Workload Placement Versus Cooling Reality
An enterprise runs a 1.2 MW data center with 60% average utilization. Batch analytics jobs are scheduled at night on whatever servers are free, spreading load across the hall. The facility has hot spots in two rows and cold aisles that are overcooled elsewhere. IT wants to raise utilization to 75% to avoid new hardware purchases. Facilities warns that higher density in the hot rows will worsen the hot spots. There is no per-rack power metering, only per-row branch circuit monitoring. The batch scheduler is unaware of cooling topology.
1. How should workload placement be changed to raise utilization without creating new thermal risk?
Read the worked answer
Make placement cooling-aware. With only per-row branch circuit data, first map which rows are hot and which are overcooled, then constrain the scheduler to prefer rows with thermal headroom and avoid concentrating new work in the two hot rows. Consolidation should be done by filling underutilized servers in cool rows before spreading to hot rows. This raises utilization while keeping density where cooling can support it. Since per-rack metering is absent, use row-level power plus rack inlet temperature checks to validate. The goal is to raise average utilization without raising peak local density in constrained rows; if the scheduler cannot see topology, add a simple row-capacity tag to placement policy.
2. What is the efficiency argument for consolidation, and what measurement would prove it worked?
Read the worked answer
Consolidating workloads onto fewer servers lets idle servers be powered down or put in low-power states, reducing IT energy and the cooling load that follows it. The efficiency argument is that fixed overhead (server idle power, fan energy, overcooling of empty rows) is spread over more useful work. To prove it, measure IT energy and facility energy before and after, compute PUE and IT utilization, and confirm hot-row inlet temperatures did not rise beyond limits. Success is higher utilization with flat or lower facility energy and no new hot spots; failure is higher utilization with higher peak row power and thermal alarms.