The reduction of high operational costs – in addition to the scaling back of the vast complexity of data center infrastructure – is a chief concern among corporations that utilize supercomputer technology. One of the most invaluable methods of accomplishing this complicated task is made possible through the use of the InfiniBand protocol, a communications link that allows interconnectivity between numerous individual computers to create an aggregate supercomputer environment. Lately, the discussion around InfiniBand has turned to its possible use in facilitating the rapid transfer of enterprise data from servers to storage systems – and it should only be a matter of time before the use of InfiniBand as a data transfer protocol becomes standard practice.
The Technology of InfiniBand
InfiniBand switch systems are capable of facilitating an incredible amount of bandwidth over a single fabric, with speeds that currently clock in at 10Gb/s using SDR (Single Data Rate), 20Gb/s using DDR (Double Data Rate), and 40Gb/s using QDR (Quad Data Rate). For this reason, it’s become an indispensible technology in HPCC (High Performance Computer Cluster) environments.
InfiniBand is capable of achieving high communication speeds through its utilization of RDMA (Remote Direct Memory Accessing) technology. That technology makes it possible to send messages from one system to another without the need to filter data through the buffers in the operating system. This capability has the impact of greatly decreasing latency times and increasing overall throughput. The by-product of this naturally results in decreased usage of CPUs when transmitting signals.
The transmission of information between a host processor and target peripheral is accomplished through the installation of Host Channel Adapters (HCA) and Target Channel Adapters (TCA). InfiniBand adapters can be installed into the PCI Express slots of most systems to enable the full operation of InfiniBand technology.
The Use of InfiniBand in Supercomputing
The role that InfiniBand technology has played and will continue to play in supercomputing isn’t something to be taken lightly. The implications of this technology when applied to HPCC environments in scientific and medical research are considerable, if not downright critical. By enabling the hardware to communicate at such high speeds, the following aims can easily be realized:
- Greatly increased computation speeds, which have and will continue to have a great impact on the ability of researchers to make important new discoveries in the fields of science and medicine.
- Quickened communications for greatly improved systems performance. The use of IB switches to speed up communications in a cluster environment is capable of outperforming Gigabit Ethernet speeds by as much as 60 percent.
- Dramatically increased speeds in the performance of simulations conducted by HPCC systems, where discovery that may have previously taken months can be completed in only minutes.
The use of InfiniBand technology to squeeze the maximum speed out of computer clusters isn’t only limited to scientific research. The practicality of the technology allows for its use in a variety of different industries, and often results in the accomplishment of several universal goals:
- Power consumption reduction. This all-important objective is enabled by the fact that InfiniBand operates on far less power than Ethernet technologies. An InfiniBand adapter, for example, uses close to 4 watts – whereas competitive technologies can draw upwards of 5 to 6 times that amount. Add to that the fact that when CPU power usage drops, the generated heat levels follow, meaning that companies will also spend far less money on cooling and ventilation.
- Management overhead can also be significantly impacted by InfiniBand’s capacity to support the operation of numerous workloads and applications.
The Future Use of InfiniBand for Storage Solutions
At the rate that corporation data centers are growing in size, there will never come a time when there won’t be a need to constantly seek out new and innovative methods of further diminishing latency times and overall speed of information transfer. To this end, storage systems manufacturers are taking a closer look at InfiniBand technology as a means of “souping up” the speeds at which critical enterprise data is transferred to storage.
The attractiveness of InfiniBand with regard to a range of issues including the low consumption of power, increased speed, and reduced latency make it tailor made for future use as the primary component of communication between data center servers and storage systems. This appeal makes it perfectly suited to become standard for use with Storage Area Networks (SANs) – and what’s even more encouraging is the fact that its implementation can be made seamless through the use of gateways that will ensure its operability with NAS, Fibre Channel, and iSCSI.
Corporate data centers typically approach the adoption of new technologies with a cautious eye, and rightly so. The last thing that a corporation wants to do is to spend tens of thousands of dollars to upgrade their IT technology, only to discover that its limitations make it ill-suited for use across newly emerging procedures. The fact that InfiniBand has been identified as an accepted technology by leading solutions providers and that its use has become widespread is a positive sign that any capital investments made today will not be regretted tomorrow.
|