Nowadays, cloud services are widespread. They are applicable not only for commercial purposes, but also for educational and scientifi c purposes. Until recently, grid infrastructures were used in scientifi c and academic fi elds to solve scientifi c problems, to provide storage for storing sensitive data and to provide computing resources. With the spread of cloud computing, a natural transition from grid to cloud technology has taken place. An example of this is the European Grid Infrastructure (EGI) -scientifi c grid infrastructure, which serves the needs for computational resources of the wide range of scientifi c projects and experiments. Since May 2014, the EGI infrastructure migrates its services to EGI, federated cloud and EGI data centers [1].

While grid is based on the resource sharing across boundaries, cloud is based on resource virtualization. In the last few years, cloud computing has also entered as a preferred approach for organizing the IT services of many universities. Furthermore, cloud computing provides opportunities of utilization of computing resources for teaching and learning purposes, for increased collaboration and for resource sharing [2].

Another direction in cloud computing, which is emphasized in [3] is related to the integration of various cloud infrastructures. The relevance of such direction in the academic fi elds is the need of controlled data sharing, access to additional computing resources and software.

In the context of these two directions: cloud integration and education, the experimental cloud of Faculty of Mathematics and Informatics at University of Sofi a (BG-SU-FMI) is presented.

The cloud of Faculty of Mathematics and Informatics at University of Sofi a (BG-SU-FMI) is based on the open source platform OpenNebula. The choice for this cloud platform follows the requirements for integration with similar structures at national and international level. The architecture of the platform contains of two main components:  Front-end node (FN), which consists of OpenNebula Core and web-based graphical user interface -OpenNebula Sunstone for access to the cloud services;  Cluster nodes (CNs) -physical servers, which are containers for virtual machines (VM) created by the user. The OpenNebula cloud platform provides two types of virtualization KVM (Kernel-based Virtual Machine), which is a full virtualization solution for Linux and vCenter, which is a VMware utility used for VM management. For the purposes of the funding research projects the experimental cloud BG-SU-FMI was established, based on the KVM virtualization. On Fig. 1 is shown BG-FMI-SU architecture. We have 3 servers with 16 cores per server with 2 threads per core -96 CPUs in total. On every server, we reserve 2 CPUs for the host that cannot be used for creation of VM. The total number of VM, which is allowable, is 90. All technical information for BG-SU-FMI cloud creation is generalized in Table 1. This cloud is open for integration with the clouds of member's states. Currently, the BG-SU-FMI cloud is in progress of technical integration with JINR cloud. There are two supported approaches for integration with the JINR cloud.

The fi rst approach -cloud bursting (Fig. 2) uses application-programming interface developed on Ruby (Ruby OCCI -rOCCI) and driver for integration based on peer-to-peer model. This approach was recently used for cloud integration between JINR member states and the JINR cloud [5]. However, the increased interest in member states for cloud integration with the JINR cloud reveals a major drawback of the approach -its low scalability and hard maintains. This obstacle has led of looking for new approaches for integration. The second approach with DIRAC -Distributed Infrastructure with Remote Agent Control is currently used approach for cloud integration with the JINR cloud. It is based on usage of DIRAC framework -a grid middleware (interware) for distributed computing. The advantage of this approach is usage of different computational resources through pilot jobs, good scalability and easier maintains. The source code of DIRAC is extended (VMDIRAC module is added to the framework) in order to provide the needed functionality for members' state clouds integration. In order to enable the LDAP Authentication in OpenNebula (BG-SU-FMI cloud) two fi les have to be modifi ed /etc/one/auth/ldap_auth.conf and /etc/one/ oned.conf and confi gured. This will permit users to have the same credentials as in LDAP and will centralize authentication.

The second step is to create VM templates for the two courses. The minimal requirements for the VM depends from the course. For example, for the Database course VM with one CPU will be suffi cient. For Grid and cloud computing course VM with at least 4 CPUs are equitable. BG-SU-FMI cloud allows every student to start his own VM and to exercise installation of database server or Grid, OpenMP and cloud usage.

Development of our cloud is founded on some ideas presented in [6], but since then more detailed and sophisticated improvements and developments happened in the Grid and clouds worlds. Current developments are given in this paper.

BG-SU-FMI is the cloud of Faculty of Mathematics and Informatics at University of Sofi a. It was applied in education for two university courses. We plan to extend its application and for other university courses. This will require additional resources to be added to the cloud in order to upgrade its abilities. Currently the cloud is in progress of technical integration with the JINR cloud. We are working also in direction of integration and on national level. As participants, we will benefi t from cloud integration with access to the projects' specifi c data, software and computational resources.