Middle East Technical University High Performance Computing Usage
Survey Report – 2021
This survey and the report are prepared to understand the current state of High Performance Computing (HPC) usage at the Middle East Technical University (METU). There were 39 researchers who responded to the survey in Spring 2021. As seen in Figure 1, HPC users are affiliated with a diverse number of departments. The largest share of HPC users among the respondents are in Computer Engineering (17.9%) and Aerospace Engineering (15.4%) Departments.
Figure 1. Departments
In another question, the participants were asked whether they are currently using ULAKBİM-TRUBA HPC resources or not. As seen in Figure 2, most of the respondents are currently TRUBA users (66.7%).
Figure 2. TRUBA users
As seen in Figure 3, most respondents (51.7%) are using HPC services for running commercial or open-source software while 29.3% and 6.9% developing code specifically for HPC systems and doing research on HPC system architectures, respectively and 3.4% indicated that they do not use HPC services. As seen in Figure 4, HPC systems being used are mainly located in centers in Turkey (42.6%) or owned by their research group (41.2%) and 13.2% reported they use HPC centers abroad.
Figure 3. HPC Usage
Figure 4. HPC systems
Figure 5. CPU and GPU usage patterns
Typical usage patterns of CPUs and GPUs are given in Figure 5. As seen in Figure 5(a), 71.8% uses between 1-128 processors and only 7.7% uses more than 512 processors. As seen in Figure 5(b), while 30.8% require at most 10K, 30.8% require 10K to 100K cpu-hours. 10.3% do not use any CPUs at all. As seen in Figure 5(c), 38.5% uses 1-4 GPU cards and 30.8% do not use any GPU cards at all. 25.6% and 15.4% require 10K to 100K and more than 500K gpu-hours, respectively.
Figure 6. Job length
As shown in Figure 6, typical jobs require days (53.8%), hours (17.9%), weeks (17.9%) and months (10.3%).
Figure 7. Research areas
Respondents represent a diverse number of research areas (Figure 7). The largest number of researchers are working on Computer Vision (10.9%), Computational Fluid Dynamics (10.9%), Deep Learning (9.1%) and Machine Learning (7.3%).
Figure 8. Preferences for high priority on TRUBA systems.
The last question was “In case there was a mechanism for higher priority usage what would be the preferences of the users”. As seen in Figure 8, most users prefer either large number of gpu-hours (22.7%) or cpu-hours (20.6%), followed by long running times (18.6%), large number of CPUs (11.3%), large number of GPUs (9.3%). Only 4.1% of the respondents indicated that the ease of application is important. 12.4% indicated that they would apply under any condition.
– HPC working group @METU