Founded in 2015, XtalPi combines quantum physics, artificial intelligence and robotic automation to support drug and materials research. Its work connects computational research with laboratory automation, creating an ongoing need for reliable computing jobs, clusters and application services. This case focuses on the technical platform and operations supporting that work.
AI researchIndustry2015FoundedShenzhenOne of its key operating bases, China
Computing jobs, cluster components and application services all support research workflows. When a job fails or a service slows down, engineers need to distinguish resource pressure from task failures and application issues.
Separate tools held logs, traces and metrics without continuous context. Investigations often stalled at the handoff between systems.
Multiple clusters and compute pools needed a shared health view. Aggregate capacity alone could not explain job status, GPU utilization or their relationship to application services.
Detecting an issue was only the beginning. Teams also needed severity-based alerts, follow-up and availability checks to coordinate investigation and recovery.
KEY RESULTS
Several×
Faster fault localization and root-cause analysis
Seconds
Business and compute-pool metric monitoring
Unified
Multi-cluster and application observability
SOLUTION
XtalPi × Guance — Solution
01
See clusters and compute pools in one place
Bring cluster health, compute-pool jobs and GPU utilization together with business metrics. Engineers can move from the overall operating picture to an individual job, using the same evidence for capacity discussions and incident investigation.
Integrate existing telemetry and progressively consolidate fragmented data into one platform. Preserving object, time and call relationships reduces repeated searches, helping make fault localization and root-cause analysis several times faster.
2
Faster investigation workflows: fault location and root-cause analysis
03
Connect detection, collaboration and verification
Severity-based alerts, inspections and incident tracking support follow-through, while synthetic tests check availability from the access side. Development and operations share evidence from the first alert through investigation and verification.
