Sequoia (supercomputer)
| Company type | Supercomputer |
|---|---|
| Industry | Computing / Supercomputing |
| Founded | 2011 |
| Defunct | 2020 |
| Headquarters | Lawrence Livermore National Laboratory , United States |
| Products | Blue Gene/Q supercomputer |
IBM Sequoia was a petascale Blue Gene/Q supercomputer constructed by IBM for the National Nuclear Security Administration as part of the Advanced Simulation and Computing Program (ASC). Delivered to Lawrence Livermore National Laboratory (LLNL) in 2011 and fully operational by June 2012, Sequoia was decommissioned in 2020. Its last recorded position on the TOP500 list was #22 in November 2019.
On June 14, 2012, Sequoia claimed the top spot on the TOP500 ranking as the fastest supercomputer worldwide, reaching 17.17 petaflops on the LINPACK benchmark. This performance was 63% higher than the previous leader, the K computer, while utilizing 123% more cores. Additionally, Sequoia demonstrated improved energy efficiency by drawing 7.9 MW compared to the K computer’s 12.6 MW.
By June 2013, Sequoia had fallen to third place behind Tianhe-2 and Titan, and by November 2017 it ranked sixth.
Sequoia facilitated groundbreaking scientific computations, sustaining performance beyond 10 petaflops. The cosmological simulation HACC achieved nearly 14 petaflops on a benchmark involving 3.6 trillion particles, and the Cardioid cardiac electrophysiology model reached close to 12 petaflops with near real-time simulations.
The system ran on Linux: over 98,000 nodes operated IBM’s CNK operating system, while 768 input/output nodes ran Red Hat Enterprise Linux, connected through the Lustre filesystem.
Dawn prototype
IBM created the "Dawn" prototype utilizing the Blue Gene/P architecture, delivering 500 teraflops to evaluate design aspects for Sequoia. Delivered in April 2009, Dawn secured the 9th position on the Top500 list by June of that year.
Purpose
Sequoia’s primary role was to support nuclear weapons simulation, succeeding both Blue Gene/L and ASC Purple at Lawrence Livermore National Laboratory (LLNL). It also contributed to scientific disciplines including astronomy, energy, lattice QCD, genomics, and studies related to climate change.
Node architecture
Built upon the Blue Gene/Q framework, Sequoia consisted of 96 racks containing a total of 98,304 compute nodes (1,024 nodes per rack). Each node housed a 16-core A2 processor paired with 16 GB DDR3 memory, culminating in an aggregate of 1,572,864 cores and approximately 1.5 PiB of memory distributed over roughly 3,000 square feet. Nodes were interconnected through a five-dimensional torus network topology.
Job scheduler
The Lawrence Livermore National Laboratory utilized the SLURM scheduler—also employed by Dawn and China’s Tianhe-IA—to effectively allocate and manage Sequoia’s computing resources.
Filesystem
LLNL implemented the Lustre parallel filesystem enhanced with a Linux-port of ZFS serving as the Lustre Object Storage Device (OSD) to boost performance and add functionality.
In September 2011, the DoE selected NetApp to supply 55 PB of storage capacity for Sequoia’s filesystem infrastructure.
Power usage
Sequoia’s power consumption was approximately 7.8 MW, achieving remarkable energy efficiency at about 2068 Mflops per watt—roughly six times more efficient than Dawn and more than two and a half times that of the June 2011 TOP500 leader, the K computer.
Application
In January 2013, Sequoia marked a milestone as the first supercomputer to operate more than one million processing cores concurrently on a single task. The Center for Turbulence Research (CTR) at Stanford Engineering leveraged this capability to address a sophisticated fluid dynamics challenge—forecasting the acoustic emissions generated by a supersonic jet engine.