Easing Scientific Computing and Federated Management in The

Easing Scientiﬁc Computing and Federated Management in the Cloud with OCCI

Zdenekˇ Sustrˇ 1, Diego Scardaci2,3, Jirˇ´ı Sitera1, Boris Parak´ 1 and V´ıctor Mendez´ Munoz˜ 4 1Department of Distributed Computing, CESNET, Zikova 4, 160 00, Praha 6, Czech Republic 2European Grid Initiative, Science Park 140, 1098 XG Amsterdam, Netherlands 3INFN Sezione di Catania, Via S. Soﬁa 64, I-95123 Catania, Italy 4Computer Architecture & Operating Systems Department (CAOS), Universitat Autonoma` de Barcelona (UAB), Bellaterra, Spain

Keywords: Federated Cloud, Cloud Standards, Cloud Interoperability, Cloud Solution Design Patterns, Cloud Application Architectures, Cloud Middleware Frameworks, Open Cloud Computing Interface.

Abstract: One of the benefits of OCCI stems from simplifying the life of developers aiming to integrate multiple cloud managers. It provides them with a single protocol to abstract the differences between cloud service implementations used on sites run by different providers. This comes particularly handy in federated clouds, such as the EGI Federated Cloud Platform, which bring together providers who run different cloud management platforms on their sites: most notably OpenNebula, OpenStack, or Synnefo. Thanks to the wealth of approaches and tools now available to developers of virtual resource management solutions, different paths may be chosen, ranging from a small-scale use of an existing command line client or single-user graphical interface, to libraries ready for integration with large workload management frameworks and job submission portals relied on by large science communities across Europe. From lone wolves in the long-tail of science to virtual organizations counting thousands of users, OCCI simplifies their life through standardization, unification, and simplification. Hence cloud applications based on OCCI can focus on user specifications, saving cost and reaching a robust development life-cycle. To demonstrate this, the paper shows several EGI Federated Cloud experiences, demonstrating the possible approaches and design principles.

1 INTRODUCTION cloud – for instance tools only available for operating systems not compatible with available grid OCCI, the Open Cloud Computing Interface (Nyren´ infrastructures, tools distributed by vendors in the et al., 2011; Nyren´ et al., 2011; Metsch and Ed- form of virtual machine images, etc. monds, 2011b; Metsch and Edmonds, 2011a), is a Lower the acceptance threshold for users trained standard developed by the Open Grid Forum to stan- • in using their existing environment without hav- dardize virtual resource management in a cloud site ing to change it or re-qualify for a different one. (OGF, 2016). It was released in 2011 and several implementations and real-world deployments followed. On reflection, this suits the need of users in the “long Among others, OCCI also became the standard of tail of science” who typically lack the resources to choice for the management of virtualized resources have their solutions tailored or adjusted to the grid in EGI’s Federated Cloud Platform (Wallom et al., and cannot invest in re-training for different tools or 2015). environments. Even relatively small-scale resources, The EGI Federated Cloud Platform has been con- available in the EGI Federated Cloud Platform at the ceived as an alternative way to access EGI’s consid- time of writing, can fit the needs of these “long-tail” erable computing resources, which are primarily ac- communities. cessed through grid middleware. Cloud-flavoured ser- Firstly, this article will introduce relevant imple- vices are meant to: mentations of OCCI, both on the server and client side Attract user communities relying on tools that are (Section 2). Secondly, Section 3 will introduce a scale • not easily ported to grid but can scale well in the of cloud usage patterns relying on OCCI for interop-

347

Šustr, Z., Scardaci, D., Sitera, J., Parák, B. and Muñoz, V. Easing Scientiﬁc Computing and Federated Management in the Cloud with OCCI. In Proceedings of the 6th International Conference on Cloud Computing and Services Science (CLOSER 2016) - Volume 2, pages 347-354 ISBN: 978-989-758-182-3 Copyright c 2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

OCCI 2016 - Special Session on Experiences with OCCI

erability and abstraction, and also briefly expand on point (Fig. 1). There are several of such server-side other mechanisms users must ensure beyond the ba- frameworks whose resources can be managed through sic OCCI functionality for their solutions to be truly OCCI: productive. Finally, Section 4 will briefly outline new OpenStack, which comes with an OCCI interface developments, and Section 5 will sum up the topic of • (occi-os) as part of the stack and is soon due to experience with OCCI in the area of scientific com- receive an all-new implementation in the form puting. of OOI – the OpenStack OCCI Interface (Garc´ıa et al., 2016). Synnefo with its own OCCI implementation (GR- 2 OCCI IMPLEMENTATIONS • NET, 2016). OpenNebula, which is accessed through an OCCI OCCI was gradually gaining support throughout its • translation service – the rOCCI-server (Parak´ lifetime and there are currently multiple implementa- et al., 2014). tions available on the server side as well as the client side. At the time of writing, the version of OCCI Although there is currently no such site participat- supported by most implementations mentioned below ing in the Federated Cloud Platform, it is also techni- is 1.1. They will all follow their own, independent cally possible to use OCCI to manage resources in a schedules to adopt OCCI 1.2, which is due to come fourth type of cloud management framework: into effect shortly. Amazon Web Services (AWS), which can be ac- • cessed with OCCI through the rOCCI-server as 2.1 Server-side OCCI Support well (CESNET, 2016). Aside from the aforementioned, there are other Server-side OCCI implementations originate mostly cloud services which support OCCI or its subsets from service providers who wish to make their re- (Fogbow, 2016), or which are going to be made sources available in a standardized way. Where appli- OCCI-capable in the foreseeable future. For instance, cable, the OCCI interfaces are being pushed upstream the development work is underway to implement an to Cloud Management Framework developers. OCCI translation layer for Microsoft Azure. In the EGI Federated Cloud Platform, multiple For the sake of completeness, it is also worth sites make their cloud resources available to users. mentioning that there are OCCI implementations cur- Although different contributing sites employ differ- rently being used in existing solutions, but apparently ent Cloud Management Frameworks (CMFs), OCCI no longer maintained. They are: (plus an X.509-based authentication mechanism) is pyssf – the Service Sharing Facility (PySSF, the unifying factor, allowing users to access any site • 2016), which strives to provide building blocks in a uniform way, indeed, without actually know- for grid, cloud, cluster or HPC services. ing what flavor of CMF is installed on the end- occi-os – the original OpenStack OCCI interface • (occi os, 2016), which is gradually being replaced Clients Servers with OOI, already mentioned above. Custom client app. http NativeNative RubyJava occi application rjOCCI-api While the latter is scheduled to phase out in fa- rjOCCI-core vor of OOI, the former – it must be acknowledged –

# Shell script has never been declared as discontinued, and devel- Application or script using rOCCI-cli http opment activity may resume; especially with the in- occi Synnefo command-line rOCCI-api executables troduction of OCCI 1.2 speciﬁcation. rOCCI-core

Custom client app. http OpenNebula Native Ruby occi 2.2 Client-side Tools and Libraries application rOCCI-api r

application e v r

rOCCI-core e Available s

- MS Azure I C C

Independent http O r OCCI is a text-based protocol. Although there are un- OCCI-compliant occi client ofﬁcial extensions to support OCCI transport through message queues (Limmer et al., 2014), the only of- Figure 1: Ecosystem of OCCI implementations envisioned ﬁcially standardized transport method for OCCI is for products relevant to EGI Federated Cloud and its user HTTP. Therefore it is technically possible to interact communities. with an OCCI-capable server through a generic HTTP

348 Easing Scientiﬁc Computing and Federated Management in the Cloud with OCCI

client, and implement at least a subset of OCCI func- around the EGI Federated Cloud. tionality with a few pre-formatted OCCI messages. This approach is often used in interoperability test- 3.1 Science Gateways and Workload ing, where a fixed set of actions is tested over and Management Frameworks over, but is highly unsuitable for real world applications wherein a general-purpose client is required. The great potential offered by OCCI is evident when At present, there are at least two independent high-level tools are built on or connected to its inter- client-side stacks available: the rOCCI framework face. End users of PaaS and SaaS services that ex- and the jOCCI library (Kimle et al., 2015). ploit cloud resources through OCCI can benefit from The rOCCI framework offers a set of Ruby li- a large amount of resources belonging to heteroge- braries (rOCCI-core and rOCCI-api), wherein the neous cloud sites adopting different cloud manage- former implements the OCCI class structure, pars- ment frameworks. ing and rendering, while the latter implements HTTP In the context of the EGI Federated Cloud (Wal- transport. Calls to these libraries may be used by na- lom et al., 2015; del Castillo et al., 2015), this oppor- tive Ruby clients to interact with any OCCI-enabled tunity has been exploited by many technical providers cloud site. who have extended their platforms to support the The rOCCI framework also comes with a general- OCCI standard providing an alternative and auto- purpose command line client – the rOCCI-cli – which mated way to access the Federated Cloud, hiding the can be used by end users to complete simple tasks, or IaaS layer from their users. to wrap around with scripts. Some user communities This has greatly increased the added value that also choose to wrap around the command line inter- the EGI Federated Cloud offers to its users, who can face if the programming language of choice cannot choose to access its resources at IaaS level or via one use Ruby or Java libraries directly. of the various OCCI-compliant tools now available. Finally the jOCCI is an independent OCCI imple- Notable high-level tools currently available in the mentation in Java. Like rOCCI, it exposes the OCCI EGI ecosystem supporting OCCI are: class structure, parsing and rendering functions, plus for PaaS: Vac/VCycle (VCycle, 2016), Slipstream HTTP transport. It is intended primarily for use by • (Slipstream, 2016) and Infrastructure Manager orchestration or submission frameworks implemented (IM) (IM, 2016) in Java. for SaaS development frameworks: VMDirac • (VMDirac, 2016) (also discussed in greater de- tail in 3.1.1), COMPSs (Lezzi et al., 2014), the 3 CLOUD USAGE PATTERNS Catania Science Gateway Framework (Ardizzone WITH OCCI et al., 2012) and WS-PGRADE (WS-PGRADE, 2016). The ability to manage cloud resources in a stan- The Vac/VCycle cloud infrastructure broker has dardized way is useful in many different scenarios. been developed by the University of Manchester and Although the benefits are greatest with automation, has been adopted by CERN, who have developed an OCCI can also be used by end users if necessary. OCCI connector for their WLCG experiments. Slip- From the users’ perspective, the main advan- stream (SixSq) is the central broker of the Helix- tage of using OCCI is that a single client-side solu- Nebula infrastructure (Helix-Nebula, 2016). IM (by tion works with multiple server-side implementations. Universitat Politecnica` de Valencia)` is a tool that de- That provides for better scaling across heterogeneous ploys complex and customized virtual infrastructures resources and helps protect users’ investment into de- on IaaS Cloud deployments, automating the VMI se- veloping their client-side tools as it allows for seam- lection, deployment, configuration, software installa- less transition between providers. There are other tion, monitoring and update of Virtual Appliances. possible approaches, but the discussion of their rel- IM will be used to implement the broker features in ative merits is out of the scope of this article. A the EGI Cloud Marketplace (EGI-CM, 2016) hosted comparison is made for instance in (Parak´ and Sustr,ˇ by the AppDB (AppDB, 2016). 2014). The OCCI based approach was chosen as a VMDirac, the Catania Science Gateway Frame- key part of the EGI Cloud Federation Platform archi- work by INFN, and WS-PGRADE by SZTAKI are tecture by the EGI Federated Cloud Task as described well-known tools to develop Science Gateways in the in the EGI Federated Cloud Blueprint (EGI, 2014). grid environment. These have been extended to ex- The following sections discuss distinct OCCI us- ploit cloud resources, too. COMPSs by BSC is a pro- age patterns as seen mainly in communities gathered gramming framework, composed of a programming

349 OCCI 2016 - Special Session on Experiences with OCCI

model and an execution runtime which supports it, requirements in their communities (Mendez et al., whose main objective is to ease the development of 2014). applications for distributed environments keeping the Hereby, dirac.egi.eu engages two roles: the one- programmers unaware of the execution environment off user and the VRE/SG support. The former ac- and parallelization details. tually uses DIRAC Web portal as a basic VRE for Furthermore, the EGI Federated Cloud and its ac- job submission, retrieval and data management oper- cess model based on OCCI has been envisioned as ations, such as searching in a catalog, data transfers the ideal infrastructure to host several community or downloading results. The latter is disaggregating platforms exposing services tailored for specific user the concept of VRE/SG in two levels, the back-end groups. Integration of several platforms into the in- is the DIRAC platform for all the resource manage- frastructure is currently underway; the most relevant ment and service science. The front-end, the spe- are listed below: cific VRE/SG providing a Web environment for hous- The Geohazard and Hydrology Thematic Ex- ing, indexing, and retrieving specifics of large data • ploitation Platforms (TEPs) (ECEO, 2016) devel- sets, as well as, eventually, suppling leverage Web 2.0 oped by the European Space Agency (ESA) (ESA, technologies and social networking solutions to give 2016); researchers a collaboration environment for resource discovery. This disaggregated VRE model eases the the D4Science infrastructure (D4Science, 2016) social component of collaborations built in a walnut • that hosts more than 25 Virtual Research Environ- shell of well-known technologies, which are dealing ments to serve the biological, ecological, environ- with the increasing complexity of the interoperability mental, and statistical communities world-wide; between different resources. Therefore, in the years a uniform platform for international astronomy to follow, generic VRE from scratch will be avoided, • research collaboration developed in collaboration relaying infrastructure management in standard prac- with the Canadian Advanced Network for Astro- tices and tools such as the DIRAC platform. nomical Research (CANFAR) (CANFAR, 2016); So far, dirac.egi.eu is connecting several third selected EPOS thematic core services (TCS) party infrastructures and services in a coherent man- • (EPOS, 2016); ner, accepting logical job and data management request, which are processed with the corresponding 3.1.1 dirac.egi.eu Case Study computing and storage resources to obtain results. In this sense, a central part of DIRAC framework is a Dirac is introduced in greater depth as an example of Workload Management System (WMS), based on the a typical OCCI-enabled scientific gateway. pull job scheduling model, also known as the pilot job DIRAC platform eases scientific computing by model (Casajus et al., 2010). The pull model only overlaying distributed computing resources in a trans- submits a pre-allocation container, which performs parent manner to the end-user. DIRAC integrates the basic sanity checks of the execution environment, in- principles of grid and cloud computing (Foster et al., tegrates heterogeneous resources and pulls the job 2009) and its ecology for virtual organizations (VOs) from a central queue. Thus, proactive measures (be- (Foster and Kesselman, 1999). Depending on context, fore job matching) can be applied in case of problems, this category is commonly referred to as either Virtual and several transfers and platform issues are avoided. Research Environments (VREs) (Carusi and Reimer, The second asset is provided from the DIRAC WMS 2010) or Scientific Gateways (SGs) (Wilkins-Diehr, service side, and the late binding of jobs to resources 2007), among other terms. The dirac.egi.eu service is allows further optimization by global load balancing. a deployment of a DIRAC instance with a particular Experience using dirac.egi.eu as a back-end ser- setup for multiple communities in EGI, accessing the vice for the WeNMR VRE (Bencivenni et al., 2014) in resources on a per-VO basis, providing storage and structural biology and life science, and the VINA vir- computing allocation, dealing with grid and cloud in tual drug screening SG (Jaghoori et al., 2015), shows an interoperable manner from a single access point. an efficiency improvement up to 99%, saving on the The interfaces to connect to dirac.egi.eu are (1) usage of important resources and alleviating daily op- a Web Portal serving as a generic human interface, erations in comparison with the previous push job (2) a DIRAC client oriented to simplify bulk opera- model. tions with the command line, and (3) a python API In the particular case of cloud computing, DIRAC and a REST interface to be used by VREs/SGs in or- treats the virtual resources as standard worker nodes der to delegate resource and service management in and assigns work to them, following the same pull the DIRAC platform, thus focusing on the high level job scheduling model. For this purpose, WMS is us-

350 Easing Scientiﬁc Computing and Federated Management in the Cloud with OCCI

ing the DIRAC cloud extension named VMDIRAC implementations and at the same time it is required to (Mendez´ et al., 2013) to pre-allocate virtual machine be supported by the cloud manager and the requested resources. This VMDIRAC scheduler has finally con- image. It is a single method to dynamically contex- verged in a common cloud driver technology for most tualize VMs for different IaaS endpoints and DIRAC of the IaaS – a rOCCI client wrapped in DIRAC client install in VO basis. Python code. There are certain historical highlights worth mentioning. Initially, VMDIRAC was designed for AWS boto python library, then the pre-standard OCCI 0.8 for OpenNebula. VMDIRAC 1.0 was adopting a flex- ible architecture for federated hybrid cloud, also inte- grating OpenStack by means of libcloud. This heterogeneous cloud driver ecosystem was lacking in software convergence. Then, OCCI 1.1 became available, with rOCCI on the client side and also as an OCCI interface for OpenNebula, followed by more OCCI implementations for OpenStack and other cloud man- Figure 3: EISCAT use case in dirac.egi.eu with fed- agement frameworks. Since OCCI APIs are in Ruby cloud.egi.eu. or Java, and DIRAC is in Python, the decision was made was to adopt the rOCCI-cli client. Among its Figure 3 shows the EISCAT use case of the fed- other traits, rOCCI-cli is the most frequently used in- cloud.egi.eu stage of Figure 2. EISCAT is a work in terface, thus, the best updated. progress in dirac.egi.eu for the data processing of in- From dirac.egi.eu experience the main assets in coherent scatter radar systems to study the interac- the rOCCI adoption are: tion between the Sun and the Earth as revealed by The releasing convergence and updated features; disturbances in the ionosphere and magnetosphere. • once a new feature is out, then it is ready for all The overall architecture in Figure 3 is showing the the underlying IaaS technologies. integration of storage and computing resources, with The alignment with EGI Federated cloud comput- the explained disaggregated VRE schema. In the top • ing model, for example, the straightforward im- left, the EISCAT user front-end is a web applica- plementation of the credentials based on X.509, tion developed for the community data and metadata avoiding dealing with HTML text request to dif- management needs, using for this purpose the We- ferent native OCCI servers with their particular bAppDIRAC framework, a tornado development kit details, as always devil is in the details. completely integrated in DIRAC engine, or alterna- tively another VRE Portal accessing by APIs to the dirac.egi.eu back-end service. An EISCAT File Cat- alog contains all the metadata of the project. The data are stored in an EISCAT filesystem, accessed by a DIRAC Storage Element built on the top of such filesystem, securing connection with VO credentials. The data processing jobs are submitted from the EISCAT front-end to dirac.egi.eu back-end to pre- Figure 2: dirac.egi.eu stage with fedcloud.egi.eu. allocate VMs in EGI FedCloud, then matching jobs, downloading input data from EISCAT file server, pro- Currently, dirac.egi.eu has two main stages sup- cessing the workload and uploading output files to the ported by rOCCI client connecting EGI Federated EISCAT file server. cloud resources. Figure 2 shows a first stage with the A second stage overall architecture is in Figure 4, pull job scheduling model draw for the dirac.egi.eu connecting dirac.egi.eu with cloud resources of the service and fedcloud.egi.eu VO resources. Any training.egi.eu VO. This stage is used in EGI train- user belonging to fedcloud.egi.eu is automatically in- ing sessions, starting with a pre-configured VM de- cluded in dirac.egi.eu by polling the VOMS server, ployment, including a DIRAC client and dynamically so that they can use the service through the Web por- requesting a temporal PUSP proxy to the EGI service tal or through the other interfaces. VMDIRAC sub- (Fernandez et al., 2015). DIRAC is matching the VM mits VMs to the IaaS endpoints of the EGI FedCloud, user and the corresponding PUSP proxy with a pool with a credential of the VO proxy. VMs are contex- of generic DIRAC users. Then, following the same tualized with cloud-init, which is supported by OCCI pull job scheduling model preallocating VM train-

351 OCCI 2016 - Special Session on Experiences with OCCI

cases also exist with so little dynamism and so simple requirements that users can easily set up their virtual resources themselves, “by hand”. This is most frequently done through the command line interface (rOCCI-cli), which makes it possible to set attributes for all common OCCI actions as arguments in the command line, and send them to a selected OCCI endpoint. Though somewhat crude, this is sufficient for use cases wherein a reasonably low number of virtual machines – a few dozen at max- imum – can be started and kept track of manually. Figure 4: dirac.egi.eu stage with training.egi.eu. Such machines are then usually created at the same time at the beginning of the experiment, not necessar- ing.egi.eu resources to latter match the temporal user ily at a single site within the federation, and disposed jobs. of once the experiment ends. This is a “poor-man’s” approach to cloud resource 3.2 Processing in a Remote provisioning and management, but it is fully sufficient Data-holding Domain for user groups or single users who require a fixed set of resources for batch processing over a limited A standard-compliant management interface can be period of time – typical in the often cited “long tail exposed by a cloud site – regardless even of whether of science”. The advantage of using OCCI here is it is part of a wider cloud federation or not – to allow the same as in use cases where OCCI is used by au- users to run prepared and pre-approved workload in tomated client-side tools – users do not need to care the provider’s domain. This may come particularly about the flavor of cloud management framework they useful in view of the needs of new communities. find on the server side. The interaction is always the An example of this is bioinformatics, where the same. principal problem with specific and very strict data policies, often subject to legal restrictions, can be 3.4 Beyond the Current Scope of OCCI solved by moving the computation into the adminis- trative domain holding the data. In other words: as Although users can use OCCI to instantiate resources long as the data must not leave a certain domain, let on demand across large heterogeneous infrastruc- the computation follow that data into the domain and tures, there are additional tasks they need to complete run there. before they can use their freshly spawned resources Providing researchers with templates for virtual efficiently, and which cannot be taken care of through compute resources to process those data gives, on one OCCI at the moment. The purpose of this section is hand, the providers a chance to precisely choose what not a thorough overview, but rather a warning for po- tools will be available and how the researcher will be tential users to think about the missing pieces, and able to use them, while on the other hand it gives the also showing where future OCCI might be addressing user more flexibility and is easier to implement than some of the areas that are currently out of its scope a full-featured interface that would have to assume all First and foremost, one needs to realize that possible methods of processing the data beforehand. procuring compute resources from an IaaS cloud Offering an open standard-based interface to do leaves the user with just that – a set of virtual ma- that is only logical, and becomes necessary if such site chines. A tool to manage the workload is always wishes to be included in a heterogeneous federation. needed: One that runs outside the pool of cloud resources: 3.3 Small-scale and One-off Use of the • Cloud – either one that is aware of the nature of the resources, having them possibly instantiated it- OCCI is primarily a machine-to-machine protocol self, (such as in use cases discussed in Sec- wherein the client side is expected to (and usually tion 3.1) in which case all the components re- does) expose a user-friendly front-end that hide the in- quired are probably already in place for the user ternals of OCCI-based communication behind a por- and the infrastructure is complete. tal. Additional scheduling and workload management – or such that treats the virtual resources as stan- logic is also often hidden behind the scenes. Yet use dard worker nodes and can assign work to

352 Easing Scientiﬁc Computing and Federated Management in the Cloud with OCCI

them. This can be a local resource manage- 5 CONCLUSIONS ment system such as TORQUE, or even a simple script. It just needs to be made aware of OCCI proves to be an invaluable binding compo- the resources, and then users can submit their nent in heterogeneous cloud federations. Although it work. Here, OCCI could be of further use in may require additional effort from service providers the future, when an intended OCCI Monitoring to achieve and maintain standard compliance, it sim- extension is finished. plifies the life and work of user communities, espe- One that runs inside the pool of cloud resources: cially the small-scale ones coming from the long tail • That borders, or even falls into the scope of PaaS of science. Traditionally short on technical support (Platform as a Service) cloud provisioning model. and development staff, or lacking it completely, an As of the upcoming OCCI 1.2 specification, an interoperability standard such as OCCI helps them by OCCI PaaS extension will be available, applicable providing a single interface to interact with multiple exactly to these cases. different cloud specifications, helping them protect their investment into whatever workload management OCCI is going to gradually address these more solution they have developed, and avoid vendor lock- complex workflows, and provide extensions to cover in. With their client side being OCCI-capable, they relevant areas for larger-scale computing. At the mo- can simply move among provider sites, or scale across ment, service providers, integrators as well as users multiple providers, without having to adjust their own must keep in mind that with OCCI, they are only man- processes. aging their resources while managing their workload is up to them. ACKNOWLEDGMENTS

4 FUTURE WORK This work is co-funded by the EGI-Engage project (Horizon 2020) under Grant number 654142. There may be usage patterns as yet unexplored, but there are also several patterns that are already known and will be made possible with the OCCI 1.2 specification. Therefore much of the foreseeable future REFERENCES work, both among cloud service developers and user AppDB (2016). [Online] https://appdb.egi.eu/. Accessed: community specialists, is going to revolve around March, 2016 OCCI 1.2 adoption. Ardizzone, V., Barbera, R., Calanducci, A., Fargetta, M., One of the most visible new usage patterns will Ingra,` E., Porro, I., La Rocca, G., Monforte, S., Ric- be enabled by improved support for resource template ceri, R., Rotondo, R., Scardaci, D., and Schenone, A. prototyping. That is, it will be possible to derive tem- (2012). The decide science gateway. Journal of Grid plates from existing virtual machines, and have them Computing, 10(4):689–707. instantiated multiple times not only on the local site, Bencivenni, M., Michelotto, D., Alfieri, R., Brunetti, R., but with sufficient support also across the federation. Ceccanti, A., Cesini, D., Costantini, A., Fattibene, E., Gaido, L., Misurelli, G., Ronchieri, E., Salomoni, D., OCCI 1.2 will also introduce a JSON rendering Veronesi, P., Venturi, V., and Vistoli, M. C. (2014). specification, which will make it possible to describe Accessing grid and cloud services through a scientific resource collections and other complicated concepts web portal. Journal of Grid Computing,13(2):159– with better precision, avoiding the possible ambiguity 175 of text/occi rendering used to-date. CANFAR (2016). [Online] http://www.canfar.phys. Finally, OCCI 1.2 is first OCCI release to intro- uvic.ca/canfar/. Accessed: March, 2016 duce a PaaS specification. This will have to be care- Carusi, A. and Reimer, T. (2010). Virtual research environ- fully assessed, and candidates for adoption will have ment collaborative landscape study. to be selected from among the plethora of server-side Casajus, A., Graciani, R., Paterson, S., Tsaregorodtsev, A., and the Lhcb Dirac Team (2010). Dirac pi- products first. Then it will be possible for user com- lot framework and the dirac workload management munities to experiment with the new model of cloud system. Journal of Physics: Conference Series, service provisioning. 219(6):062049. CESNET (2016). rOCCI ec2 backend documentation. [On- line] https://wiki.egi.eu/ wiki/rOCCI:EC2 Backend. Accessed: March, 2016 D4Science (2016). [Online] https://www.d4science.org/. Accessed: March 10, 2016.

353 OCCI 2016 - Special Session on Experiences with OCCI

del Castillo, E. F., Scardaci, D., and Alvaro´ Lopez´ Garc´ıa 2014, Revised Selected Papers, Part II, chapter Per- (2015). The egi federated cloud e-infrastructure. Pro- formance Investigation and Tuning in the Interopera- cedia Computer Sceince, (68):196–205. ble Cloud4E Platform, pages 85–96. Springer Inter- ECEO (2016). [Online] http://www.ca3-uninova.org/ national Publishing, Cham. project eceo. Accessed: March, 2016 Mendez,´ V., Casaju’s, A., Graciani, R., and Fernandez,´ V. EGI (2014). Egi federated cloud blueprint. [Online] https:// (2013). How to run scientific applications with dirac documents.egi.eu/document/2091. Accessed: March, in federated hybrid clouds. ADVCOMP. 2016 Mendez, V., Casajus Ramo, A., Graciani Diaz, R., and Tsaregorodstsev, A. (2014). Powering Distributed Ap- EGI-CM (2016). [Online] https://appdb.egi.eu/browse/ plications with DIRAC Engine. In Proceedings of cloud. Accessed: March, 2016. the symposium ”International Symposium on Grids EPOS (2016). [Online] http://www.epos-eu.org/. Accessed: and Clouds (ISGC) 2014”(ISGC2014). 23-28 March, March, 2016. 2014. Academia Sinica, Taipei, Taiwan, page 42. ESA (2016). [Online] http://www.esa.int/ESA. Accessed: Metsch, T. and Edmonds, A. (2011a). Open cloud March, 2016. computing interface – http rendering. [Online] Fernandez, E., Scardaci, D., Sipos, G., Chen, Y., and Wal- http://ogf.org/documents/GFD.185.pdf. Accessed: lom, D. C. (2015). The user support programme and March 10, 2016. the training infrastructure of the egi federated cloud. Metsch, T. and Edmonds, A. (2011b). Open cloud In the 2015 International Conference on High Perfor- computing interface – infrastructure. [Online] mance Computing & Simulation (HPCS 2015). IEEE. http://ogf.org/documents/GFD.184.pdf. Accessed: Fogbow (2016). [Online] http://www.fogbowcloud.org/. March 10, 2016. Accessed: March, 2016. Nyren,´ R., Edmonds, A., Papaspyrou, A., and Metsch, T. Foster, I. and Kesselman, C., editors (1999). The Grid: (2011). OCCI specification. [Online] http://occi- Blueprint for a New Computing Infrastructure. Mor- wg.org/about/specification. Accessed: March, 2016. gan Kaufmann Publishers Inc., San Francisco, CA, Nyren,´ R., Edmonds, A., Papaspyrou, A., and Metsch, USA. T. (2011). Open cloud computing interface – core. [Online] http://ogf.org/documents/GFD.183.pdf. Ac- Foster, I. T., Zhao, Y., Raicu, I., and Lu, S. (2009). Cloud cessed: March, 2016. computing and grid computing 360-degree compared. CoRR, abs/0901.0131. occi os (2016). [Online] https://wiki.openstack.org/ wiki/Occi. Accessed: March, 2016. Garc´ıa, A. L., del Castillo, E. F., and Fernandez,´ P. O. (2016). ooi: Openstack occi interface. In Soft- OGF (2016). Open grid forum. [Online] http://www. wareX(2016). ScienceDirect. ogf.org. Accessed: March, 2016. ˇ GRNET (2016). Synnefo OCCI documentation. [Online] Parak,´ B., Sustr, Z., Feldhaus, F., Kasprzak, P., and Srba, M. https://www.synnefo.org/docs/snf-occi/latest/. Ac- (2014). The rOCCI project – providing cloud interop- cessed: March, 2016. erability with OCCI 1.1. In ISGC14, The International Symposium on Grids and Clouds 2014. PoS. Helix-Nebula (2016). [Online] http://www.helix- ˇ nebula.eu/. Accessed: Helix-Nebula Parak,´ B. and Sustr, Z. (2014). Challenges in achieving iaas cloud interoperability across multiple cloud man- IM (2016). [Online] http://www.grycap.upv.es/ agement frameworks. In Utility and Cloud Comput- im/index.php. Accessed: March, 2016. ing (UCC), 2014 IEEE/ACM 7th International Con- Jaghoori, M. M., Altena, A. J. V., Bleijlevens, B., ference on, pages 404–411. Ramezani, S., Font, J. L., and Olabarriaga, S. D. PySSF (2016). [Online] http://pyssf.sourceforge.net/. Ac- (2015). A multi-infrastructure gateway for virtual cessed: March, 2016. drug screening. Concurrency and Computation: Slipstream (2016). [Online] http://sixsq.com/products/ slip- Practice and Experience, 27(16):4478–4490. stream/. Accessed: March, 2016. Kimle, M., Parak,´ B., and Sustr,ˇ Z. (2015). jOCCI VCycle (2016). [Online] https://www.gridpp.ac.uk/vcycle/. – general-purpose OCCI client library in java. In Accessed: March, 2016. ISGC15, The International Symposium on Grids and VMDirac (2016). [Online] https://github.com/ DIRAC- Clouds 2015. PoS. Grid/VMDIRAC/wiki. Accessed: March, 2016. Lezzi, D., Lordan, F., Rafanell, R., and Badia, R. M. (2014). Wallom, D., Turilli, M., Drescher, M., Scardaci, D., and Euro-Par 2013: Parallel Processing Workshops: Big- Newhouse, S. (2015). Federating infrastructure as a DataCloud, DIHC, FedICI, HeteroPar, HiBB, LS- service cloud computing systems to create a uniform DVE, MHPC, OMHI, PADABS, PROPER, Resilience, e-infrastructure for research. In IEEE 11th Interna- ROME, and UCHPC 2013, Aachen, Germany, August tional Conference on e-Science, 2015. IEEE. 26-27, 2013. Revised Selected Papers , chapter Execu- Wilkins-Diehr, N. (2007). Special issue: Science gateways tion of Scientific Workflows on Federated Multi-cloud - common community interfaces to grid resources. Infrastructures, pages 136–145. Springer Berlin Hei- Concurrency and Computation: Practice and Expe- delberg, Berlin, Heidelberg. rience, 19(6):743–749. Limmer, S., Srba, M., and Fey, D. (2014). Euro-Par 2014: WS-PGRADE (2016). [Online] https://guse-cloud- Parallel Processing Workshops: Euro-Par 2014 Inter- gateway.sztaki.hu/. Accessed: March, 2016. national Workshops, Porto, Portugal, August 25-26,

354