Software Engineer in Warm Storage team, Meta

Seattle, WA, United States

• Technical leader of metadata team on distributed Warm Storage (a.k.a. Tectonic) having 100EB+ and 70 trillion queries/day. Supported Meta’s GenAI (LLaMA 3/4) with flash, and unblocked 96K GPU model training on checkpoint and data loading congestions by developing request prioritization and root-causing/fixing P100 tail latencies. Also ran power efficiency program for better GPU training capacity in data centers.

• Led various Metadata projects for scalability, reliability, and efficiency; regional-scale metadata storage project (20EB per cluster) to achieve 10x horizontal scalability, multi-tenancy overload protection for metadata’s critical resources (DB shard limit, CPU, capacity), file API SLO program and improved its P99 latency SLO by 66%.

• Developed core features/fixes across dependencies; obtained 20x vertical scale for open file operation, improved client’s RPC strategy, fixed ZippyDB’s inconsistent results on coalescing workflow, improved daily statistics landing time by 75%, batched traffic from GC components to reduce overload in Zippy, etc.

Software Development Engineer in Automated Profitability Management team, Amazon

Vancouver, BC, Canada

• Led integration of automated vendor negotiation system into an organization-wide vendor data analysis platform. Clarified business/technical requirements with counterparts and developed/designed seamless UX/UI with backends. Resolved technical challenges such as latency reduction with ElasticSearch to fulfill SLAs and multi-threading to optimize workflow.

• Designed and implemented the multiple components across the team – user permission system with simple/scalable onboarding mechanism, notifier for negotiation status updates, tracking system for customer agreements to prove actions, dynamic term selection for negotiators to choose the order of negotiating terms with validating correctness, etc.

Software Engineer in Cloud Platform Part, Kakao Corporation

Seongnam, Republic of Korea

• Designed and implemented the automatic data migration system that transferred old/unmodified data to cold storage reducing size by 33% with an optimized scanning strategy in hundreds of petabyte scale distributed file system.

• Designed a caching file system with a regex cache invalidation strategy, reducing the fluctuation of IO performance by 80%, and optimizing lock logic with a bloom filter to prevent slowdown from overloaded contents, increasing speeds by 300%.

• Developed and operated a large-scale cache service (~20k TPS per host, 500+ hosts) with 2 layered distributed clusters, cache invalidation worker servers, the serverless service (Openwhisk) for in-house developers.

• Developed a fast memory cache plugin for high traffic – mid volume services, allowing 400% of traffic without delay.

• Managed a key-value storage component (HBase) for file metadata that recovers lost storage items, gathered/analyzed the statistics of storage usage/traffic by services, and reformed meta structure to achieve better consistency.

Yunho Kim, Taejin Kim, Moonzoo Kim, et al., Effective Integer Promotion Bug Detection Technique for Embedded Software, Journal of KIISE: Software and Applications, Vol. 43, Num 6, Jun 2016 

Taejin Kim, Moonzoo Kim, et al., Detecting Integer Promotion Bugs with Embedded Software using Static Analysis Technique, Korea Computer Congress, Dec 17-19, 2015 (Best Paper Award)