Organizations Turn to New Data Center Technologies to Drive Innovation
- 01.24.14
- 1:05 PM
Purdue’s data center-in-a-box (shipping container.) Image: Purdue University
In today’s computing business environment, everyone is looking for more: more computing capacity, more storage, more data transfers, and more bandwidth. To remain static is to fall behind; therefore, it is crucial that organizations take advantage of the latest advances in technology to stay competitive.
Interconnect technology is a critical element responsible for the overall system capability and performance. Fast Interconnect technology powers the world’s fastest supercomputers and most successful organizations, as it allows lightning-fast transfers of massive amounts of data. Without high bandwidth interconnects that allow data to move quickly and reliably from point-to-point, we wouldn’t have the level of supercomputing and social media technologies we have today.
The most prominent development in this respect is Remote Direct Memory Access (RDMA). Whether used across InfiniBand or Ethernet protocols, RDMA is the fastest available interconnect technology on the market. RDMA enables high-throughput, low-latency networking without involving the CPU by allowing the network adapter to independently transfer data directly from one machine’s memory to another.
By incorporating RDMA into networking solutions, more is now easily achievable within a single data center. However, while data centers have traditionally been located locally because of power and space limitations, the data explosion has forced companies to build multiple data centers, sometimes miles away from the primary site. How can an organization achieve more with its interconnect when its data centers are spread across long distances?
Long-haul interconnect is a new innovation that makes it possible for organizations of all sizes — from corporate giants and large universities to small technology providers — to move from the paradigm of multiple, disconnected data centers to a single multi-point meshed mega-datacenter. Long-haul connectivity also addresses key concerns for these organizations: faster, more reliable disaster recovery; improved utilization of remote storage; and combined compute infrastructures across geographic sites. Connecting remote data center sites with long-haul interconnect technology provides fast compute, high volume data transfer, and improved business continuity at long distances.
It also allows organizations to take advantage of the space they already have, instead of building brand new data center facilities.
Take Purdue University, for example. Purdue deployed a Mellanox long-haul interconnect solution over six kilometers, to connect a remotely-sited compute cluster to central networks and storage facilities. This access allowed the university to take advantage of existing facilities to house a state-of-the-art supercomputer, without increased construction or retrofit costs. The university’s savings also came without sacrificing performance. Purdue researchers are able to take advantage of the fastest interconnect technology at a distance, and can therefore run complex simulations and further advance their cutting-edge research in areas such as climate change, aerospace, and molecular biology.
For research organizations like Purdue, real-time processing is vital to their ability to process the massive data inputs that are required to run complex simulations and perform advanced modeling. Furthermore, these organizations cannot combine the computing power of thousands of CPUs unless the data transfers between them are instantaneous and reliable. Long-haul interconnect enables high speed computing as if centralized storage were local, enabling real-time data processing across large distances without the associated performance penalties.
“A common problem facing data-driven researchers is the time cost of moving their data between systems, from machines in one facility to the next, which can slow their computations and delay their results,” said Mike Shuey, HPC systems manager at Purdue. “Mellanox’s long-haul solution lets us unify systems across campus, and maintain the high-speed access our researchers need for intricate simulations – regardless of the physical location of their work.”
Moreover, it is even possible to boost supercomputing performance by using long-haul connectivity. Mellanox recently implemented a solution for a large federal research client that had hundreds of thousands of CPUs spread over multiple facilities separated by a number of kilometers. Using long-haul interconnect to combine the facilities into a mega-datacenter, the client saw huge improvements in both bandwidth and latency, while also realizing significant savings in capital expenditures by reducing the amount of hardware required to maintain the connection.
One primary concern for universities like Purdue and other organizations dealing with massive quantities of data is that of disaster recovery. In a disaster situation, data centers can become inaccessible, paralyzing an organization’s capabilities until connectivity to and from the primary site can be restored. Imagine the chaos that would ensue if social media and internet search sites were unavailable or exceptionally slow whenever an especially large storm hit their headquarters. Long-haul interconnect solutions ensure the lowest possible amount of data loss in such a situation.
Often, large organizations designate a secondary site that functions when the primary site is down; however, if the secondary site is many kilometers away, there is the potential for severe performance degradation and the data loss that results from such reduced performance. Long-haul interconnect ensures that there is no drop-off in bandwidth even when the backup site is located far from the primary site.
Long-haul interconnect allows companies to control and manage sites that are tens of kilometers apart, building a more efficient and more powerful network. By connecting remote facilities and transforming individual data centers into a mega-datacenter, long-haul interconnect enables faster computing and yields higher volume data transfer with no performance penalty. It also improves business continuity by facilitating faster backup and more reliable disaster recovery.
Most of all, long-haul interconnect keeps today’s top companies and research facilities at the cutting edge of innovation by providing more data transfers at higher speeds, leading to more innovations such as product development and cures for diseases at the molecular level. Thanks to long-haul interconnect, it really is possible to achieve more.
Tal Roll is a Product Manager at Mellanox Technologies.
Images courtesy of Purdue University.
