While the natural habitat of HPC storage is supercomputing and enterprise datacenters, the growing popularity of edge computing means sometimes that hardware must live outside its comfort zone. One such example is using these storage systems on ships performing undersea oil and gas exploration, a use case that has expanded the horizons for at least one well-known HPC vendor.
Performing seismic imaging of subsurface geologic structures is the conventional way energy companies look for oil and gas underneath the sea these days. Sonar receivers deployed on the seabed are used to collect seismic data of the underlying geological structures, which indicate the presence or absence of hydrocarbon reservoirs. What’s changed in the past few years is that with the advent of higher resolution 3D and 4D seismic imaging and the use of larger survey areas, data volumes have grown tremendously. And that means storing that data has become increasingly difficult.
And time-consuming. In the good old days, the collected data was physically brought back to shore and uploaded into storage residing in a traditional datacenter, where the seismic processing and reservoir modeling would take place. But the time involved in taking the data acquisition hardware offline and bringing the ships back to land is costly. How costly? Running these vessels with a full crew can cost over $200,000 a day.
That is plenty of motivation for energy exploration firms to streamline their data collection workflow, and one way to do that is to place fast, high-capacity storage on-board. By doing so, ships can stay at sea long enough to perform an entire survey before coming back to land. With the larger geographical areas now being covered, such surveys can now take up to six months and collect hundreds of terabytes.
That scenario describes what is being done at Magseis Fairfield, a Norwegian seismic data acquisition company that serves exploration and production firms. In this case, the company is using Panasas storage, both in its datacenters and, more recently, on board its data collection vessels. For the company’s off-shore data project, each vessel is equipped with between 250 and 600 terabytes of ActiveStor gear, for a total capacity of 1.6 petabytes spread across its fleet.
According to Dale Brantly, Panasas’s director of Storage Systems Engineering, obviously this is not a typical use case for ActiveStor, since here, the system is being employed solely as an I/O engine for capturing raw seismic data. There’s no compute cluster attached to it performing in situ analysis. That’s still being performed at a traditional datacenter once the files are uploaded – some of which is done over a satellite link while still at sea.
The whole idea, says Brantley, is to keep the ships at sea and those seismic collectors deployed on the seabed as long as possible. And with data volumes on the rise, both capacity and I/O speed are critical factors for Magseis Fairfield. Since ActiveStor and its PanFS parallel file system were designed with those two aspects in mind, they would appear to be well-suited to provide an I/O engine in this kind of environment.
“The resolution associated with this data collection continues to climb,” he explains.” If they increase the data collect capacity of their acquisition devices 10 times, now their storage requirement has grown 10 times.”
Magseis Fairfield has been a Panasas customer for 15 years but has only been employing on-board storage for about half that time. Besides the obvious compatibility advantage of having ActiveStor on both the ship and the datacenter, according to Brantly, the company has found its appliance-like design a good fit for the on-board computer room and personnel.
Surprisingly, the storage gear are not ruggedized or otherwise hardened for sea duty. This is the same ActiveStor hardware – in this case, ActiveStor 20 systems – as you would find in an HPC facility. Whatever is required to protect the hardware from the marine environment is provided by the ships’ own machine rooms.
And since there are no system administrators taking sea cruises to baby-sit the hardware, all storage maintenance is in the hands of the on-board data scientists, who are equipped with only an ActiveStor instruction booklet. Usually maintenance amounts to replacing a failed storage blade or a dead battery, says Brantly, and that’s about as complicated as it gets. “There are only seven replaceable parts in the whole box,” he notes.
Panasas Sales Director Michael Tortorello tells us they have a few other customers needing this sort of edgey storage solution, including Petroleum Geo-Services (PGS) and TGS. Like Magseis Fairfield, both provide seismic data services for the energy industry. Tortorello says they’re also engaged with another energy company collecting well data via drones using RFID technology. The company was using the cloud to upload its data but was limited to a 200Mbps data pipe. With Panasas, they are able to exceed their target of 2Gbps, an order of magnitude more. The storage-maker also has edge deployments in the classified DoD realm, but for obvious reasons is rather tight-lipped about the details.
Will Panasas start building more customized edge storage solutions? At this point, there are no plans to do so, but as Tortorello points out, since PanFS is now portable, it would relatively straightforward to migrate the file system onto other, more customized hardware platforms. And since most of Panasas’s business is now on the commercial side of HPC, it might make sense to pursue a strategy that taps into this still-nascent edge storage market. Time will tell.