Research Agenda

On this page

Active research areas

Ignara Labs is conducting research across five primary areas. Each area addresses a fundamental infrastructure challenge for AI workloads.

Memory architecture

• CXL-based memory disaggregation for AI inference clusters
• KV-cache subsystem design for LLM serving (paged attention, eviction policies)
• Memory tiering strategies for heterogeneous memory hierarchies
• Bandwidth-compute tradeoffs in disaggregated memory systems

Scheduling research

• Gang scheduling algorithms for large distributed training jobs
• Network-topology-aware placement for collective communication workloads
• Fairshare scheduling with preemption in multi-tenant GPU clusters
• Spot/preemptible workload admission control and checkpointing

Storage research

• Prefetching strategies for training data pipelines
• Distributed checkpoint protocols with fast recovery
• Storage tiering for training workloads (NVMe, DRAM, remote object store)
• I/O amplification patterns in large-scale model training