Biotechnology and Genomics · Enterprise · $260K · owned by Yusuf Abdel-Rahman
Steve has not written one for this account. Everything it would read is below.
Every record Steve is allowed to cite
Head of Computational Biology
Research Platform Engineering Lead
Biotechnology and Genomics · Enterprise · $260K · owned by Yusuf Abdel-Rahman
Steve has not written one for this account. Everything it would read is below.
Every record Steve is allowed to cite
Head of Computational Biology
Research Platform Engineering Lead
Director of Research IT
Next step: Storage tiering and egress design review on 2026-07-30
A whole-genome variant-calling run across 9,400 samples exceeded the 24 hour maximum job duration and was killed twice. Split the pipeline into four checkpointed stages with resumable state, and the customer completed a full run in three stages of under 14 hours each.
Harborview moved a 412TB cohort between regions and the egress line came in well above the figure their team had modelled. Walked through the billing detail, confirmed the charges were correct, and documented an in-region staging pattern that avoids the transfer for future cohorts.
Restores from the cold object tier are taking eight to twelve hours against the team's working expectation of under one hour. Confirmed the behaviour is within the tier's published retrieval window. Customer has asked whether an intermediate tier exists. Still open.
Ben requested a written tiering recommendation across hot, warm and cold for the 2.4PB archive, including expected restore times per tier and the egress implications of the in-region staging pattern. Awaiting the design review.
Grete said the expansion has to fit inside a flat research infrastructure budget and mentioned that a grant renewal decision lands in September. Ben noted that pipeline submissions have fallen since the first week of July. He also asked again about an intermediate storage tier, referencing the eight to twelve hour cold restores. Agreed a design review at the end of the month to close out both the tiering model and the restore expectations.
Harborview holds a 2.4PB cohort archive of which roughly 70 percent has not been read in 12 months, and Ben wants a tiering design that keeps restore times predictable for the samples that do get recalled. Amara described the analysis pattern as a small number of very long pipelines rather than many short queries, so job durability matters more to her than concurrency. Agreed to produce a tiering and egress model covering the three tiers and the in-region staging pattern.
Director of Research IT
Next step: Storage tiering and egress design review on 2026-07-30
A whole-genome variant-calling run across 9,400 samples exceeded the 24 hour maximum job duration and was killed twice. Split the pipeline into four checkpointed stages with resumable state, and the customer completed a full run in three stages of under 14 hours each.
Harborview moved a 412TB cohort between regions and the egress line came in well above the figure their team had modelled. Walked through the billing detail, confirmed the charges were correct, and documented an in-region staging pattern that avoids the transfer for future cohorts.
Restores from the cold object tier are taking eight to twelve hours against the team's working expectation of under one hour. Confirmed the behaviour is within the tier's published retrieval window. Customer has asked whether an intermediate tier exists. Still open.
Ben requested a written tiering recommendation across hot, warm and cold for the 2.4PB archive, including expected restore times per tier and the egress implications of the in-region staging pattern. Awaiting the design review.
Grete said the expansion has to fit inside a flat research infrastructure budget and mentioned that a grant renewal decision lands in September. Ben noted that pipeline submissions have fallen since the first week of July. He also asked again about an intermediate storage tier, referencing the eight to twelve hour cold restores. Agreed a design review at the end of the month to close out both the tiering model and the restore expectations.
Harborview holds a 2.4PB cohort archive of which roughly 70 percent has not been read in 12 months, and Ben wants a tiering design that keeps restore times predictable for the samples that do get recalled. Amara described the analysis pattern as a small number of very long pipelines rather than many short queries, so job durability matters more to her than concurrency. Agreed to produce a tiering and egress model covering the three tiers and the in-region staging pattern.