CENTRO DE SUPERCOMPUTACIÓN DE GALICIA

1. CENTRO DE SUPERCOMPUTACIÓN DE GALICIA OpenNebula: Experience at Cesga Iván Díaz, Esteban Freire, Alejandro Feijoo, Álvaro Simón, Javier López Cacheiro, Pablo Rey and Carlos Fernández EGI Community Forum 2014: Helsinki, Finland, 22/05/2014

2. Outline • Introduction • CESGA Cloud Infrastructure • EGI FedCloud Infrastructure • Hadoop • CloudPYME • Accounting system • Perks and Quirks

3. Introduction • CESGA site must adapt and react to very complicated requirements. • Necessary to economize on energy consumption. • Our collaboration on different projects (EGI FedCloud, CloudPyme, Hadoop...) requires new testing and verification resources, which must be assigned flexibly depending on demand. • The solution is to use a hybrid cloud based infrastructure. • The infrastructure is hybrid because the VMs used differ in image content, purpose, users and life cycle.

4. 40 HP ProLiant SL2x170z G6 (Virtualization Nodes)‏ Processor 2 x INTEL Processor E5520 (2.26 GHz, 8MB Cache, 4 cores per processor)‏ Memory 16 GB Storage 211 GFlops per system 500GB (35 nodes) / 1000GB (5 nodes)‏ Networking Embedded 2 Ports Intel 1Gb 6 HP ProLiant DL180 G6 (Server & Storage)‏ Processor 2 x INTEL Processor E5520 (2.26 GHz, 8MB Cache, 4 cores per processor)‏ Memory 16 GB Storage 211 GFlops per system 5700GB (5 nodes) / 2000GB (1 nodes)‏ Networking Embedded 2 Ports Intel 1Gb (6 nodes) / 2 Ports NetXen 10G (4 nodes)‏ HP ProLiant SL2x160 G6 (Server)‏ Processor 2 x INTEL Processor E5504 (2 GHz, 4MB Cache, 4 cores per processor)‏ Memory 32 GB Storage 211 GFlops per system 1200 GB Networking Embedded 2 Ports Intel 1Gb HP ProLiant SL2x165 G6 (Server)‏ Processor 2 x AMD Processor 2435 (2.6 GHz, 6 cores per processor)‏ Memory 32 GB Storage 211 GFlops per system 1200 GB Networking Embedded 2 Ports Intel 1Gb Cloud Infrastructure

5. Cloud Infrastructure

6. EGI FedCloud Infrastructure • Cesga is collaborating as resource provider and developer within EGI FedCloud Task Force. • EGI FedCloud auth its based on x509 authentication and OGF OCCI 1.1 (rOCCI server) • ON OCCI service is not OGF compliant... • Virtual Organizations and users management is based on PERUN service developed by CESNET.

7. EGI FedCloud Infrastructure • EGI Fedcloud OpenNebula accounting plugin is still under development • ON inserts new records into MySQL database in XML format. The accounting plugin must parse this info. • ONEACCT is a good tool but does not display all the information required by Fedcloud accounting infrastructure (User DN, Image UUID...)

8. Hadoop • CESGA is providing a testing Hadoop infrastructure for local and Fedcloud users thanks to ON framework. • Hadoop is an emerging technology that could benefit from existing developments in cloud. • It represent a real challenge to the Fedcloud infrastructure due to its heavy usage of disk and network I/O. • Runing Big Data calculations over federated clouds is attracting the interest of the research community.

9. Hadoop • Three steps to deploy a Hadoop cluster inside the Cloud infrastructure: • Cluster startup script: obtaining the necessary resources. • Hadoop configuration and contextualisation: configuring the Hadoop cluster. • Benchmark execution from Hadoop headnode: Load data and run MapReduce job.

10. Hadoop • This service was already tested using Fedcloud infrastructure:

11. Hadoop • Pros: • Initial benchmarks performed by CESGA suggest the suitability of a private/federated infrastructure to run certain type of Big Data analytics. • Small VM instances are good enough to run Hadoop. • About the future work is possible to offer Hadoop clusters on-demand through a simple web frontend (Sunstone cloud view?)

12. Hadoop • Cons: • There were some disappointing results about cluster startup (Ceph tm is recommended instead of SSH. SSH tm driver is very slow to deploy simultaneously a large number of machines). • Only a few resource providers are able to provide the required resources. • In general cloud Hadoop clusters are not suitable for long term big data experiments (I/O performance and HDFS data distribution time)

13. CloudPYME • The project aims to create an infrastructure to support innovation in small and medium enterprises (SME). • By providing SME access to CESGA HPC and Cloud facilities. • By encouraging the implementation of Product Lifecycle Management (PLM) tools and techniques, seeking to improve efficiency in new product developments. • It provides a complete PaaS solution which includes computing resources, storage and billing.

14. CloudPYME

15. CloudPYME

16. Accounting system • This scenario tries to integrate accounting records from resource providers, in order to aggregate user resource consumption across all the federation. • This is an area where interoperability has been not reached. No standards, official or otherwise. • CESGA uses a custom internal cloud accounting service (Cloud-PYME project). • A script was developed to transform this data and transfer it via SSM to the APEL repository site for EGI Fedcloud project. • SSM transfers information between computers using python, ActiveMQ and STOMP.

17. OpenNebula 4.6 MySQL database Accounting system EGI Apel Global Repo cloud.cesga.es ac-consumo.cesga.es Summarized records Running /stopped/finished VMs

18. Perks and Quirks • Perks: • OpenNebula infrastructure is easy to install and maintain. • ONE endpoint could be used by different services and OCCI servers (cloud interoperability between projects) • ONE service upgrades are independent of compute nodes or DOM0s upgrades. • Easy LDAP integration. • Since v4.6 it includes new federation features.

19. Perks and Quirks • Quirks: • ONE user clients require .one/one_auth “user:password” file even for LDAP auth. • ONE OCCI is not OGF OCCI compliant. • ONE database fields are messy (XML internal format). • Accounting API is a bit slow. • ONE VM image names do not support common characters like “:” or “/”.

20. Questions? Thank you for your attention! Questions?