How to Make Solving I/O Bottlenecks Look Easy

We live for a good customer challenge. And you can’t get a much better challenge than one that turns into a new product with broad application.

The challenge came to us by way of Los Alamos National Laboratory (LANL) and the National Nuclear Security Administration (NNSA).

LANL launched their “Trinity” supercomputer in 2015. Based on Cray® XC™ series architecture, Trinity is the first installation of the NNSA Simulation and Computing program’s initiative to develop and deploy a series of advanced technology systems. Together they’re exploring compute technology so they can provide platforms with higher performance for predictive capability — all while solving national security challenges. In short, NNSA and LANL are working on some pretty tough, forward-looking problems.

As part of their work they run high-resolution 3D simulations that generate extremely large and valuable datasets, but they found those datasets were limiting application performance and impacting throughput. Read our case study to learn about the performance challenges facing LANL.

Restrictive data flow and I/O bottlenecks made matters worse. LANL was left with only two options: 1) frequent checkpointing with reduced recovery periods but highly inefficient run times; or 2) bare minimum number of checkpointing runs and long recovery periods.

Neither option was sustainable. They’d reached the financially feasible limits of scaling with traditional HDD technology configurations. They needed a better solution.

LANL approached us to build a solution based on flash storage. Their goal was to achieve speeds of 3.2 TB/s, enable hourly flushing to the Lustre® file system and shorten recovery periods.

Spoiler alert: The solution to their problem turned into the Cray® DataWarp™ applications I/O accelerator — a major technology for us. DataWarp technology delivers a balanced and cohesive system architecture from compute to storage. It allocates storage dynamically in either private or shared modes. What it means for the user is streamlined workflows because applications process faster and access storage on demand. Efficient workflows mean productive users.

Read the case study to see the staggering results Cray and LANL achieved with the DataWarp solution and hear what LANL scientists had to say about it.

Speak Your Mind

Your email address will not be published. Required fields are marked *