Space attracts not only scientists, but also companies working in the field of information technology. The very idea of placing computing power in outer space has a number of unique advantages. One of the pioneers in this area can rightfully be considered HPE, which in 2017 sent the HPE Spaceborne system to the ISS. HPE today announces the transition from a test phase to a commercial phase and unveils the second generation of the platform, Spaceborne-2.
As a rule, space electronics has a number of specific features associated with the operating conditions. First, it must be resistant to high levels of radiation, and secondly, consume as little energy as possible. Such electronics are very expensive, and the computing power does not shine, so the decision of HPE to adapt classical computing platforms to the conditions prevailing on the ISS looks quite logical. In addition, scientific experiments are constantly being carried out on the ISS, which would not be in the way of extra teraflops.
The first variant of Spaceborne was already superior in performance to anything that had ever been in orbit. This space supercomputer consisted of two HPE Apollo 40 servers connected by a 56 Gbps InfiniBand network, but more importantly, each compute node contained four NVIDIA Tesla P100 accelerators. This made it possible to bring the power of Spaceborne to 1 teraflops – modest by ground standards, but a record for space.
The first generation of HPE Spaceborne was not used for scientific purposes or to control the subsystems of the ISS, its task was to prove the possibility of the normal functioning of server equipment in a space station. The experiment ended successfully, and today HPE is ready to present a new generation of “space supercomputers” – Spaceborne-2.
It will be based on the HPE Edgeline EL4000 Converged Edge Computing Platform, and the latest generation HPE ProLiant DL360 servers with two Intel Xeon Cascade Lake processors and NVIDIA T4 accelerators will be used as the main compute nodes. It is expected that the performance of Spaceborne-2 will at least double its predecessor, but more importantly, this is no longer an experiment and the new space supercomputer will bring real benefits.
There will be two racks with EL4000 and DL360, and both will monitor the condition of both themselves and the neighbor. All data is duplicated between racks, and SSDs will be combined in software and hardware into RAID arrays. Despite the fact that SSDs are less resistant to space radiation (in Spaceborne-1, 9 out of 20 disks died during operation), they have an advantage in operating speed. In addition, some spare server components will be available at the station itself so that they can be quickly replaced.
The servers will use a 10GbE network to communicate with each other. The systems will be powered from two independent lines, which are connected to solar panels and batteries. There is a stepwise dynamic regulation of the energy consumption level. Cooling in the new systems is hybrid – the heat exchanger in the rack is directly connected to the ISS water cooling circuit.
The ISS is a middle-aged station, the first modules were launched into orbit back in 1998, but all this time it was constantly being modernized taking into account new scientific needs. During this time, as we know, computing has managed to make a huge leap forward, but until now, its capabilities were inaccessible to scientists working on the ISS and in need of more and more serious computing power. The station was overgrown with numerous instruments and sensors, the amount of data received increased, but even for primary processing they still had to be transmitted to Earth, which negatively affected the time of research.
The appearance of Spaceborne-2 will solve this problem: with a capacity of more than 2 teraflops, it will be possible to carry out at least primary data processing in the shortest possible time, and in some cases, this capacity will be enough for research without using ground computing power. HPE’s named use cases include tracking terrestrial traffic from space to identify trends, assessing pollution of the Earth’s atmosphere, and tracking air and space traffic, also in real time.
Spaceborne-2 will also take an important part in ensuring the health of astronauts working at the station: the capacity of the new platform will be enough to constantly monitor their health indicators, including complex ones – X-rays and sonograms. This will make it possible to make a diagnosis in the shortest possible time and, thereby, minimize the damage from a possible illness, and, perhaps, save the life of a sick astronaut.
The new “space supercomputer” HPE will be able to work not only alone, but also as a peripheral platform – Microsoft Azure Space will provide two-way communication between Spaceborne-2 and ground computing complexes, which will allow scientists to use it around the world. The studies named by Microsoft Research include modeling and predicting dust storms on Earth, which will help to better understand similar processes on Mars, estimating water costs for growing plants in zero gravity, and studying thunderstorm patterns that lead to large-scale forest fires.
But the ISS still has one mankind, and the needs for space computing power clearly exceed its cargo and passenger capabilities. HPE is developing the SatFrame platform, which can be used on uninhabited satellites – the launch of such a satellite will be significantly cheaper than a manned flight to the ISS, which means that more companies and scientific organizations will be able to afford a space-based peripheral micro-data center.
Currently, the launch date for Spaceborne-2 has already been set for February 20. If all goes according to plan, the 15th Northrop Grumman Cargo Expedition will deliver the new “space supercomputer” to the ISS. The ship Cygnus “SS. Katherine Johnson “. It is assumed that the operation of Spaceborne-2 will last at least 2-3 years. Applications are already open for experiments on the new space computing platform. The average waiting time for launching tasks will be 1-2 weeks.