#2212 - [PR #1388] [MERGED] [Store] - Add comprehensive storage backend benchmark suite - kvcache-ai/Mooncake

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

kvcache-ai/Mooncake#2212

bsakaguchi commented

2026-10-29 21:33:42 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/Mooncake/pull/1388
Author: @maheshrbapatu
Created: 1/15/2026
Status: ✅ Merged
Merged: 1/22/2026
Merged by: @ykwd

Base: main ← Head: benchmark

📝 Commits (2)

ac2fbfa benchmark
cd34aa3 fix formatting

📊 Changes

2 files changed (+2472 additions, -0 deletions)

View changed files

📝 mooncake-store/benchmarks/CMakeLists.txt (+12 -0)
➕ mooncake-store/benchmarks/storage_backend_bench.cpp (+2460 -0)

📄 Description

Description

Add extensive benchmarking infrastructure for evaluating storage backend
performance across multiple scenarios critical to LLM inference workloads.

Key features:

Support for all storage backends: OffsetAllocator, Bucket, and FilePerKey
Comprehensive test coverage:
- Initialization and recovery time
- BatchOffload (write) and BatchLoad (read) performance
- Multi-threaded concurrent operations
- Metadata lookup (IsExist) with hit/miss tracking
- Mixed read-write workloads with realistic contention
- Churn testing at near-capacity with overwrites
- Restart/recovery benchmarks

Technical highlights:

Thread-local statistics for lock-free hot path
4KB-aligned buffer pools for future O_DIRECT support
Data integrity verification via deterministic patterns
Multiple access patterns (sequential, zipf)
Cache mode support for cold vs warm testing

This benchmark is essential for performance validation and regression
testing of TTFT-critical read paths and storage backend optimizations.

Type of Change

Types
- Bug fix
- New feature
  - Transfer Engine
  - Mooncake Store
  - Mooncake EP
  - Integration
  - P2P Store
  - Python Wheel
- Breaking change
- CI/CD
- Documentation update
- Other

How Has This Been Tested?

Benchmarked with small block sizes.

Checklist

I have performed a self-review of my own code.
I have formatted my own code using ./scripts/code_format.sh before submitting.
I have updated the documentation.
I have added tests to prove my changes are effective.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/Mooncake/pull/1388 **Author:** [@maheshrbapatu](https://github.com/maheshrbapatu) **Created:** 1/15/2026 **Status:** ✅ Merged **Merged:** 1/22/2026 **Merged by:** [@ykwd](https://github.com/ykwd) **Base:** `main` ← **Head:** `benchmark` --- ### 📝 Commits (2) - [`ac2fbfa`](https://github.com/kvcache-ai/Mooncake/commit/ac2fbfaf28aa93893834189b5e8b0e741773aa5f) benchmark - [`cd34aa3`](https://github.com/kvcache-ai/Mooncake/commit/cd34aa38849e412bae3365697fa0296d628d0d74) fix formatting ### 📊 Changes **2 files changed** (+2472 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `mooncake-store/benchmarks/CMakeLists.txt` (+12 -0) ➕ `mooncake-store/benchmarks/storage_backend_bench.cpp` (+2460 -0) </details> ### 📄 Description ## Description Add extensive benchmarking infrastructure for evaluating storage backend performance across multiple scenarios critical to LLM inference workloads. Key features: - Support for all storage backends: OffsetAllocator, Bucket, and FilePerKey - Comprehensive test coverage: * Initialization and recovery time * BatchOffload (write) and BatchLoad (read) performance * Multi-threaded concurrent operations * Metadata lookup (IsExist) with hit/miss tracking * Mixed read-write workloads with realistic contention * Churn testing at near-capacity with overwrites * Restart/recovery benchmarks Technical highlights: - Thread-local statistics for lock-free hot path - 4KB-aligned buffer pools for future O_DIRECT support - Data integrity verification via deterministic patterns - Multiple access patterns (sequential, zipf) - Cache mode support for cold vs warm testing This benchmark is essential for performance validation and regression testing of TTFT-critical read paths and storage backend optimizations. ## Type of Change * Types - [ ] Bug fix - [X] New feature - [ ] Transfer Engine - [X] Mooncake Store - [ ] Mooncake EP - [ ] Integration - [ ] P2P Store - [ ] Python Wheel - [ ] Breaking change - [ ] CI/CD - [ ] Documentation update - [ ] Other ## How Has This Been Tested? Benchmarked with small block sizes. ## Checklist - [X] I have performed a self-review of my own code. - [ ] I have formatted my own code using `./scripts/code_format.sh` before submitting. - [ ] I have updated the documentation. - [ ] I have added tests to prove my changes are effective. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

closed this issue
added the
pull-request
label

Rows
Columns

[PR #1388] [MERGED] [Store] - Add comprehensive storage backend benchmark suite #2212