MIRAGE: KV Cache Optimization through Parameter Remapping for Multi-tenant LLM Serving

2025-07-15

AI Summary

In multi-tenant LLM serving, dynamic KV cache updates induce high CPU memory bandwidth pressure and frequent GPU–CPU data transfers. To address this, we propose Parameter Remapping: leveraging the static nature of model parameters, it virtualizes their idle memory space at runtime and dynamically repurposes it for KV caching—enabling efficient reclamation of memory from inactive models in multi-tenant settings. The method exploits high-bandwidth interconnects (e.g., NVIDIA Grace Hopper) and lightweight memory virtualization to minimize cache management overhead. Experiments show that, compared to vLLM, our approach reduces tail-token latency by 44.8%–82.5%, first-token latency by 20.7%–99.3%, and improves throughput by 6.6%–86.7%.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

KV cache accelerates LLM inference by avoiding redundant computation, at the expense of memory. To support larger KV caches, prior work extends GPU memory with CPU memory via CPU-offloading. This involves swapping KV cache between GPU and CPU memory. However, because the cache updates dynamically, such swapping incurs high CPU memory traffic. We make a key observation that model parameters remain constant during runtime, unlike the dynamically updated KV cache. Building on this, we introduce MIRAGE, which avoids KV cache swapping by remapping, and thereby repurposing, the memory allocated to model parameters for KV cache. This parameter remapping is especially beneficial in multi-tenant environments, where the memory used for the parameters of the inactive models can be more aggressively reclaimed. Exploiting the high CPU-GPU bandwidth offered by the modern hardware, such as the NVIDIA Grace Hopper Superchip, we show that MIRAGE significantly outperforms state-of-the-art solutions, achieving a reduction of 44.8%-82.5% in tail time-between-token latency, 20.7%-99.3% in tail time-to-first-token latency, and 6.6%-86.7% higher throughput compared to vLLM.

Problem

• Optimizes KV cache memory usage in multi-tenant LLM serving
• Reduces CPU-GPU swapping overhead via parameter remapping
• Improves latency and throughput in dynamic inference environments

Innovation

• Remaps model parameter memory for KV cache
• Avoids KV cache swapping in multi-tenant environments
• Leverages high CPU-GPU bandwidth for performance