, networking, and storage resources
considering the energy consumption minimization;
To use federated identity authentication based on SAML 2.0 (Security Assertion Markup
Language) to make easy user access.
It is planned to develop and provide user-specific high-level services, like SaaS solutions for
all those communities, conducting the experiments without accessing the computing resources.
Instead, the communities may use the browser to access any domain-specific service and run the
experiment from it, further simplifying cloud resource usage. The OpenStack Ironic will be
implemented for economic and most efficient use of computing resources focusing on HPC Cloud
solutions provisioning based on completely virtual, bar-metal, and hybrid architectures. The future
ultimate goal is the establishment of a National Open Science Cloud Initiative and its further
integration with the European Open Science Cloud and European Research Infrastructures, like
ELIXIR or Openaire.
120
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
References
[1] M.F. Mergen, V. Uhlig, O. Krieger, J. Xenidis. Virtualization for high-performance computing //
ACM SIGOPS Operating Systems Review, 2006 Apr 1, vol. 40, issue 2, pp. 8-11
[2] M. Armbrust, et al. A view of cloud computing // Communications of the ACM, 2010 Apr 1, vol.
53, issue 4, pp. 50-58
[3] V. Chang, G. Wills, D. De Roure. A review of cloud business models and sustainability // In2010
IEEE 3rd International Conference on Cloud Computing 2010 Jul 5, pp. 43-50
[4] Y. Jadeja, K. Modi Cloud computing-concepts, architecture and challenges // In2012
International Conference on Computing, Electronics and Electrical Technologies, 2012 Mar 21, pp.
877-880
[5] S.K. Sowmya, P. Deepika, J. Naren. Layers of Cloud–IaaS, PaaS and SaaS: A Survey //
International Journal of Computer Science and Information Technologies. 2014, vol. 5, issue 3, pp.
4477-4480
[6] R.R. Expósito, et al. Performance analysis of HPC applications in the cloud // Future Generation
Computer Systems, 2013 Jan 1, vol. 29, issue 1, pp 218-229
[7] H. Astsatryan, , et al. Environmental science federated cloud platform in the bsec region //
International Journal of Scientific & Engineering Research, 2014, vol. 1, issue 1, pp. 1130–1133
[8] H. Astsatryan, Yu. Shoukouryan, V. Sahakyan. Grid activities in Armenia // In Proceedings of the
International Conference Parallel Computing Technologies. Novgorod, Russia, March, 2009
[9] H. Astsatryan, Yu Shoukourian, V. Sahakyan. Creation of High-Performance Computation
Cluster and DataBases in Armenia // In Proceedings of the Second International Conference on
Parallel Computations and Control Problems (PACO ‘2004), pp. 466–470, 2004
[10] H. Astsatryan, V. Sahakyan, Yu. Shoukourian, P.H. Cros, M. Dayde, J. Dongarra, P. Oster.
Strengthening Compute and Data intensive Capacities of Armenia // IEEE Proceedings of 14th
RoEduNet International Conference - Networking in Education and Research, Craiova, Romania, pp.
28-33, September 24-26 2015
[11] A.M. Mickaelian, et al. Ten Years of the Armenian Virtual Observatory // ASPC, vol. 505, issue
16, 2016
[12] Sh. Asmaryan, et al. Paving the way towards an armenian data cube // Data, vol. 4, issue. 3, pp.
1–10, 2019
[13] H. Astsatryan, et al. An interoperable cloud-based scientific GATEWAY for NDVI time series
analysis // Elsevier Computer Standards & Interfaces, vol. 41, pages 79-84, 2015
[14] H. Astsatryan, W. Narsisian and Sh. Asmaryan. SWAT hydrological model as a DaaS cloud
service // Springer Earth Science Informatics, vol. 9, issue 3, pages 401–407, 2016
121