ROOCS 2022 annual summary
This report summarizes monthly usage of the ROOCS subsetting services operated by DKRZ, IPSL, and CEDA during 2022. Together, the services processed 357,821 requests, of which 335,145 were successful, and delivered 27.16 TB of subsetted data.
| Scope | Requests | Successful | Failures | Subsetted data | Peak concurrency |
|---|---|---|---|---|---|
| All sites | 357,821 | 335,145 | 22,676 (6.34%) | 27.16 TB | 143 |
| DKRZ | 238,873 | 225,346 | 13,527 (5.66%) | 16.24 TB | 28 |
| IPSL | 112,734 | 103,684 | 9,050 (8.03%) | 10.50 TB | 97 |
| CEDA | 6,214 | 6,115 | 99 (1.59%) | 0.42 TB | 143 |
Monthly request outcomes
The green portion of each bar represents successful requests; failures are shown in red.
All sites
DKRZ
IPSL
CEDA
Understanding the failures
Most failed requests were caused by users accessing the services directly through the CDS API rather than through the CDS portal. In this workflow there is no CDS catalogue interface to guide or validate the available parameters, so users commonly rely on trial and error to discover valid request combinations. These invalid requests are counted as failures even though the service itself is operating normally.
A smaller share resulted from temporary infrastructure problems, such as full temporary disks or unavailable Lustre storage. The statistics also include genuine processing failures discovered during normal operation, including issues related to calendar handling. The available dashboard data does not provide a reliable numerical breakdown among these causes.
Monthly subsetted data
The monthly values combine the data delivered by DKRZ, IPSL, and CEDA.
Monthly peak concurrency
For each month, overall concurrency is the highest peak reported by DKRZ, IPSL, and CEDA. The site peaks are not added because they may have occurred at different times.
Data and methodology
Request, failure, and subsetted-data totals are sums of the monthly DKRZ, IPSL, and CEDA dashboard exports. Successful requests are calculated as total requests minus failed requests. Peak concurrency is not additive: the combined monthly series uses the higher site value, and the annual value is its maximum.
Source note: CEDA contributed during the service transition through April. Its January export omits transfer volume, so the value is inferred from the published Q1 CEDA total of 420 GB. The April IPSL monthly export is missing; request outcomes and concurrency were recovered from the Q2 daily series, and its transfer value completes the published Q2 IPSL total of 1,366 GB. The September IPSL export contains only 49.88 GB, so this summary uses the value implied by the published Q3 IPSL total of 2,340 GB. The unusually high peaks of 143 for CEDA in January and 97 for IPSL in October come directly from the raw monthly exports.