Performance/Sysbench: baseline gating, MB/sec normalization, clock sanity + docs/YAML updates

Runner/suites/Performance/Sysbench_Performance/README_sysbench.md

@@ -0,0 +1,227 @@
+# Sysbench_Performance
+
+This suite runs a **repeatable Sysbench performance baseline** on Yocto/QLI targets in a CI-friendly way.
+
+It runs the following test cases for one or more thread counts and for N iterations:
+
+- **CPU** (`sysbench cpu`)  
+- **Memory bandwidth** (`sysbench memory`)  
+- **Threads** (`sysbench threads`)  
+- **Mutex** (`sysbench mutex`)  
+- **File I/O throughput** (`sysbench fileio`) — **seqwr / seqrd / rndwr / rndrd**
+
+For each case it:
+- prints sysbench output live to console (and saves per-iteration logs),
+- records per-iteration KPIs into `*.values` files,
+- prints a compact per-iteration KPI line (`ITER_SUMMARY`) to console,
+- computes averages and writes `OUT_DIR/sysbench_summary.txt`,
+- optionally appends per-iteration + avg rows to a CSV (only if `CSV_FILE` is provided),
+- optionally enforces PASS/FAIL **baseline gating** (when a baseline file is provided or auto-detected).
+
+---
+
+## KPIs collected (and “better” direction)
+
+| Case | KPI name (run.sh / baseline key) | Unit | Better |
+|---|---|---:|---|
+| CPU | `cpu_time_sec` | seconds | **lower** |
+| Memory | `memory_mem_mbps` | MB/s | **higher** |
+| Threads | `threads_time_sec` | seconds | **lower** |
+| Mutex | `mutex_time_sec` | seconds | **lower** |
+| File I/O seq write | `fileio_seqwr_mbps` | MB/s | **higher** |
+| File I/O seq read | `fileio_seqrd_mbps` | MB/s | **higher** |
+| File I/O rnd write | `fileio_rndwr_mbps` | MB/s | **higher** |
+| File I/O rnd read | `fileio_rndrd_mbps` | MB/s | **higher** |
+
+> Note: even if you think of reads as “GB/s”, the suite **stores and gates in MB/s** for consistency.  
+> If you have a baseline in GB/s, convert to MB/s (GB/s × 1024).
+
+---
+
+## Baseline gating (PASS/FAIL)
+
+Baseline gating is enabled when either:
+- `BASELINE_FILE` / `--baseline FILE` is set, **or**
+- `./sysbench_baseline.conf` exists in the **same folder as `run.sh`** (auto-detected).
+
+`ALLOWED_DEVIATION` / `--delta` controls tolerance:
+
+- **Higher-is-better** metrics (MB/s):  
+  **PASS** if `avg >= baseline * (1 - delta)`  
+  (recommend **0.05** → “>= 95% of baseline”)
+- **Lower-is-better** metrics (seconds):  
+  **PASS** if `avg <= baseline * (1 + delta)`  
+  (recommend **0.05** → “<= 105% of baseline”)
+
+The console shows a per-metric GATE line and `sysbench_summary.txt` includes a `gate_*` section.
+
+---
+
+## Baseline file format (`sysbench_baseline.conf`)
+
+Plain `key=value` format (comments allowed with `#`). Keys are per-thread-count:
+
+```
+# ---- Threads = 4 ----
+cpu_time_sec.t4=3.483
+memory_mem_mbps.t4=4120.000
+threads_time_sec.t4=3.703
+mutex_time_sec.t4=0.004
+
+fileio_seqwr_mbps.t4=0.400
+fileio_seqrd_mbps.t4=29.300
+fileio_rndwr_mbps.t4=0.610
+fileio_rndrd_mbps.t4=29.300
+```
+
+Use `t1`, `t4`, `t8`, … matching your `THREADS_LIST`.
+
+---
+
+## Parameters (env vars / CLI)
+
+The suite is controlled by environment variables (and equivalent CLI flags). Typical params:
+
+### Common
+- `OUT_DIR` (default `./sysbench_out`) – where logs / values / summary go
+- `ITERATIONS` (default `1`)
+- `TIME` (default `30`)
+- `RAND_SEED` (default `1234`)
+- `THREADS_LIST` (default `"4"`) – space-separated list of thread counts
+- `TASKSET_CPU_LIST` (default empty) – passed to `taskset -c`, e.g. `"6-7"`
+- `BASELINE_FILE` (default empty; auto-detects `sysbench_baseline.conf` next to `run.sh`)
+- `ALLOWED_DEVIATION` (default `0.10`; recommended `0.05`)
+- `CSV_FILE` (default empty → **no CSV by default**)
+
+### CPU
+- `CPU_MAX_PRIME` (default `20000`)
+
+### Threads
+- `THREAD_LOCKS` (default `20`)
+- `THREAD_YIELDS` (default empty → sysbench default)
+
+### Memory
+- `MEMORY_OPER` (default `write`) – `read|write`
+- `MEMORY_ACCESS_MODE` (default `rnd`) – `seq|rnd`
+- `MEMORY_BLOCK_SIZE` (default `1M`)
+- `MEMORY_TOTAL_SIZE` (default `100G`)
+
+### File I/O
+File I/O uses sysbench `fileio` with a **prepare → run → cleanup** flow.
+
+Recommended knobs (names match the suite params):
+- `FILEIO_DIR` – directory where files are created (recommended: **tmpfs** like `/tmp`)
+- `FILE_TOTAL_SIZE` – total size per file, e.g. `1G`
+- `FILE_BLOCK_SIZE` – e.g. `4K`, `1M`
+- `FILE_NUM` – number of files, e.g. `1`, `4`
+- `FILE_IO_MODE` – e.g. `sync` (stable), `async`
+- `FILE_FSYNC_FREQ` – optional, e.g. `0` (disable) or `1` (every op)
+
+> Your `run.sh` may expose additional `FILE_*` knobs; keep them documented here as you add them.
+
+---
+
+## Safety warning for File I/O
+
+`sysbench fileio` **creates files and performs reads/writes** in `FILEIO_DIR`.
+
+To avoid accidental wear / regression noise:
+- Prefer `FILEIO_DIR=/tmp/...` (tmpfs) for CI gating.  
+- If you must test real storage, point to a **dedicated mount** (not `/`), and keep sizes small.
+- Do **NOT** point fileio at your root filesystem (`/`) unless you fully understand the impact.
+
+---
+
+## Outputs
+
+In `OUT_DIR/` you will see:
+
+- `*_iterN.log` — full sysbench stdout/stderr per test case + iteration
+- `*.values` — one KPI value per iteration (for averaging)
+- `sysbench_summary.txt` — averages + gating summary
+- Optional CSV (only when `CSV_FILE` is set)
+
+Result file:
+- `./Sysbench_Performance.res` — `PASS` / `FAIL` / `SKIP`
+
+---
+
+## Examples (using our `run.sh`)
+
+All examples are run from:
+
+```
+cd Runner/suites/Performance/Sysbench_Performance/
+```
+
+### 1) Full suite (CPU + memory + threads + mutex + fileio)
+```
+./run.sh
+```
+
+### 2) 3 iterations, pin to big cores, tighter gating (95% baseline)
+(Uses `./sysbench_baseline.conf` automatically if present.)
+```
+./run.sh --iterations 3 --threads-list "4" --taskset-cpu-list "6-7" --delta 0.05
+```
+
+### 3) Explicit baseline file path
+```
+./run.sh --baseline ./sysbench_baseline.conf --delta 0.05
+```
+
+### 4) Enable CSV (NOT default)
+```
+./run.sh --csv ./sysbench_out/sysbench.csv
+```
+
+### 5) CPU time focus (reduce noise)
+You can’t “run only CPU” without editing the script, but you can reduce runtime:
+```
+./run.sh --iterations 1 --time 15 --cpu-max-prime 20000 --threads-list "4"
+```
+Then read:
+- console `ITER_SUMMARY ... cpu_time_sec=...`
+- `OUT_DIR/cpu_t4.values`
+- CSV rows (if enabled)
+
+### 6) Memory bandwidth focus (use seq vs rnd)
+```
+MEMORY_ACCESS_MODE=seq ./run.sh --iterations 1 --threads-list "4"
+MEMORY_ACCESS_MODE=rnd ./run.sh --iterations 1 --threads-list "4"
+```
+
+### 7) File I/O on tmpfs (recommended for CI)
+```
+FILEIO_DIR=/tmp/sysbench_fileio ./run.sh --iterations 1 --threads-list "4"
+```
+
+### 8) File I/O heavier workload (be careful)
+```
+FILEIO_DIR=/tmp/sysbench_fileio FILE_TOTAL_SIZE=1G FILE_BLOCK_SIZE=4K FILE_NUM=4 ./run.sh --iterations 1 --threads-list "4"
+```
+
+### 9) Verifying seq/rnd IO KPIs
+After the run, check:
+- `sysbench_summary.txt` for:
+  - `fileio_seqwr_mbps`, `fileio_seqrd_mbps`, `fileio_rndwr_mbps`, `fileio_rndrd_mbps`
+- per-iteration logs:
+  - `fileio_*_iter*.log`
+
+---
+
+## LAVA YAML integration
+
+Your LAVA test definition should:
+- `cd Runner/suites/Performance/Sysbench_Performance/`
+- run `./run.sh`
+- publish `Sysbench_Performance.res` via `send-to-lava.sh`
+
+Keep `CSV_FILE` empty by default (to avoid extra artifacts) and enable only when you explicitly need CSV.
+
+---
+
+## License
+
+Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.  
+SPDX-License-Identifier: BSD-3-Clause-Clear

Runner/suites/Performance/Sysbench_Performance/Sysbench_Performance.yaml

@@ -0,0 +1,68 @@
+metadata:
+  name: sysbench-performance
+  format: "Lava-Test Test Definition 1.0"
+  description: "Sysbench performance suite (CPU / memory / threads / mutex + fileio seqwr/seqrd/rndwr) with per-iteration logs + summary averages + optional baseline gating."
+  os:
+    - linux
+  scope:
+    - performance
+    - functional
+
+params:
+  OUT_DIR: "./logs_Sysbench_Performance"
+
+  # Iteration controls
+  ITERATIONS: "3"
+  TIME: "30"
+  RAND_SEED: "1234"
+  THREADS_LIST: "4" # space-separated list, e.g. "1 4 8 16"
+
+  # CPU
+  CPU_MAX_PRIME: "20000"
+
+  # Threads test
+  THREAD_LOCKS: "20"
+  THREAD_YIELDS: "" # optional override; empty uses sysbench default
+
+  # Memory test
+  MEMORY_OPER: "write" # read|write
+  MEMORY_ACCESS_MODE: "rnd" # seq|rnd
+  MEMORY_BLOCK_SIZE: "1M"
+  MEMORY_TOTAL_SIZE: "100G" # keep explicit for stability
+
+  # Mutex test
+  MUTEX_NUM: ""
+  MUTEX_LOCKS: ""
+  MUTEX_LOOPS: ""
+
+  # Core pinning (optional)
+  TASKSET_CPU_LIST: ""
+
+  # Baselines + thresholds (optional gating; empty => report-only)
+  # If empty, run.sh may auto-pick ./sysbench_baseline.conf from the test folder (if present)
+  BASELINE_FILE: ""
+  ALLOWED_DEVIATION: "0.10" # 0.05..0.10 typical
+
+  # Optional CSV append (NOT default). If empty, no CSV is written.
+  CSV_FILE: ""
+
+  VERBOSE: "0" # 1|0
+
+  # ---------------- FileIO (storage-backed) ----------------
+  # IMPORTANT: use ext4-backed path (NOT /tmp or other tmpfs paths)
+  FILEIO_DIR: "/var/tmp/sysbench_fileio"
+
+  # Recommended stable fileio parameters
+  FILEIO_TOTAL_SIZE: "1G"
+  FILEIO_NUM_FILES: "64"
+  FILEIO_BLOCK_SIZE: "4K"
+  FILEIO_IO_MODE: "sync" # sync|async
+  FILEIO_MODE_LIST: "seqwr seqrd rndwr" # fileio modes to run
+  FILEIO_EXTRA_FLAGS: "" # optional additional sysbench fileio flags
+
+run:
+  steps:
+    - REPO_PATH=$PWD
+    - cd Runner/suites/Performance/Sysbench_Performance/
+    - ./run.sh || true
+    - $REPO_PATH/Runner/utils/send-to-lava.sh Sysbench_Performance.res || true