Weather Service Doubles Up on Cray Supercomputer

When it comes to weather forecasting and climate modeling, supercomputer maker Cray has plenty of stories to tell...

To add to that bucket, South African Weather Service’s (SAWS) has performed a major upgrade of its Cray XC30 supercomputer, doubling its computational horsepower and tripling the system’s storage capacity. The expanded system now has 48 Xeon-powered compute blades and 1.8 petabytes of ClusterStor storage. The upgrades were delivered in March 2018 and the entire system was relocated to a new datacenter the following month.

The rationale for the expansion is the weather agency’s increased workload, which entails both daily operational forecasts and weather/climate research. Those services are delivered to various South African government agencies, as well as to aviation and marine organizations. The forecasts are also disseminated through radio and television station outlets. SAWS has the additional burden of being the onlyng

World Meteorological Organization (WMO) long-range global modeling center on the entire African continent.

In general, any additional computational power enables weather and climate models to run at greater fidelity, which directly translates into more accurate forecasts. So it’s fairly easy for these centers to make the case to their government sponsors and other customers that you get what you pay for in the weather prognostication business. There is also an underlying motivation for these forecasting centers not to fall behind in this technology as other weather agencies upgrade their systems. All of the top national centers in the US, Europe and Asia are now employing multi-petaflop supercomputers to run their models.

According to SAWS, the original XC30 system installed in 2014 was “inadequate to conduct the necessary research to improve on its modelling forecasts and applications research in a bid to stay up to date with other international organisations.” As result, the agency offloaded some of its modeling research to Lengau, the country’s only petascale supercomputer, which is managed by South Africa’s Centre for High Performance Computing (CHPC). Lengau also serves as a backup system for operational weather forecasts, in the event of a system failure at SAWS.

As it turns out, that has yet to occur. The original XC30 has been running continuously with no downtime since it booted up five years ago. According to Ilene Carpenter, earth sciences segment director at Cray, it’s not atypical for a system the size of the SAWS machine to run flawlessly for several years. “A small configuration like this is extremely stable,” Carpenter told us.

Cray routinely deploys much larger XC systems for weather forecasting centers – UK’s Met Office operates an 8-petaflop XC40, for example – and all of these machines are required to have availability north of 99 percent. Unlike most HPC applications, weather modeling must be performed in near-real-time since daily forecasts have to be delivered in a timely manner to be of any use. As an aside, a number of these weather/climate centers buy two identical systems, one for operational forecasts and one for research. If the forecast system goes down or has to be patched, the operational work is temporarily shifted to the research machine.

Thanks to that kind of reliability, over the past several years, Cray has come to dominate the weather/climate modeling space. The XC40 is the workhorse in that regard, claiming nine such supercomputers of a petaflop or better, installed across the UK, Germany, India, and Korea. In aggregate, Cray supplies supercomputing machinery to over 85 percent of the weather forecasting centers on the planet, according to Carpenter.

The SAWS machine is not in petascale territory, and in fact, even after the recent expansion probably tops out at less than 100 teraflops. The new blades are equipped with Intel “Ivy Bridge” processors, in this case, 2.7 GHz Xeon E5-2697 v2 CPUs. The original 24 blades are powered by the marginally slower 2.4 GHz Xeon E5-2695 v2.

Unless you’re stuck in a time warp, you’ll notice that these are not exactly the latest and greatest Xeons Intel has to offer. Since the Ivy Bridge chips debuted in 2013, they have been succeeded by the Haswell, Broadwell and now the Skylake processors. Likewise, the XC30 is a couple of generations back in Cray’s XC portfolio. Most customers have moved on to the newer XC40 and XC50 lines. The differences mainly have to do with processor support; all three employ the same Aries interconnect.

Which probably means SAWS got a pretty good deal on the new hardware. How good a deal will remain a mystery, however, since Cray and the South Africans are keeping the contract details to themselves.