Magerit
| Company type | Supercomputer |
|---|---|
| Industry | High-performance computing |
| Founded | 2005 |
| Headquarters | Madrid , Spain |
| Parent | Technical University of Madrid (CeSViMa) |
Magerit ranks among the most advanced supercomputers in Spain, situated at CeSViMa, a research facility affiliated with the Technical University of Madrid. It holds the distinction of being the second highest-ranked Spanish system in the TOP500 supercomputer listings.
Initially deployed in 2006, Magerit secured the position of 9th fastest across Europe and 34th globally, representing Spain’s second top placement in the TOP500 rankings. Additionally, it appeared at 275th place on the first-ever Green500 list. Following a significant upgrade in 2011, Magerit rose to become Spain’s foremost supercomputer, ranking 44th in Europe and 136th worldwide, while climbing to 18th on the Green500 for energy-efficient performance.
The term Magerit originates from Madrid’s old Arabic designation, meaning "place of abundant water," symbolizing its historical significance.
Initial Phase (2005)
Magerit was established through collaboration between the Technical University of Madrid and IBM, initially located within the Computer Science School at Madrid’s newly founded CeSViMa. The original system comprised 124 nodes, funded by the Spanish Ministry of Education and Science alongside the Autonomous Region of Madrid.
Integration into the Spanish Supercomputing Network (2006–2007)
By late 2006, CeSViMa became part of the Spanish Supercomputing Network (RES), expanding Magerit to 1204 nodes and reaching a performance of 14 TFLOPS. This upgrade secured a 34th position on the TOP500 list, a notable achievement for Spanish supercomputing resources. Resource allocation was divided with 68% dedicated to RES users and 32% reserved for CeSViMa local users.
Relocation and Incremental Enhancements (2008–2010)
In May 2008, both Magerit and CeSViMa relocated to a new campus facility approximately 500 meters from their former location. Upgrades included new communication switches, storage systems, and replacement blades, increasing computing power by nearly 2 TFLOPS to a total of 15.95 TFLOPS. Despite these improvements, Magerit was no longer listed in the TOP500 by November 2008.
During this period, resource usage was split with 59.7% allocated through RES and 40.3% managed locally by CeSViMa. In 2009, the system software transitioned to SUSE Linux Enterprise Server version 10. By 2010, CeSViMa enhanced storage capacity by adding a new system with one petabyte, complementing Magerit’s existing infrastructure.
Major Upgrade (2011)
At the beginning of 2011, Magerit underwent a comprehensive overhaul that replaced all computing nodes and networking components using state-of-the-art technology within a record timeframe of one month. This revamped configuration positioned Magerit at 136th on the TOP500 list and elevated it to 18th place on the Green500 ranking, establishing it as Spain’s most powerful and energy-efficient supercomputer.
The distribution of usage shifted so that CeSViMa-UPM managed 80% of resources while RES accounted for 20%. Although this represented a smaller proportion for RES users, their allocated resources increased four- to fivefold. The storage subsystem installed in 2008 remained intact with plans for minor upgrades to meet future requirements.
Architecture
Magerit exists in two principal configurations:
- The initial setup from 2006 (which incorporated the original 124 nodes from 2005) with minor enhancements made in 2008.
- The comprehensive overhaul completed in 2011 that positioned Magerit as Spain’s leading supercomputing system.
Initial Configuration (2005–2010)
The original setup achieved 34th place on the TOP500 list in November 2006, ranking second nationally within Spain and ninth across Europe. Additionally, it was positioned 275th on the inaugural Green500 ranking.
Following the 2008 enhancement, the system comprised a cluster of 1204 IBM eServer BladeCenter units, including 1036 JS20 and 168 JS21 nodes, all operating on SUSE Linux Enterprise Server 9:
- JS20 nodes contained two IBM PowerPC single-core 970FX processors (totaling two cores) clocked at 2.2 GHz, equipped with 4 GB of RAM and 40 GB of local disk space.
- JS21 nodes featured two IBM PowerPC dual-core 970FX processors (four cores in total) running at 2.2 GHz, paired with 8 GB RAM and 80 GB of local storage.
Storage was provided by a distributed GPFS system offering 190 TB capacity, linked through a high-speed switch capable of peak data transfer rates up to 1 Tbit/s. The nodes communicated via a low-latency Myrinet network (with latency between 2.6 and 3.2 microseconds) dedicated to MPI traffic, supplemented by an auxiliary Ethernet network for administrative purposes.
Updated Version (2011)
The 2011 upgrade elevated Magerit to Spain’s most powerful supercomputer, securing the 44th position in Europe and ranking 136th globally.
The cluster architecture was revamped to include 245 PS702 nodes, each outfitted with dual 64-bit POWER7 processors providing a total of sixteen cores per node at a frequency of 3.0 GHz. Each node had 32 GB RAM and 300 GB of local storage. Individual cores delivered performance of 18.38 Gflops.
Interconnect technology advanced to a 40 Gbit/s Infiniband fabric featuring ultra-low latency of approximately 0.3 microseconds. Two separate Gigabit Ethernet networks were deployed to handle auxiliary operations such as image deployment and storage access.
The GPFS-based storage capacity remained around 192 TB with bandwidth close to 1 Tbit/s.
Software improvements included upgrading to SLES11SP1 operating system, implementation of the xCAT deployment framework, and refreshed sets of libraries and development tools.
Latest Version (2019)
In its most recent iteration, Magerit incorporated Lenovo ThinkSystem SD530 nodes.
Applications
Magerit is primarily utilized for executing batch processing tasks that demand significant computational resources, including simulations in universe modeling, material science, and climate forecasting. A key project supported is the Cajal Blue Brain initiative, representing Spain’s participation in the broader Blue Brain Project.
Computations typically run over multiple days across hundreds of CPUs, requiring continuous operation throughout the year. A batch job queue management system ensures efficient scheduling of resources globally while maintaining fair access among users.
Resource Access
Access to Magerit’s computing capabilities is available to individuals, research centers, or enterprises through several channels:
- Submitting direct applications via CeSViMa’s official website.
- Establishing collaboration agreements with CeSViMa.
- Participating in competitive resource allocation managed by the Spanish Supercomputing Network, which controls approximately one-fifth of Magerit’s capacity through RES.
External links
- CeSViMa – Centro de Supercomputación y Visualización de Madrid, host institution for Magerit
- Detailed hardware and software specifications
- Time-lapse video documenting construction of Magerit's second generation preserved at Ghostarchive.org as of April 17, 2022