Sunsetting of 2013 hardware on taki

To all taki users,

tl;dr: Equipment from 2013 (both the batch and gpu partitions) will be disabled on August 15, 2023.

These Dell PowerEdge R620 & R720 nodes were purchased in 2013 as a part of an NSF MRI “Acquisition of Hybrid CPU/GPU Nodes for the Interdisciplinary UMBC High Performance Computing Facility”. After installation, this hardware constituted the bulk of the UMBC community resource known as the maya cluster as a part of the UMBC HPCF. While these nodes represented the majority of the computing power of the cluster from 2013-2018, they have been supplanted by the 2017 MRI purchase, and the 2019 GPU ada cluster purchase and no longer see much usage. 

In addition, for the past year and a half, the utilization of the 2013 hardware partitions (batch and gpu) has been persistently decreasing to what is now a small fraction of their historical monthly utilization*. Also, the advanced age of the hardware is leading to more frequent hardware failures in these partitions, many of which cause jobs to fail, leading to lost work in addition to staff-hours associated with triage and repair. These nodes are also the least energy efficient of the cluster, using more power and cooling than the newer, more reliable nodes. 

Given all of these points, we will be disabling these partitions on August 15, 2023 so we can focus on keeping a high-performing and reliable cluster for the campus community.

Since 2013, other NSF MRIs (in 2017 & 2019) and various community buy-ins orchestrated by dedicated faculty and staff have grown the UMBC HPCF into what is now two clusters, ada and taki. As noted above, these machines represent comparable compute power with less demand of power, cooling, and maintenance by UMBC HPCF System Administrators.

We encourage any remaining workflows utilizing the 2013 nodes to transition to the newer hardware. As always, if you have any questions, comments, or concerns–please submit an RT ticket via the form found at 

(https://doit.umbc.edu/request-tracker-rt/doit-research-computing/) or by navigating to hpcf.umbc.edu > “Forms” > “HPCF Help Request Form”.

Roy Prouty

UMBC HPCF System Administrator

*For more information about usage trends on the taki cluster, see (http://hpcf-files.umbc.edu/research/usage_statistics.pdf) or navigate to hpcf.umbc.edu > “Research” > “Usage Statistics”