Editorial commentary on distributed-systems research
From Gandiva and Pollux to Kueue and DRA, a survey of GPU scheduling research and the gaps Kubernetes schedulers still leave open for LLM training and serving.
Read more →
SOSP 2026 met in Prague September 29 to October 2. A reading list from the accepted papers on verification, LLM serving, shared logs and operations.
Read more →
SafarDB moves replicated-data-type transaction logic onto the FPGA-resident NIC itself, reporting 6.8x latency and throughput gains over RDMA-based replication.
Read more →
gcsfs 2026.8.0 makes concurrent prefetching the default read path, turning cloud storage reads from a network wait into a compute-bound operation.
Read more →
A systematic review of 122 papers on serverless HPC, AI, and big data maps eight years of research trying to fit compute-intensive workloads onto FaaS.
Read more →
A USENIX ATC 2025 paper called Barre shows how three small changes to an old AIMD control loop cut AI-training congestion at 400 Gbps with no new hardware.
Read more →
Archive page for the C4BB4C project — applying cloud infrastructure to computational biology workloads.
Read more →
Archive page for the TinOva project — hypervisor abstraction research from the DSA Research Group.
Read more →From Gandiva and Pollux to Kueue and DRA, a survey of GPU scheduling research and the gaps Kubernetes schedulers still leave open for LLM training and serving.
Read more →SOSP 2026 met in Prague September 29 to October 2. A reading list from the accepted papers on verification, LLM serving, shared logs and operations.
Read more →SafarDB moves replicated-data-type transaction logic onto the FPGA-resident NIC itself, reporting 6.8x latency and throughput gains over RDMA-based replication.
Read more →