KV Offloading Usage Guide

This guide covers configuration of the OffloadingConnector, which extends the prefix cache by offloading completed KV blocks to slower but larger tiers (CPU host memory, plus optional secondary tiers) as they are produced. Hits in the offload tiers are promoted back to GPU on demand. Transfers between GPU and CPU use DMA (cudaMemcpyAsync) and run asynchronously alongside model computation, so offloading adds minimal CPU- and GPU-core overhead.

!!! note The OffloadingConnector currently supports CUDA, ROCm, and XPU only.

Overview

Two specs are available, selected by the spec_name key in kv_connector_extra_config:

• CPUOffloadingSpec (default): single CPU tier. Completed GPU blocks are copied into pinned host memory.
• TieringOffloadingSpec: multi-tier. A CPU primary tier plus one or more secondary tiers.

Only the CPU primary tier has direct GPU access. Secondary tiers cannot read from or write to GPU memory; all GPU↔secondary transfers are staged through the CPU primary tier.

flowchart LR
    GPU <--> CPU["CPU primary tier"]
    CPU <--> S0["Secondary tier 0"]
    CPU <--> S1["Secondary tier 1"]
    CPU <--> SN["..."]

Single-Tier Setup (CPU Only)

vllm serve <model> \
  --kv-transfer-config '{
    "kv_connector": "OffloadingConnector",
    "kv_role": "kv_both",
    "kv_connector_extra_config": {
      "block_size": 64,
      "cpu_bytes_to_use": 1000000000
    }
  }'

Multi-Tier Setup

Set spec_name to "TieringOffloadingSpec" and supply a secondary_tiers list. Each entry is a dict with a required type key plus tier-specific fields. The list is ordered: tier 0 is consulted before tier 1, and so on. See Secondary Tiers for tier-specific keys.

vllm serve <model> \
  --kv-transfer-config '{
    "kv_connector": "OffloadingConnector",
    "kv_role": "kv_both",
    "kv_connector_extra_config": {
      "spec_name": "TieringOffloadingSpec",
      "cpu_bytes_to_use": 10737418240,
      "block_size": 16,
      "eviction_policy": "lru",
      "secondary_tiers": [
        {
          "type": "fs",
          "root_dir": "/mnt/kv_cache",
          "n_read_threads": 32,
          "n_write_threads": 16
        }
      ]
    }
  }'

kv_connector_extra_config Reference

Secondary Tiers

Each entry in secondary_tiers is a dict with a required type field plus tier-specific fields.

The filesystem and object-store tiers can publish hash-only BlockStored KV events for blocks they successfully store, tagged with a stable per-tier medium (FS for the filesystem tier, OBJ for the object-store tier). Set enable_kv_events: true in the tier's entry to opt in; events are published only when KV cache events are also enabled globally via --kv-events-config.

Set the optional locality tier field to LOCAL or REMOTE to describe the tier's storage location relative to the publishing vLLM instance. LOCAL marks storage local to that instance, while REMOTE marks storage that is not local to it. When the setting is omitted, locality is unspecified. vLLM does not infer it from the tier type, so an OBJ tier is not implicitly REMOTE. A KV event includes locality only when the tier explicitly configures it. This metadata describes the tier property without implying that a consumer can already route requests to its blocks.

Filesystem (FS)

The filesystem tier (type: "fs") writes blocks to a filesystem directory.

Each thread group prefers its own queue but pulls from the other when its primary queue is empty, so a write-heavy or read-heavy burst won't leave the off-priority queue waiting. Size the totals to your storage's effective concurrency.

On-Disk Layout

Under root_dir, vLLM creates a subdirectory <model>_<digest>, where <model> is the model name with / replaced by _ (so HuggingFace IDs like meta-llama/Llama-3-8B don't nest), and <digest> is a short SHA256 prefix derived from the run configuration (model, block size, parallelism, dtype, etc.). Runs with the same configuration share the same subdirectory; runs with different configurations live side-by-side under the same root_dir without colliding.

Inside that subdirectory, blocks are sharded across hash-prefix subdirectories to limit directory fan-out:

<root_dir>/
  <model>_<digest>/
    config.json
  <model>_<digest>_r<rank>/
    <hhh>/                    # first 3 hex chars of the block hash
      <hh>_g<group_idx>/      # next 2 hex chars + KV cache group index
        <hash_hex>.bin        # full block hash (in hex)

config.json records the run (block size, number of KV groups, etc.) and is written on first start. Each rank writes blocks under its own _r<rank> sibling directory, so multiple ranks can safely share the same root_dir.

Cross-Process Sharing

To enable KV cache sharing between multiple vLLM instances using the same root_dir (e.g., via a shared PVC), the PYTHONHASHSEED environment variable must be set to the same fixed value (e.g., "0") on every instance. Without this, each process initializes NONE_HASH (the chain-hash seed for block content hashes) with random bytes, producing different block filenames for identical token content.

PYTHONHASHSEED=0 vllm serve ...

Object Store (OBJ)

The object-store tier (type: "obj") offloads blocks to an S3-compatible object store through the NIXL OBJ backend.

store_config fields:

Object keys follow the same run-configuration digest scheme as the filesystem tier (see On-Disk Layout) and are stored under the optional prefix. The Cross-Process Sharing requirement (PYTHONHASHSEED) applies to shared buckets as well, so instances sharing a bucket produce identical keys for identical content. At startup the tier probes object store connectivity and fails fast with a configuration error if the bucket is unreachable.

P2P (Including P/D)

The P2P tier (type: "p2p") shares completed KV blocks between vLLM instances over RDMA via NIXL. Each instance binds a control socket on host:port and exchanges blocks directly with peers — no shared filesystem required.

PYTHONHASHSEED environment variable must be set to the same fixed value on all nodes.

The backends and num_threads options mirror the conditional logic used by NixlConnector: when any non-UCX backend is configured, NIXL is initialised with backends=...; otherwise it falls back to a UCX-only agent with the configured num_threads. This lets the P2P tier use a different transport (e.g. MOONCAKE, GDS_MT, LIBFABRIC) than the main NixlConnector running in the same process.

Environment Variables

Rather than embedding host/port in each secondary_tiers entry, set them once at deploy time via environment variables (mirroring VLLM_NIXL_SIDE_CHANNEL_HOST/VLLM_NIXL_SIDE_CHANNEL_PORT). Explicit host/port config keys, when present, take precedence.

• VLLM_P2P_SIDE_CHANNEL_HOST (default localhost): address the P2P control socket binds to. It is used verbatim as both the bind address and the identity peers dial back — there is no auto-detection (this mirrors VLLM_NIXL_SIDE_CHANNEL_HOST). The default binds the loopback interface only, so peers on another host cannot reach it. For any cross-host P2P deployment you must set this explicitly to the node's routable IP (e.g. the pod IP) before launching vllm serve — otherwise remote peers will fail to connect. The NIXL agent name is a separate per-process identifier, so peers sharing a host:port never collide.
• VLLM_P2P_SIDE_CHANNEL_PORT (default 5710): base port for the P2P control socket. The port actually bound is VLLM_P2P_SIDE_CHANNEL_PORT + data_parallel_index — one socket per DP replica, matching NIXL (for DP=1 the offset is 0). The peer's port is passed as remote_port in kv_transfer_params; the router/EPP that selects the DP rank (e.g. via the X-data-parallel-rank header) computes remote_port = base + rank. The DP-index offset separates replicas within one deployment; two co-located deployments (a prefiller and a decoder on the same host) still need distinct base ports (e.g. decoder base 5711) to avoid a bind collision.

Tuning Tips

• cpu_bytes_to_use: a bigger CPU tier means fewer trips to slower secondary tiers and a higher hit rate. The value is total across all workers, not per-worker. Leave headroom for the rest of the host workload.
• For single-tier (CPU-only) setups, set cpu_bytes_to_use larger than the aggregate GPU KV cache. Because offloading is immediate, a smaller CPU tier just mirrors what the GPU already holds and adds no hit rate.
• block_size: larger offloaded blocks reduce per-block bookkeeping overhead but increase the granularity of lookups. Must be a multiple of the GPU block size.
• FS thread counts: tune n_read_threads and n_write_threads to the parallelism your storage can sustain. Reads are latency-sensitive on the prefill path, so prefer more read threads when prefill hit rates are high.
• Sharing root_dir across runs: runs with the same model, block_size, parallelism layout, and dtype share files under the same <digest> subdirectory. Changing any of these produces a new subdirectory; old ones are orphaned but harmless. Delete them to reclaim disk.

Per-Request Selective Offload

Individual requests can cap how many of their tokens are eligible for offload by setting max_offload_tokens in the request's kv_transfer_params. Only the first max_offload_tokens tokens of the request are offloaded; blocks beyond that point are skipped on the store path. This is useful when a known prefix (e.g., a system prompt or shared context) is worth caching but later request-specific tokens are not.

!!! note max_offload_tokens is experimental and subject to change.

Example (OpenAI-compatible completions request):

{
  "model": "<model>",
  "prompt": "...",
  "kv_transfer_params": {
    "max_offload_tokens": 1024
  }
}

Further Reading

• vLLM blog: KV Offloading Connector — motivation, architecture (DMA-based async transfer), and benchmarks (TTFT and throughput).