Measuring Metadata Operation Throughput on Cloud Filesystems
Metadata operations dominate AI filesystem performance, not data throughput.
Desmond Kwarteng-Asare
Staff Writer
Desmond Kwarteng-Asare
Desmond came up through academic systems research, completing a postdoc on fault-tolerant state machines before joining the publication in 2021. He writes about how agent runtimes encode, checkpoint, and recover state, with a particular eye toward the failure scenarios that production deployments rarely anticipate until it is too late.
8 stories
Files vs. Databases vs. In-Memory State for Agent Context Storage
Different storage types prevent the memory failures that derail agents in production.
Desmond Kwarteng-Asare
Securing Agent Data Access Without Bespoke Authorization Middleware
Authorization must move outside the agent's process to prevent compromise from prompt injection.
Desmond Kwarteng-Asare
Iterative Context Discovery vs. Front Loaded Prompt Injection
Context engineering, not prompt writing, decides how well agents actually perform at scale.
Desmond Kwarteng-Asare
Replication Lag and Read-After-Write Inconsistency in Distributed Storage
Async replication creates windows where reads return stale data before replicas catch up.
Desmond Kwarteng-Asare
Data Corruption From Concurrent Uncoordinated Writes on Shared Filesystems
Multiple writers on shared filesystems corrupt data silently and predictably.
Desmond Kwarteng-Asare
POSIX Compliance Gaps in S3-Compatible Object Storage
Atomic rename and file locking remain incompatible with S3's distributed architecture.
Desmond Kwarteng-Asare
Mmap on Cloud-Backed Filesystems for Agent Inference Workloads
Memory-mapped inference on cloud storage creates latency traps that agents cannot afford.
Desmond Kwarteng-Asare