Indico: https://indico.cern.ch/event/1595861/#2-tcb-report
The TCB is working on the presentation for CHEP26. There was a talk rehearsal for comment at OTF#11 today (Zoom only), just before this meeting. The presentation is a week from today so we can still collect comments.
Updated FTS4 timelines were presented in the CMS Offline and Computing Week Open Session [Indico] on March 19th and the April 29th DOMA General Meeting [Indico]. The user input freeze will happen in June and the first official release is planned in Q4 of 2026, in time for DC27 in February 2027. Deployment in production will take place first at CERN before DC27 and coordinated with other sites afterwards in tandem with the Alma10 rollout.
OTF#10, with the theme of medium-to-long-term evolution of sites, was co-located with the HEPiX Spring Workshop in Lisbon, Portugal on Tuesday, April 21, 2026 [Indico]. Live notes are attached to the Indico agenda page.
There were presentations and discussions on operational effort at sites, tape systems and performance evolution, storage capacity and throughput evolution, and a panel discussion on deploying services with k8s.
A survey on operational effort is planned to be completed before the HEPiX Fall Workshop in October in Nebraska, since the last one was done in 2014. This survey will be coordinated by WLCG (Ops and TechCoord) and Pepe.
Understanding the storage I/O requirements of the experiments is important (e.g., for the WLCG Technical Roadmap) especially noting that capacity is increasing significantly quicker than throughput both for tape and disk. Flash storage may help performance in this respect, but there is no unified approach (caching, tiered storage, or standalone).
K8s was seen as a useful way to orchestrate services but by no means the only way to do so and that it’s important to choose the “right tool for the job.” Advantages include the potential for sharing Helm charts and configurations, increasing the knowledge base of operations staff, possible room for distributed operations though recognizing that the learning curve is steep, which can challenge small site teams, and that a multilayered infrastructure complicates the debugging in case of issues.
Two WLCG federations, U.S. CMS and IHEP (China) presented their facility evolution plans.
U.S. CMS presented their plans to concentrate future storage purchases at the Fermilab Tier-1 and processing (including for analysis facilities) at the university Tier-2 sites, driven by datacenter limitations and simplification of sites to allow scaling up for HL-LHC with existing levels of effort. There were questions about Alpaka (performance portability layer) and analysis facility plans.
China will run university network paths though IHEP, taking advantage of the recently-upgraded LHCOPN/LHCONE connections at 100 Gbps to the international research networks. IHEP is also planning more vendor-heterogeneous (locally produced) CPU and GPU purchases to support ML/AI.
Also at the HEPiX Spring Workshop, O. Smirnova (Lund) presented on behalf of the TCB about the December 2025 WLCG Workshop on Heterogeneous Architectures [Indico] on April 22nd.
This presentation was repeated on May 13th at the EPIC Workshop on Heterogeneous Computing [Indico] by D. Benjamin (BNL), where there were questions and comments about the potential for collaboration across experiments or with governments to develop ML training infrastructure, about portability layers and vendor choice, and the cost evolution of GPUs.