Over the past year we have noticed that water circulating through two of the Cooling Distribution Units (CDU) in Row 8a and their compute nodes has become significantly discoloured. This impurity causes cooling problems which as a result increases chances of node failure. To remedy FASRC has scheduled a full flush of these CDUs and their attached nodes. Unfortunately to do the full flush we have to fully power down all the nodes and drain the water, which is a process that takes several days to complete.

To minimize disruption we have staggered this work over two weeks. The work on CDU5 will take place from 9/22 - 9/25, while CDU6 will take place from 9/29 - 10/2. Lists of impacted partitions are below. For those impacted we recommend using other resources on the cluster during that time such as shared and gpu_h200 or any of the requeue partitions. Note this work impacts both Cannon and FASSE.

No jobs will be cancelled, rather blocking reservations are in place to naturally drain the nodes prior to the work.

Thank you for your patience as we work to improve cluster stability and hardware longevity.

CDU 5 (9/22 - 9/25) -Complete

Cannon:

arguelles_delgado_gpu_a100

arguelles_delgado_gpu_mixed

bigmem_intermediate

blackhole_gpu

eddy

gershman

hejazi

hernquist_ice

hoekstra

huce_ice

iaifi_gpu

itc_gpu

jshapiro

kovac

kozinsky

kozinsky_gpu

murphy_ic

ortegahernandez_ice

rivas

seas_compute

seas_gpu

siag

siag_gpu

siag_combo

sur

zhuang

FASSE:

fasse_ultramem

CDU 6 (9/29 - 10/2)

Cannon:

arguelles_delgado_h100

bigmem

dvorkin

eddy

enos

gpu

hsph

hsph_gpu

intermediate

itc_cluster

janson_sapphire

joonholee

jshapiro

olveczky_sapphire

sapphire

test

yao

yao_alphatns

yao_gpu

FASSE:

cnl