Over at Desktop Engineering, Kenneth Wong has published “The Zombie Computer Survival Guide,” or “things to consider when shopping for HPC for simulation.”
Wong writes that “sooner or later, advanced CAE software user will have to face the invasion of the zombie desktop computers.” This is the inevitable point when underpowered desktop computers forced to run complex simulation jobs come to a near standstill, hovering between life and death.
In a recent reader survey conducted by Desktop Engineering on behalf of IBM Platform Computing, 39 percent of the 1,000-plus respondents said they run their applications exclusively on workstations, while only 11 percent use clusters exclusively. The simulations run the gamut from complete systems (60 percent) to large assemblies and models (58 percent) and parts and small assemblies (52 percent). According to Wong, the survey results indicate a lot of high-level simulation work is being performed on workstations, which means that many of the participants may be all too familiar with the zombie workstation syndrome.
Wong suggests two methods to address this inefficiency:
1. Set up an in-house cluster to tackle the CAE jobs and free up the workstations.
2. Offload them to a remote cluster or on-demand cloud computing vendor.
Yes, it’s the old buy or rent paradigm, but there are some new developments to watch out for, leading to some not-so-obvious choices to make when shopping for additional compute resources, according to Wong.
One decision is whether to opt for open source or commercial cluster-management software. The usual tradeoff here is that cheaper cost (free) open source versions generally require more in-house expertise. When it comes to the job-scheduling option, however, Wong suggests going with a commercial solution that is sanctioned by the simulation software vendor.
Companies that do not wish to invest in on-site HPC infrastructure or those who just need occasional extra oomph might want to take a look at the new generation of on-demand clusters. Rescale, for example, is an HPC cloud simulation platform that targets the CAE sector.
This kind of on-ramp to HPC might be a particularly good idea for this technical community because according to the DE-Platform Computing study, only 18 percent of respondents stated they are familiar with the benefits of HPC.
Wong also covers whether to go with Windows or Linux (there are benefits to each), the need to balance core counts or node counts with interconnect speed (watch out for performance bottlenecks), and the GPU-accelerated approach (compliments of NVIDIA’s new GRID system).
In his concluding paragraphs, Wong writes that “complex CAE jobs will almost always test the limits of individual workstations. When these jobs are running, the burden on the CPU is significant.” As manufacturers push the envelope on digital simulation to exploit its competitive advantages, CAE clusters, both on and off-site, promise a zombie-free future.