A seismic and volcanic monitoring agency that needed to accelerate its scientific simulations
An Ecuadorian government institution founded in 1983, responsible for assessing and monitoring seismic and volcanic hazards across the country, was operating with simulation processing times that could extend to weeks or even months on conventional architectures. With an annual operating budget of approximately $850,000, the institution sought to determine whether the Power10 platform could radically reduce these processing times.
Infodataserv worked directly with the institution and the University of Bristol (UK), developer of the Kestrel software, to conduct performance tests on Power10 infrastructure and validate the operational feasibility of the environment for real-world use cases.
La Gasca landslide data for simulation testing
Kestrel performance tests were conducted using real data from a mass movement event that occurred in Quito. This made it possible to validate the system under computational load conditions equivalent to critical operational scenarios.
A mud and rock flow triggered by heavy rainfall that caused the drainage system in the basin to collapse. The event left at least 28 people dead, 48 injured, and 12 missing. A total of 555 people from 160 families were affected, with 7 homes destroyed and 27 declared uninhabitable.
The computational workload required to simulate this type of event — with high spatial resolution and multiple iterations — is precisely the kind of scenario where Power10's advantage over conventional Intel architectures becomes most relevant for emergency decision-making.
Two clear technical goals: compatibility and performance
Objective 1 — Validate Kestrel on Power10
Confirm Kestrel's compatibility, successful compilation, and stable execution on the Power10 RISC platform across all three software versions.
Objective 2 — Optimize processing times
Measure and compare Kestrel execution times across Intel and Power10 architectures, evaluating the actual performance improvement using the Speedup metric.
LPAR environment configuration
Create an LPAR with Power VM, configure an NL SAS disk array, install Ubuntu 22.04, and prepare all the libraries required by Kestrel.
Secure remote access via VPN
All installation, configuration, and validation activities were performed entirely remotely through a VPN connection to the institution's environment.
Three Kestrel versions evaluated across three test groups
The project validated three Kestrel compiler/execution variants on the Power10 environment to identify the optimal configuration for complex operational environments.
-
Serial Version (baseline) — Standard sequential execution. Used as the reference for calculating Speedup. On Intel, some Cotopaxi volcano simulations took up to 77 days to complete.
-
Parallel Version (OpenMP) — Process parallelization using multiple threads. On Power10, the La Gasca event simulation was reduced from days to just 9 hours.
-
AT16 Version (Advance Toolchain 16) — Advanced optimization compiler for POWER architecture. Evaluated in Group 1 alongside Serial and OpenMP for direct comparison.
| Group | Resolution | Versions Compared | Purpose |
|---|---|---|---|
| Group 1 (3m) | Standard | Serial · OpenMP · AT16 | Full comparison of the three variants |
| Group 2 (1m) | High Resolution | OpenMP | Evaluate behavior under higher computational load |
| Group 3 (2m) | Corrected | Corrected Version | Confirm consistency and reproducibility of results |
Power10 vs. Intel: a dramatic reduction in simulation times
The comparative results, provided by the client, demonstrate the advantage of the Power10 architecture over the Intel environments previously used, both in absolute processing times and in the Speedup achieved with the parallel version.
| Simulation | Intel — Serial | Power10 — Serial | Power10 — Parallel | Speedup |
|---|---|---|---|---|
| Cotopaxi (Case A) | 77 days | — | — | — |
| Cotopaxi (Case B) | 35 days | — | — | — |
| La Gasca | 3 days (Bristol) | 2d 15h | 9 hours | ~7× |
Specifications of the deployed Kestrel LPAR
The infrastructure was configured as a dedicated logical partition (LPAR) on the Power10 server, with remote VPN access and Fibre Channel storage connectivity. Validation confirmed full compatibility, sustained stability, and software portability on the RISC architecture.
| Component | Specification |
|---|---|
| Server | IBM Power10 — RISC · Power ISVs Alliances Leader Latin America |
| Processor (LPAR) | 1.6 physical cores converted into 24 logical threads |
| RAM | 48 GB allocated to the Kestrel LPAR |
| Storage | 60 GB · IBM FS5200 · NL SAS Array · 16 Gb/s Fibre Channel I/O |
| Operating System | Ubuntu 22.04 LTS on LPAR with Power VM |
| Virtualization | Power VM with additional LPARs available (LPAR 2…n) |
| Connectivity | IBM SAN Switch · Secure remote VPN access |
| Scientific Software | Kestrel (University of Bristol, UK) · OpenMP · AT16 Advance Toolchain |
Three-way collaboration: Infodataserv, client, and IBM Chile
The project involved the coordination of three teams with complementary roles: Infodataserv's technical team, the institution's infrastructure lead, and specialized support from IBM Chile's Technical Specialist Power.