Get This Page | NCBS

Computing Clusters - Computing & Storage

1. Cray CS300-AC:

`Tiramisu` as it is normally called, uses the CentOS 6.4 operating system and vSMP from the SCALEMP foundation. It is comprises of a single racks with 2u twin architecture and has 16 physical chassis in total. Each node is driven by 2 nos. of 8 core AMD processors. Tiramisu uses InfiniBand (IB) switching at a rate for 36,288 Gbps. Tiramisu CPUs total 64 amouting to 1024 cores, and there is a total of 16 chassis, 8 TB of RAM, and around 30 TB of storage. System provides a single system image using vSMP foundation.

As of December 2015, Tiramisu has used more than 6 million CPU hours.

`Tiramisu` scheduler is Torque. The cluster is mostly being used to run application which requires huge datasets to be put in memory. The details on how to run jobs in the cluster is being provided in and is being uploaded at `/opt/HOW-TO-SUBMIT-JOBS` and also provided at the time of login.

A networked, high-performance 30 storage system is connected via IB fabric. The IB is FDR with a 100 percent non blocking architecture.

2. Intel Resilienté:

`Nargis` as it is normally called, uses the CentOS 6.4 based ROCKS distributed operating system. It is comprises of a 2 racks one housing the nodes while the other housed the IB switching equipment, networking and the storage. It also uses with 2U node twin architecture and has 16 physical chassis in total. Each node is driven by 2 nos. of 8 core Intel processors. Nargis uses InfiniBand (IB) switching at a rate for 36,288 Gbps. Nargis CPUs total 64 amounting to 1024 cores, and there is a total of 16 chassis, 8 TB of RAM, and around 30 TB of storage.

As of December 2015, Nargis has used more than 12 million CPU hours.

`Tiramisu` scheduler is SGE (Grid Engine). The cluster is mostly being used to run application which are CPU intensive and can run parallely. The details on how to run jobs in the cluster is being provided in and is being uploaded at `/opt/HOW-TO-SUBMIT-JOBS` and also provided at the time of login.

A networked, high-performance 30 storage system is connected via IB fabric. The IB is FDR with a 100 percent non blocking architecture. The centralized storage with GPFS is also mounted in all the  nodes through IB for additional data/compute store.

3. DDN SFA 12K-20E:

DDN’s Storage Fusion Architecture (SFA) utilizes the most advanced processor technology, busses and memory with an optimized RAID engine and sophisticated data management algorithms. The SFA12K-20E product family is designed to derive peak performance out of the compute investments with a massive I/O infrastructure and multi-media disk drives that maximize system performance. The SFA12K-20E simplifies big data growth with future scalability in a more intelligently, efficiently and cost effective manner.

The storage comprises of 750 TeraBytes of which 50 TB is on SAS while the rest 700 TB on SATA. It uses GPFS to provide parallel file system. It is also connected to a tape archival system and a WOS for disk to disk backup.

With 40Gb Ethernet and InfiniBand connectivity, the SFA12K-20E delivers 20GB⁄s of bandwidth and 700K Flash IOPS. It also eliminates data access latency and the cost and complexity of external computing and storage networking for various HPC applications.

WOS(Web Object Scaler) leverage's a unique ObjectAssure data protection, which is optimized with replication to provide self-healing capabilities.