Koa

 

Koa came online June of 2023. It is the second University of Hawai‘i (UH) high performance computing (HPC) cluster –  a collection of many computers called nodes connected together with a network –  that solves computational problems too large for standard computers. Koa is operated by UH Information Technology Service Cyberinfrastructure, which serves as a central university-wide computational resource that supports data and computationally intensive research in over 90 disciplines. Koa consists of 263 nodes (7219 cores) with a total of 52 TB of RAM, 146 GPUs and more than 6 PB of storage. Koa is a continuation of Mana containing equipment deployed in late 2014, with investments from the University, an NSF MRI award ( #1920304) in 2019, two CC* awards (#2201428 and #2232862) in 2022 and 2023, and ongoing Principal Investigator (PI) purchases of condo compute nodes. Koa is free to use for all UH users, with the Slurm Workload Manager managing access with a fair share algorithm. UH community users get uninterrupted access to the all university-owned machines, but can also schedule work on the PI-owned nodes via partitions subject to preemption by node owners' job submissions.


 

Koa is FREE to use by all UH faculty, staff and students across the 10-campus system, but unlike UHUNIX, access is not granted by default to everyone at UH.

Prerequisites to access and use Koa:

  • Be an active UH faculty, staff or students affiliated with at least one of the 10 campuses in the UH system

  • Sign up for and attend an onboarding session.  Registration

  • SSH client or web browser

  • Koa’s hostname: koa.its.hawaii.edu


 


Koa Related News

News related to Koa and featured stories of research done with the help of Koa can be found at the Hawaii Data Science Institute's website. 


Hardware

 

8,842  processor cores

Intel processor types:

AMD processor types:

298 nodes

Single, Dual and Quad-socket nodes

2 login nodes ( koa.its.hawaii.edu )

4 vCPUs, 16 GB RAM

2 Open Ondemand nodes ( koa.its.hawaii.edu )

4 vCPUs, 24 GB RAM

1 Data transfer node (non-Globus) ( koa-dtn.its.hawaii.edu )

10 vCPUs, 24 GB RAM, 100 Gbit/s Ethernet

4 Data transfer nodes (Globus)

32 vCPUs, 64 GB RAM, 100 Gbit/s Ethernet

53 TB of total system memory

Memory per computational nodes vary from : 72 GB, 96 GB, 128 GB, 192 GB, 256 GB, 512 GB and 1 TB of RAM

Mellanox HDR Infiniband

  • Topology: Fat tree with a 4:1 blocking factor

  • Bandwidth: 200 & 100 Gbit/s

  • In Use: Yes

  • Expanding: Yes

1/25/100 Gbit Ethernet network

Computational nodes are not provided public IPs, but are capable of accessing external networks and websites


 

Â