CINECA powers up high-performance virtualized storage with IBM

Italiano

CINECA is the largest high-performance computing (HPC) centre in Italy, and one of the most important worldwide. Run by a consortium of 25 leading Italian universities, CINECA employs more than 250 people and is located on the western outskirts of Bologna. The organisation provides HPC services to major public and private research institutions, and to commercial enterprises, helping studies in fields such as physics, chemistry, bioengineering, medicine and astronomy.

With dozens of major HPC projects running on several supercomputing clusters, CINECA has demanding requirements for data storage, in terms of speed, capacity and flexibility. The challenge is to meet rapidly changing storage needs without incurring high costs, and to support multiple client platforms while keeping administration as simple and rapid as possible.

Tommaso Cecchi, Storage Administrator at CINECA, comments, “We deal with enormous volumes of scientific data for numerous completely separate computing environments. In the past, this meant that we spent a lot of time laboriously transferring data from one machine to another, which was costly and caused delays. Our aim was to create a single environment for data storage, capable of serving a mixed set of client platforms, and offering excellent performance.”


Acceleration through virtualization
CINECA chose to work with IBM on the creation of an ambitious new storage infrastructure, built around the concept of virtualization. The goal was to create a single pool of virtual storage that could be shared flexibly between multiple systems, with policy-driven automation to keep management costs low.

IBM proposed a storage area network (SAN) based around six IBM TotalStorage DS4500 disk arrays, three of which use ultra-high-performance Fibre Channel disks, and three of which use the low-cost SATA standard for applications with less intensive input/output requirements. The storage capacity on the DS4500 systems is managed by IBM TotalStorage SAN Volume Controller (SVC) software, which virtualizes the available capacity, making it easier for CINECA to allocate new storage to applications.

IBM TotalStorage SAN File System (SAN FS) completes the core of the solution, enabling multiple servers to share data stored on the SAN, and uses policy-based automation to move data between physical devices without disruption to users or applications.

“From the user’s point of view, the IBM virtualized storage solution represents a significant improvement,” says Cecchi. “There is now a single point of access to all data, regardless of which operating system owns that data, so it is quicker and easier for users to run multi-stage simulations.

“You can think of this work in terms of a supply chain made of different blocks of data on different machines: one creates an output, and the next machine uses that as its input. Previously, that was a manual process, slow and prone to error. With the new IBM solution, the data can be shared more easily, helping researchers to complete projects more quickly.”


Powerful cost savings
The new virtualized environment provides storage services for several HPC platforms, including two IBM supercomputing clusters. The newest is an IBM Cluster 1600 with 64 nodes, each of which is an 8-way IBM System p5 575 server, for a total of 512 POWER5 processors. CINECA also has an IBM Cluster 1350 composed of 512 IBM eServer xSeries 335 servers, each of which has two Intel Xeon processors.

A key requirement for the new storage solution was the ability to support multiple client platforms, including various UNIX and Linux distributions. The use of SAN FS provides a single file system for all of these different platforms, enabling a standardized solution for easy management and efficient operations.

“With SVC and SAN FS, we have created a storage environment that is much easier to administer, saving us time and money,” comments Cecchi. “We can effectively see the status of all our storage at a glance, and we can increase the virtual capacity available to any system from a single point of control. The IBM solution also delivers cost benefits through physical consolidation: by enabling us to have a single pool of storage shared between multiple systems, it has improved our utilization of disk resources, so we can avoid buying new disks until the existing ones are practically full.

“We estimate that the IBM solution has significantly reduced our overall costs for data storage management, while improving the availability, performance and flexibility.”


Anticipating growth
The IBM storage solution at CINECA currently manages around 30TB of research data, and the organization expects rapid growth, to around 200TB within the first eighteen months of operation. With a fully virtualized architecture, the organization can add new physical disks or even entire new disk arrays without disruption to applications or users, and without significantly increasing the future administrative workload.

Cecchi concludes, “The virtualized storage solution provided by IBM will help us to serve the growing needs of our clients without a corresponding increase in cost in administration. As simulations become more complex and require more data, this scalability is an important benefit.”