llmtrafficlens 0.3.5

Profile LLM gateway logs for prefix-cache reuse potential, and export anonymized replayable traces

llmtrafficlens

Analyze an LLM gateway's request log for prefix-cache reuse, then export the log as an anonymized trace that standard benchmark tools can replay.

Two commands:

• profile — how much a prefix cache could save on this traffic, and how much cache memory that takes.
• export — the same log as a Mooncake- or bailian-format trace, with all text removed.

Install

pip install llmtrafficlens

No runtime dependencies. Python ≥ 3.10.

profile

llmtrafficlens profile gateway.csv -o report

Input. A CSV with a request_json column holding the OpenAI-format request body. These columns are used when present: response_json, model_name, status_code, and a timestamp column (timestamp, ts, created_at, ...; epoch or ISO-8601, or set --ts-column). Mooncake and qwen-bailian traces are accepted as input too (--format mooncake|bailian).

Output. report.html and report.json, containing:

• hit rate at each cache size, ending with the unbounded case — the ceiling;
• top prefixes by reuse count, and separately by reusable token volume;
• share of requests carrying a session identifier;
• input/output token distributions, streaming ratio, model mix, QPS.

How to read it. The ceiling bounds what prefix-aware routing can achieve on this traffic. A few percent means the work will not pay off.

The session-identifier share says whether session affinity alone would reach that ceiling. If most requests carry an identifier, affinity suffices. If none do, the reuse sits in prefixes shared between unrelated requests, and only prefix-aware routing can exploit it.

The two leaderboards rank differently, and the token-volume one determines savings: a short prefix reused very often contributes almost no reusable volume, while a long prefix reused a few dozen times can account for most of it.

The capacity curve is the input to cache sizing: each row is the hit rate if the engine can keep that many tokens of KV resident. Read where the curve flattens toward the ceiling, then convert tokens to bytes as shown in the next section. Two limits apply to that reading; both are in Methodology.

export

llmtrafficlens export gateway.csv --to mooncake -o trace.jsonl   # or --to bailian

A benchmark written by hand, with equally popular prompt groups, reports a higher hit rate than production reaches, because every group stays warm. Replaying the real log avoids that, and the export can be shared, because each request is reduced to one line of structure:

{"timestamp": 1753340000123, "input_length": 1994, "output_length": 117,
 "hash_ids": [4251731047194047120, 8125214104179782736, ...]}

hash_ids is a salted chained block hash: two requests share their first N hashes exactly when they share their first N blocks. Replay tools generate one synthetic block per hash, so the prefix-sharing structure is preserved while the content is not real.

# replay at the recorded pace
aiperf profile --custom-dataset-type mooncake_trace --input-file trace.jsonl --fixed-schedule ...
# replay at 2x the pace, for rate sweeps
aiperf profile --custom-dataset-type mooncake_trace --input-file trace.jsonl --synthesis-speedup-ratio 2.0 ...
# or SGLang
python -m sglang.bench_serving --dataset-name mooncake --dataset-path trace.jsonl ...

--to bailian adds chat_id, parent_chat_id and turn for AIPerf's bailian_trace mode; use it when the log carries session identifiers. Tell the consumer which block size the export used — AIPerf defaults to 512 for mooncake and 16 for bailian (--prompt-input-tokens-block-size).

We have not round-tripped an export through AIPerf or SGLang yet. The formats match their documented schemas, but replay is unverified.

Try it on a public trace

This needs no data of your own, and doubles as a correctness check:

curl -LO https://raw.githubusercontent.com/kvcache-ai/Mooncake/main/FAST25-release/traces/conversation_trace.jsonl
llmtrafficlens profile conversation_trace.jsonl --format mooncake -o mooncake-report

The curve over those 12,031 requests:

cache size (tokens)	hit rate
16K	3.3%
64K	4.2%
256K	4.3%
1M	5.6%
4M	18.5%
16M	34.2%
unbounded	37.4% (ceiling)

Cache size is how many tokens' worth of KV blocks the engine can keep resident at once; a block is evicted (LRU) once the cache is full. The unit is tokens because bytes depend on the model: multiply by layers × kv_heads × head_dim × 2 (K and V) × dtype_bytes, which is 128 KB/token for a 32-layer, 8-KV-head, 128-dim model in fp16. At that rate the 4M row is about 512 GB.

So this workload gains almost nothing below 1M tokens, and reaching most of its ceiling takes KV storage in the terabytes — more than device memory alone provides, which is the case tiered KV stores (host memory, SSD) are built for.

To cross-check, feed the same trace to the official KV Cache Hit Rate Simulator and compare its infinite-capacity ceiling. Our figure agrees with third-party analysis of this trace (a prefix-caching study reports ≈40% for the conversation workload, arXiv:2505.21919). We have not run an automated comparison against the simulator itself.

The repository also contains a synthetic sample log (examples/, not shipped in the pip package): 360 requests, three shared system prompts with skewed popularity, sparse session identifiers, real timestamps. It reports a 66.4% ceiling, and it shows the two leaderboards disagreeing — 16 tokens × 180 requests is 2.3% of the reusable volume, while 2400 tokens × 30 requests is 55.6% of it.

Methodology

Each metric follows a published construction. Deviations are listed.

• Hit rate — requests are replayed in timestamp order against an LRU block cache. Unbounded capacity gives the "ideal hit rate" of KVCache in the Wild (ATC'25), which is also the Mooncake simulator's infinite-capacity ceiling; bounded capacities give that simulator's capacity curve. LRU is the default eviction policy in vLLM and SGLang.
  • The simulation models one global cache, i.e. a single worker or perfect routing. With N workers behind hash routing, each sees a partition, so read the curve as fleet-level guidance.
  • An engine's KV pool also holds in-flight requests. Only the remainder is available for reuse.
• Block hashing — chained per-block hashing (hash(parent, block), 16-token blocks) follows the vLLM prefix-cache block scheme and the qwen-bailian trace convention. Deviation: in approximate mode we hash fixed-size character chunks rather than tokenizer output, so two prompts match only if they are byte-identical.
• Token counts — characters divided by four, unless the log carries usage, in which case exact counts are used and the report says so. CJK-heavy text tokenizes closer to 1–2 characters per token, so absolute counts run low while ratios remain usable.
• Coverage — the ceiling covers sharing between requests only. Intra-session reuse is not measurable without session identifiers in the log. Without a timestamp column, row order is used instead: arrival statistics are omitted and the export is flagged as unfit for fixed-schedule replay.

Privacy

Processing is local. Reports contain aggregates only. An export contains lengths, timestamps, anonymized session identifiers and salted block hashes; prompts cannot be reconstructed from it. The hash key is written to .ltl-salt on first run — keep it unchanged so runs stay comparable, and treat it as a secret.

Related tools and scope

We do not replay traffic, control request rate, or search deployment configurations. AIPerf, SGLang bench_serving and simulators such as Vidur do that, and all of them take a trace as input.

Observability platforms (Langfuse, Helicone, Datadog LLM Observability) record token counts and cost per request, not prefix structure. Trace analyzers such as the Mooncake simulator and aiperf analyze-trace require an already-hashed trace. As of July 2026 we found no tool that goes from a raw gateway log to a hashed trace, which is the step this fills.

License

Apache-2.0