Mooncake Performance

This section collects Mooncake performance evaluations and benchmark results for Mooncake core components.

Document	Area	Key Findings
Storage Benchmark	Mooncake Store storage	Measures end-to-end storage performance across Mooncake Store operations and deployment configurations.
Allocator Benchmark	Segment allocation	The optimized OffsetAllocator significantly improves utilization for uniform-size LLM KV cache allocation patterns.
Allocation Strategy Benchmark	Allocation routing	Compares random and free-ratio-first allocation across segments, replicas, skewed capacity, and DSA-style KV+indexer workloads.
SSD Offload Benchmark	Cache hierarchy	SSD offload extends the KV cache hierarchy with NVMe, reducing the performance cliff after DRAM cache capacity is exhausted in long multi-turn conversations.
Guide for tebench	Transfer Engine	End-to-end bandwidth and latency benchmarking for classic TE and TENT backends across block size, batch size, and concurrency.