Resource Allocation Reductions
Learn about resource allocation reductions for NREL's high-performance computing (HPC).
Fiscal Year 2022 Quarterly Allocation Reductions
NREL shifted to a new allocation reduction policy at the end of Q2 of fiscal year 2021 (March 31). The goal of the new policy was to simplify the reductions, and account for the fact that more and more of our users have requested that their allocations vary from quarter to quarter in response to changing project needs. This policy remains in effect for fiscal year 2022.
Shortly after Q1, Q2, and Q3 ends, allocations will be automatically adjusted to account for low utilization against planned usage. At the end of each quarter, the total allocation units (AUs) used for the year to date will be compared to the total planned usage for the year to date for each allocation. At the end of Q1, the total used for the year to date is compared to the Q1 planned usage; at the end of Q2, the total used for the year to date is compared to the Q1+Q2 planned usage, and at the end of Q3, the total used for the year to date is compared to the Q1+Q2+Q3 planned usage.
AUs are then removed based on the table below. Note that allocation reductions are meant to be cumulative over the course of a year, and not compounding. If, for instance, a project was reduced by 10,000 AUs for low usage in Q1, and the reduction table suggests the project should be reduced by 25,000 AUs at the end of Q2, 15,000 (25,000-15,000) AUs should be removed at the end of Q2 to make the total removal for the year to date equal to 25,000 AUs.
|Percentage of planned AUs used to date||Percentage of planned to date AUs removed|
|More than 70%||0% (No reduction)|
|Less than 70% but greater than 55%||20%|
|Less than 55% but greater than 40%||35%|
|Less than 40% but greater than 20%||55%|
|Less than 20%||80%|
To understand how this process would work, we consider the following two 100,000 AU allocations, one with a uniform distribution of planned AUs throughout the year, and one with a distribution designed to enable development in Q1 and production runs in Q2 through Q4. The two allocations are described in the table below.
|Quarter||Allocation "Renewables"||Allocation "Efficiency"|
If "Renewables" uses 9,000 AUs in Q1, 20,000 AUs in Q2, and 25,000 AUs in Q3, it would be reduced in the following manner:
After Q1, the project will have used 36% (9,000/25,000) if its allocation, leading to 13,750 (55% x 25,000) AUs being removed.
After Q2, the project will have used 58% [(9,000+20,000)/(25,000+25,000)] of its allocation, leading to 10,000 AUs (20% x 50,000) being potentially removed. However, because 12,500 AUs were removed in Q1, no removal is performed. Note that AUs are not restored in this case.
After Q3, the project will have used 72 % [(9,000+20,000+25,000)/(25,000+25,000+25,000)] of its allocation. No reduction would be made. The project would then begin Q4 with 33,500 AUs (100,000-9,000-20,000-25,000-12,500).
If "Efficiency" uses AUs under the same schedule, after Q1, the project will have used 90% (9,000/10,000) of its allocation and will not be penalized.
After Q2, the project will have used 72.5% [(9,000+20,000)/(10,000+30,000)] of its allocation and will not be penalized.
After Q3, the project will have used 77% [(9,000+20,000+25,000)/(10,000+30,000+30,000)] of its allocation, and will not be penalized. The project would then begin Q4 with 46,000 AUs.
Note that "Renewables" and "Efficiency" have the same total (100,000 AUs) but lose very different amounts of AUs over the course of the year. This is because the allocation request "Efficiency" is more closely tuned to the user's actual us of HPC resources. NREL allows users to tune their allocation request through the use of "profiles" in the allocation request to avoid this sorts of reductions.