South Africa’s Center for High-Performance Computing (CHPC) hosts the largest and fastest computer in Africa, called Lengau. (“Lengau” means “Cheetah” in Setswana). And over the past year CHPC has increased Lengau’s core count by more than 33% — to more than 32,000 currently. As a result, CHPC has also increased its commitment to two key areas of earth science research notable both for their importance to humanity and their heavy computational load: climate change projections and weather forecasting.
Climate change research is particularly relevant in Africa — the continent is experiencing climate change much faster than other parts of the world but yet, as a developing region, has fewer resources with which to adapt to climate change. Hence, the urgency to support earth science research with more robust computing resources.
“Earth science projects have now started pushing to 14% of our cluster utilization,” says CHPC Director, Dr Happy Sithole. “We can now provide the number of cores they need for the large simulations they run. So they’ve become more prominent users.” Fast I/O is another requirement of earth science modelling, so in addition to increasing the core count, CHPC also migrated this past year to the latest Lustre distributed file system. “Performance is much better and users can do better computations because more simulations can run simultaneously,” Dr Sithole says.
Good for Africa
The new model has been named the Variable-resolution Earth System Model (VrESM) and is the first earth system model to originate from Africa, he says. “Several earth system models have been developed over the last two decades, but ours is only the second model of its kind to be developed in the Southern Hemisphere. Having an earth system model developed through an African lens is an important leap forward for African climate science and for building human capacity in the climate sciences in Africa.”
A reliable climate model is good for Africa for two reasons. Prof Engelbrecht says. “The first one is the fact that Africa has what we call a limited adaptive capacity. If you are a rich developed country it is easier to make extensive adaptation investments to prepare for the climate change that is upon us. But if you are a developing country where climate change is only one of a number of challenges to work on, it may be very difficult to make sufficient investments in climate change adaptation. So because adaptive capacity is relatively low in Africa it means that we are more vulnerable to present-day climate variability and future climate change.
“The second reason is that Africa has quite a strong climate change signal. With that we mean that in many parts of Africa climate is changing exceptionally fast — at a much higher rate than the global rate of climate change. And that is specifically true in Southern Africa, in the subtropics, and also in subtropical North Africa. In these two very vast regions of Africa, the so-called African subtropics, temperatures are rising at about twice the global rate of temperature increase. That means that if the Paris agreement is trying to keep the global temperature rise below 2 °C, it still means a temperature increase of about 4 °C on the average in subtropical Africa.”
More Upgrades Coming
Dr Sithole, the CHPC director, is already taking steps with his vendor partners to ensure the Lengau cluster will handle even more complex modelling scenarios in the future. He cites the 4-node test cluster Intel recently provided to CHPC.
“One thing that we have also been doing with Intel is to look at what happens with the technologies going forward. We have a smaller test system from Intel that has got newer technologies, like Intel Xeon scalable processors, the next-generation Xeon Phi processor, and Intel Omni-Path Architecture interconnect fabric. So with that we’ll do some benchmarks of our current workloads to determine what other configurations we can be looking at.”
The test cluster also incorporates Intel FPGA chips, which CHPC will leverage for evaluating new test cases in artificial intelligence and machine learning.
Given CHPC’s heavy workloads, Dr Sithole sees such vendor support as vital for supporting its current users — including the earth science researchers — while also preparing for future commitments.
“Intel is providing the test cluster. If I did not have this support I would have to go out and buy all these things and do the setup and testing myself. It’s a lot of effort. So for them to work together with us to provide access to the newer technologies — that gives us an opportunity to do the validation better and also make better decisions.”
By Randall Cronk, Owner, LLC