perf: replace O(n) TCP RX buffer scan with SIMD memchr + carry buffer (Sprint 3)

This commit fixes the most significant hot-path bottleneck in the
client: the tcp_client_rx_loop was scanning up to 256KB byte-by-byte
on every read() call looking for interleaved 12-byte status messages.

Changes:
- client.rs (tcp_client_rx_loop): Replace the O(n) for-loop scan
  with a three-stage approach:

  1. Split-message check: An 11-byte carry buffer stores trailing
     bytes from the previous read. We check every possible alignment
     where a status message (0x07 + cpu_byte) could span the carry
     and the start of the current buffer. This fixes a latent bug
     where the old code would miss status messages split across TCP
     read boundaries.

  2. Fast scan: memchr::memchr (AVX2/NEON SIMD) finds 0x07 bytes
     in the 256KB buffer. On all-zero data packets this exits in
     ~4096 SIMD-width operations instead of 262,144 byte compares.
     ~64x faster scan path.

  3. Carry save: Save up to 11 trailing bytes for the next read.

- client.rs (unit tests): Add scan_status_message() helper and
  five unit tests covering:
  - Status message fully within buffer
  - Status message split across reads (5+7 bytes)
  - Status message split at boundary (1+11 bytes)
  - All-zero buffer (no false positive)
  - Short buffer (no panic)

- Cargo.toml / Cargo.lock: Add memchr as an explicit dependency.

Verified against live MikroTik RouterOS (TCP both + receive modes
with EC-SRP5 auth). Status messages detected correctly. No wire
protocol changes — 100% MikroTik compatible.

Cargo.lock

@@ -11,6 +11,15 @@ dependencies = [
  "memchr",
 ]
 
+[[package]]
+name = "android_system_properties"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "819e7219dbd41043ac279b19830f2efc897156490d7fd6ea916720117ee66311"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "anstream"
 version = "1.0.0"
@@ -235,10 +244,12 @@ dependencies = [
  "askama",
  "axum",
  "bytes",
+ "chrono",
  "clap",
  "hostname",
  "ldap3",
  "md-5",
+ "memchr",
  "num-bigint",
  "num-integer",
  "num-traits",
@@ -283,6 +294,19 @@ version = "1.0.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9330f8b2ff13f34540b44e946ef35111825727b38d33286ef986142615121801"
 
+[[package]]
+name = "chrono"
+version = "0.4.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c673075a2e0e5f4a1dde27ce9dee1ea4558c7ffe648f576438a20ca1d2acc4b0"
+dependencies = [
+ "iana-time-zone",
+ "js-sys",
+ "num-traits",
+ "wasm-bindgen",
+ "windows-link",
+]
+
 [[package]]
 name = "clap"
 version = "4.6.0"
@@ -748,6 +772,30 @@ dependencies = [
  "tower-service",
 ]
 
+[[package]]
+name = "iana-time-zone"
+version = "0.1.65"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e31bc9ad994ba00e440a8aa5c9ef0ec67d5cb5e5cb0cc7f8b744a35b389cc470"
+dependencies = [
+ "android_system_properties",
+ "core-foundation-sys",
+ "iana-time-zone-haiku",
+ "js-sys",
+ "log",
+ "wasm-bindgen",
+ "windows-core",
+]
+
+[[package]]
+name = "iana-time-zone-haiku"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f31827a206f56af32e590ba56d5d2d085f558508192593743f16b2306495269f"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "icu_collections"
 version = "2.1.1"
@@ -1997,12 +2045,65 @@ dependencies = [
  "semver",
 ]
 
+[[package]]
+name = "windows-core"
+version = "0.62.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b8e83a14d34d0623b51dce9581199302a221863196a1dde71a7663a4c2be9deb"
+dependencies = [
+ "windows-implement",
+ "windows-interface",
+ "windows-link",
+ "windows-result",
+ "windows-strings",
+]
+
+[[package]]
+name = "windows-implement"
+version = "0.60.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "053e2e040ab57b9dc951b72c264860db7eb3b0200ba345b4e4c3b14f67855ddf"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "windows-interface"
+version = "0.59.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f316c4a2570ba26bbec722032c4099d8c8bc095efccdc15688708623367e358"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
 [[package]]
 name = "windows-link"
 version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f0805222e57f7521d6a62e36fa9163bc891acd422f971defe97d64e70d0a4fe5"
 
+[[package]]
+name = "windows-result"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7781fa89eaf60850ac3d2da7af8e5242a5ea78d1a11c49bf2910bb5a73853eb5"
+dependencies = [
+ "windows-link",
+]
+
+[[package]]
+name = "windows-strings"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7837d08f69c77cf6b07689544538e017c1bfcf57e34b4c0ff58e6c2cd3b37091"
+dependencies = [
+ "windows-link",
+]
+
 [[package]]
 name = "windows-sys"
 version = "0.52.0"

PERFORMANCE_AUDIT.md

@@ -0,0 +1,302 @@
+# btest-rs Performance Audit
+
+**Date:** 2026-04-30  
+**Scope:** Full codebase (`src/`, `tests/`, `Cargo.toml`)  
+**Methodology:** Static code analysis, hot-path tracing, lock/contention review, algorithmic complexity analysis  
+
+---
+
+## Executive Summary
+
+The codebase is generally well-structured for a network I/O tool, with good use of atomics in the per-packet hot path and zero-allocation protocol serialization. However, **three critical bottlenecks** significantly limit throughput and scalability:
+
+1. **O(n) buffer scan on every TCP read** in the client RX loop (up to 256KB scanned per `read()` call)
+2. **Expensive EC curve reconstruction on every authentication** (heavy `BigUint` modular arithmetic)
+3. **Single SQLite connection mutex** serializing all DB operations in `server_pro`
+
+Additionally, there is **no benchmark or profiling infrastructure** in the project, making it impossible to measure improvements or catch regressions.
+
+---
+
+## Severity Legend
+
+| Icon | Severity | Impact |
+|------|----------|--------|
+| 🔴 | **Critical** | Direct throughput/latency hit in hot path; fix immediately |
+| 🟠 | **High** | Significant overhead under load; fix in next sprint |
+| 🟡 | **Medium** | Noticeable at scale or under specific conditions |
+| 🟢 | **Low** | Cosmetic / easy wins; batch with other work |
+
+---
+
+## 🔴 Critical Bottlenecks
+
+### 1. O(n) Linear Buffer Scan in `tcp_client_rx_loop` (`src/client.rs:210-216`)
+
+**Problem:** On every TCP `read()` call (up to 256KB), the client performs a byte-by-byte scan looking for interleaved 12-byte status messages:
+
+```rust
+for i in 0..=(n - STATUS_MSG_SIZE) {
+    if buf[i] == STATUS_MSG_TYPE && buf[i + 1] >= 0x80 {
+        // ...
+    }
+}
+```
+
+Since data packets are all zeros and status bytes are extremely rare, this **almost always scans the entire 256KB buffer** uselessly. At high bandwidth (many reads/sec), this wastes massive CPU cycles and pollutes cache lines.
+
+**Impact:** CPU-bound slowdown on the client RX side during bidirectional TCP tests. The compiler *may* auto-vectorize the simple loop, but it still processes ~256K bytes per read.
+
+**Fix Options (pick one):**
+- **Best:** Use `memchr` (crate or `std::slice::memchr` on nightly) to find `0x07` bytes. On all-zero buffers this exits after a few SIMD-width checks.
+- **Alternative:** Since status messages are injected at `write_all` boundaries and data is all zeros, maintain a small 12-byte sliding ring buffer across reads. Process the stream with a tiny state machine instead of scanning the whole buffer.
+- **Alternative:** Track read bytes modulo expected packet size. Status messages are injected between full packets, so they will appear at predictable offsets *if* the client knows the server's `effective_size`. This requires protocol coordination.
+
+---
+
+### 2. `WCurve::new()` Recomputes Generator on Every Auth (`src/ecsrp5.rs:363,499`)
+
+**Problem:** Every EC-SRP5 authentication (client and server) calls `WCurve::new()`, which performs `lift_x(9)` → `prime_mod_sqrt()` — heavy `BigUint` modular arithmetic to derive the curve generator point.
+
+```rust
+pub async fn client_authenticate<S: ...>(stream: &mut S, username: &str, password: &str) -> Result<()> {
+    let w = WCurve::new();  // <-- expensive, same result every time
+    // ...
+}
+```
+
+The curve constants (`P`, `CURVE_ORDER`, `WEIERSTRASS_A`) are already cached as `LazyLock` statics, but the generator point is not.
+
+**Impact:** Auth latency spikes, especially on the server under many concurrent connections. Each auth does redundant `BigUint` allocations and modular square roots.
+
+**Fix:** Cache `WCurve` (or at least the generator point) in a global `LazyLock`:
+
+```rust
+static WCURVE: std::sync::LazyLock<WCurve> = std::sync::LazyLock::new(WCurve::new);
+```
+
+Then use `&*WCURVE` in both `client_authenticate` and `server_authenticate`.
+
+---
+
+### 3. Single SQLite Mutex Serializes All DB Operations (`src/server_pro/user_db.rs:15-18`)
+
+**Problem:** The entire `server_pro` database layer uses a single shared `Connection` behind a `std::sync::Mutex`:
+
+```rust
+pub struct UserDb {
+    conn: Arc<Mutex<Connection>>,
+}
+```
+
+While SQLite WAL mode is enabled (allowing readers to proceed during writes), **the Rust mutex still serializes all access to the connection object**. Under concurrent load with many tests starting/finishing, this becomes the primary bottleneck.
+
+**Critical sub-issue:** `QuotaManager::remaining_budget()` (`src/server_pro/quota.rs:387`) performs **up to 15 separate SQLite queries** in sequence, locking the mutex 15+ times per pre-test check.
+
+**Impact:** Connection setup/teardown latency increases linearly with concurrency. Quota checks and usage recording block each other.
+
+**Fix Options:**
+- **Connection pooling:** Use `r2d2_sqlite` or `deadpool-sqlite` to maintain a small pool of connections (SQLite handles this well in WAL mode).
+- **Separate read/write paths:** Open a read-only connection for quota checks (`remaining_budget`) and a dedicated write connection for usage recording. SQLite WAL allows this.
+- **Batch quota checks:** Cache quota results for a few seconds per user/IP to avoid redundant queries.
+- **Channel-based writer:** Use a single dedicated DB writer task with an `mpsc` channel so only one task ever touches the connection, eliminating lock contention entirely.
+
+---
+
+## 🟠 High Severity Issues
+
+### 4. 100ms Busy-Poll Wait in Multi-Connection TCP (`src/server.rs:313-332`)
+
+**Problem:** When waiting for secondary TCP connections to join a multi-connection session, the primary connection busy-polls the session map every 100ms:
+
+```rust
+loop {
+    let count = { let map = sessions.lock().await; ... };
+    if count + 1 >= conn_count as usize { break; }
+    tokio::time::sleep(Duration::from_millis(100)).await;
+}
+```
+
+This adds up to **100ms of unnecessary latency** to every multi-connection test startup. It also hammers the async mutex needlessly.
+
+**Fix:** Replace with `tokio::sync::Notify`. When a secondary connection registers itself, it calls `notify_one()`. The primary waits on `notified().await` with a timeout, waking instantly when ready.
+
+---
+
+### 5. FreeBSD CPU Sampling Spawns Process Every Second (`src/cpu.rs:142`)
+
+**Problem:** On FreeBSD, `get_cpu_times()` spawns `sysctl -n kern.cp_time` via `std::process::Command` every second:
+
+```rust
+fn get_cpu_times() -> (u64, u64) {
+    if let Ok(output) = std::process::Command::new("sysctl")
+        .arg("-n").arg("kern.cp_time").output()
+    { ... }
+}
+```
+
+`fork()` + `exec()` is extremely expensive relative to the work being done (reading 5 integers).
+
+**Fix:** Use `libc::sysctl()` via FFI, matching the macOS implementation style. Cache the `mib` array and call the syscall directly.
+
+---
+
+### 6. Per-Call Timer Registration in UDP RX Loops (`src/client.rs:393`, `src/server.rs:925`)
+
+**Problem:** Both UDP RX loops create a new `tokio::time::timeout` timer on **every single `recv`/`recv_from` call**:
+
+```rust
+match tokio::time::timeout(Duration::from_secs(5), socket.recv(&mut buf)).await
+```
+
+At high packet rates (e.g., 100K pps), registering and canceling timers on the Tokio timer wheel adds measurable overhead.
+
+**Fix:** Use `tokio::select!` with a long-lived `tokio::time::sleep` future that is reset, or use the socket's built-in SO_RCVTIMEO if available via `socket2`. Alternatively, since UDP is connectionless, consider whether a 5-second timeout is needed on every call or if the outer test duration timer is sufficient.
+
+---
+
+## 🟡 Medium Severity Issues
+
+### 7. String Error Matching with Allocation (`src/server_pro/enforcer.rs:157-161`)
+
+```rust
+match format!("{}", e).as_str() {
+    s if s.contains("daily") => ...
+}
+```
+
+`format!("{}", e)` allocates a `String` from the error just to do substring matching. Use `e.to_string().contains(...)` or match on error types directly if possible.
+
+---
+
+### 8. `ip.to_string()` Called Repeatedly in Quota Checks (`src/server_pro/quota.rs:389`)
+
+```rust
+let ip_str = ip.to_string();
+// ... used in 6+ DB calls
+```
+
+This allocates a `String` on every quota check. Accept `&str` or `IpAddr` directly in DB methods, or cache the string.
+
+---
+
+### 9. `chrono_date_today()` Recomputes Calendar from Epoch (`src/server_pro/user_db.rs:617-638`)
+
+A hand-rolled date calculation loops through years from 1970 and months every time it's called (which is before almost every DB write). The `chrono` crate is already used indirectly by `rusqlite`; add it as a direct dependency and replace with `chrono::Local::now().format("%Y-%m-%d")`.
+
+---
+
+## 🟢 Low Severity / Easy Wins
+
+### 10. CSV File Reopened on Every Write (`src/csv_output.rs:77`)
+
+```rust
+if let Ok(mut f) = OpenOptions::new().append(true).open(path) {
+    let _ = writeln!(f, "{}", row);
+}
+```
+
+Called once per test, not per-packet, but still suboptimal. Consider keeping a lazily-initialized `Mutex<Option<File>>` or using `std::fs::OpenOptions` once at init and storing the handle.
+
+---
+
+### 11. Global Syslog Mutex Held During I/O (`src/syslog_logger.rs`)
+
+```rust
+static SYSLOG: Mutex<Option<SyslogSender>> = Mutex::new(None);
+```
+
+The global `std::sync::Mutex` is held while formatting the timestamp (expensive manual calendar math) and sending UDP. Switch to `parking_lot::Mutex` (faster) or `tokio::sync::Mutex` if async, and format the message outside the lock. Better yet, use `tracing`'s built-in syslog integration or a structured appender.
+
+---
+
+### 12. `hash_password()` Uses `format!` + `format!` in Hex Loop (`src/server_pro/user_db.rs:612-614`)
+
+```rust
+hasher.update(format!("{}:{}", username, password).as_bytes());
+result.iter().map(|b| format!("{:02x}", b)).collect()  // N allocs for N bytes
+```
+
+The hex encoding allocates one `String` per byte. Use a small fixed buffer or `hex` crate (already used elsewhere in `ecsrp5.rs`).
+
+---
+
+### 13. Redundant `Instant::now()` Calls in TX Loop (`src/server.rs:593,606`)
+
+```rust
+if send_status && Instant::now() >= next_status {
+    // ...
+    next_status = Instant::now() + Duration::from_secs(1);
+}
+```
+
+Two monotonic clock reads per loop iteration. Cache `let now = Instant::now();` at the top of the loop.
+
+---
+
+## Architecture Observations
+
+### What the Code Does Well
+- **Zero-allocation protocol layer:** `serialize()` returns fixed-size stack arrays (`[u8; 12]`, `[u8; 16]`). Excellent.
+- **Atomic bandwidth tracking:** `BandwidthState` uses `AtomicU64` with `Relaxed` ordering in the per-packet path. No locks in the data plane.
+- **Buffer reuse:** TX/RX loops allocate `vec![0u8; ...]` once before the loop. Good.
+- **Aggressive release profile:** `lto = true`, `codegen-units = 1`, `opt-level = 3`.
+
+### Async Runtime Usage
+- `tokio` with `full` features is used. For a primarily I/O-bound tool, this is appropriate.
+- `tokio::task::yield_now().await` is used in unlimited-rate mode to prevent starving the runtime. This is correct but consider whether `tokio::task::spawn_blocking` or dedicated CPU pinning is needed for the EC-SRP5 math (which is CPU-bound and currently runs on the async runtime during auth).
+
+### Memory Safety
+- Several `unwrap()`/`expect()` calls in setup paths (socket binding, address parsing). These are acceptable for config errors but should use `?` propagation where possible to allow graceful degradation.
+
+---
+
+## Missing Performance Infrastructure
+
+| Infrastructure | Status | Recommendation |
+|----------------|--------|----------------|
+| **Benchmarks** | ❌ None | Add `criterion` + `benches/` for `BandwidthState`, protocol ser/de, and EC-SRP5 auth |
+| **Profiling hooks** | ❌ None | Add optional `pprof` or `dhat` dev-deps for heap profiling |
+| **Throughput regression tests** | ⚠️ Partial | Integration tests assert `tx > 0` and `rx > 0` but don't measure sustained throughput |
+| **Load tests** | ❌ None | Add a `benches/load_test.rs` that spawns 100+ concurrent tests against a local server |
+| **CI performance gates** | ❌ None | Consider a benchmark action that fails on >5% regression |
+
+---
+
+## Priority Action Plan
+
+### Phase 1: Hot-Path Fixes (1-2 days)
+1. Replace buffer scan with `memchr` or ring-buffer approach in `tcp_client_rx_loop`
+2. Cache `WCurve` in a global `LazyLock`
+3. Replace 100ms poll with `tokio::sync::Notify` in multi-conn wait
+
+### Phase 2: Scalability (2-3 days)
+4. Add SQLite connection pooling or channel-based writer in `server_pro`
+5. Cache `remaining_budget()` results for 5-10 seconds
+6. Fix FreeBSD CPU sampling to use `libc::sysctl` FFI
+
+### Phase 3: Polish & Tooling (1-2 days)
+7. Replace manual date arithmetic with `chrono`
+8. Add `criterion` benchmarks for auth and bandwidth state
+9. Fix low-severity allocation issues (CSV, syslog, hex encoding)
+
+---
+
+## Appendix: File-by-File Quick Reference
+
+| File | Lines | Hot Path? | Key Concern |
+|------|-------|-----------|-------------|
+| `src/client.rs` | 531 | ✅ Yes | O(n) 256KB scan per TCP read |
+| `src/server.rs` | 1094 | ✅ Yes | 100ms poll wait, status injection timing |
+| `src/ecsrp5.rs` | 660 | ✅ Yes (auth) | `WCurve::new()` recomputed per auth |
+| `src/bandwidth.rs` | 263 | ✅ Yes (atomics) | Well-designed; no issues |
+| `src/protocol.rs` | 214 | ✅ Yes (ser/de) | Zero-allocation; excellent |
+| `src/cpu.rs` | 215 | ⚠️ Periodic | FreeBSD `fork+exec` every second |
+| `src/server_pro/quota.rs` | 470 | ⚠️ Periodic | 15 DB queries per budget check |
+| `src/server_pro/user_db.rs` | 641 | ⚠️ All DB ops | Single mutex serializes everything |
+| `src/server_pro/server_loop.rs` | 449 | ✅ Yes | DB auth locks during connection setup |
+| `src/server_pro/enforcer.rs` | 411 | ⚠️ Periodic | String error matching allocates |
+| `src/csv_output.rs` | 86 | ❌ No | File reopen per write |
+| `src/syslog_logger.rs` | 154 | ❌ No | Global mutex + manual calendar math |
+| `src/auth.rs` | 164 | ⚠️ Auth only | Minor; double MD5 per auth |
+| `src/main.rs` | 243 | ❌ No | Entry point only |

PERFORMANCE_PRDS.md

@@ -0,0 +1,634 @@
+# Performance Improvement PRDs
+
+**Project:** btest-rs  
+**Constraint:** 100% MikroTik BTest protocol compatibility — no wire-format or behavioral changes visible to MikroTik devices  
+**Date:** 2026-04-30  
+
+---
+
+## How to Read This Document
+
+Each PRD is sorted by **recommended execution order**, which balances:
+- **Effort** (development + review + test time)
+- **Risk** (probability of regression or compatibility break)
+- **Performance Effect** (measured or estimated throughput/latency improvement)
+- **MikroTik Compatibility Risk** (whether the change could affect interoperability)
+
+**Sorting rationale:** Execute *quick wins* first to build velocity and reduce risk surface, then tackle *high-impact* items with full attention.
+
+---
+
+## Summary Matrix
+
+| # | PRD | Effort | Risk | Perf Impact | MikroTik Risk | Tier |
+|---|-----|--------|------|-------------|---------------|------|
+| 1 | WCurve Global Cache | 30 min | None | Medium | None | Quick Win |
+| 2 | Redundant `Instant::now()` | 15 min | None | Low | None | Quick Win |
+| 3 | `hash_password` Hex Fix | 30 min | None | Low | None | Quick Win |
+| 4 | CSV File Handle Cache | 30 min | None | Low | None | Quick Win |
+| 5 | Error String Matching | 30 min | None | Low | None | Quick Win |
+| 6 | `chrono_date_today` Replace | 1 hr | Low | Low | None | Quick Win |
+| 7 | Syslog Mutex + Timestamp | 1 hr | Low | Low | None | Quick Win |
+| 8 | `ip.to_string()` Cache | 1 hr | Low | Low | None | Quick Win |
+| 9 | FreeBSD CPU FFI | 3 hrs | Medium | Medium | None | Platform Fix |
+| 10 | Multi-Conn Notify Wake | 2 hrs | Medium | Medium | None | Latency Fix |
+| 11 | UDP Timer Reuse | 2 hrs | Medium | Medium | None | Throughput Fix |
+| 12 | TCP RX Scan Optimization | 4 hrs | Medium | **High** | Low | Hot Path Fix |
+| 13 | SQLite Connection Pool | 1–2 days | High | **High** | None | Scalability Fix |
+
+---
+
+## Tier 1: Quick Wins (Do These First)
+
+---
+
+### PRD-001: Cache `WCurve` in Global `LazyLock`
+
+**Background:**
+`WCurve::new()` is called on every EC-SRP5 authentication (client and server). It recomputes the Weierstrass curve generator point via `lift_x(9)` → `prime_mod_sqrt()`, which performs heavy `BigUint` modular arithmetic. The result is deterministic and immutable.
+
+**MikroTik Compatibility:**
+- **100% safe.** This is pure internal mathematics. The wire bytes, auth handshake order, and hash outputs are identical. No protocol-visible change.
+
+**Objective:**
+Eliminate redundant `BigUint` modular square root computation per authentication.
+
+**Design:**
+```rust
+// src/ecsrp5.rs
+static WCURVE: std::sync::LazyLock<WCurve> = std::sync::LazyLock::new(WCurve::new);
+```
+
+Replace all call sites:
+- `src/ecsrp5.rs:363` (`client_authenticate`)
+- `src/ecsrp5.rs:499` (`server_authenticate`)
+
+Change `let w = WCurve::new();` to `let w = &*WCURVE;`. Update any `WCurve` methods that take `self` to take `&self` if they don't already.
+
+**Acceptance Criteria:**
+- [ ] `ecsrp5_test.rs` passes unchanged.
+- [ ] `full_integration_test.rs` EC-SRP5 tests pass unchanged.
+- [ ] `WCurve::new()` is called exactly once per process lifetime.
+- [ ] No change to serialized auth bytes on the wire.
+
+**Effort:** 30 min  
+**Risk:** None — stateless deterministic cache  
+**Performance Impact:** Medium — reduces per-auth CPU time by ~30-50% (estimated), especially noticeable under concurrent logins.
+
+---
+
+### PRD-002: Deduplicate `Instant::now()` in `tcp_tx_loop_inner`
+
+**Background:**
+The TCP TX loop calls `Instant::now()` twice per iteration (status check and interval scheduling). Monotonic clock reads are cheap but not free, and occur in the hottest loop in the system.
+
+**MikroTik Compatibility:**
+- **100% safe.** Timing granularity remains identical.
+
+**Objective:**
+Reduce syscalls in the per-packet hot path.
+
+**Design:**
+```rust
+let now = Instant::now();
+if send_status && now >= next_status { ... next_status = now + Duration::from_secs(1); }
+// ... reuse `now` for interval math
+```
+
+**Acceptance Criteria:**
+- [ ] TCP send/receive/both integration tests pass.
+- [ ] No behavioral change in status injection timing.
+
+**Effort:** 15 min  
+**Risk:** None  
+**Performance Impact:** Low — micro-optimization, but trivial.
+
+---
+
+### PRD-003: Fix `hash_password()` Hex Encoding Allocations
+
+**Background:**
+`user_db.rs:614` allocates one `String` per byte when hex-encoding a 32-byte SHA256 hash:
+```rust
+result.iter().map(|b| format!("{:02x}", b)).collect()
+```
+
+**MikroTik Compatibility:**
+- **100% safe.** Output string is identical.
+
+**Objective:**
+Replace N-allocation hex encoding with a single-allocation approach.
+
+**Design:**
+Use `hex` crate (already in dependency tree via `ecsrp5.rs` debug logging) or a small `[u8; 64]` buffer with `write!` to a `String::with_capacity(64)`.
+
+**Acceptance Criteria:**
+- [ ] Same hex string output for all inputs.
+- [ ] `pro` feature tests pass.
+
+**Effort:** 30 min  
+**Risk:** None  
+**Performance Impact:** Low — removes 32 allocations per password hash.
+
+---
+
+### PRD-004: Cache CSV File Handle
+
+**Background:**
+`csv_output::write_result()` re-opens the file via `OpenOptions::new().append(true).open(path)` on every call (once per test). Safe but wasteful.
+
+**MikroTik Compatibility:**
+- **100% safe.** No protocol involvement.
+
+**Objective:**
+Hold the file handle open for the process lifetime.
+
+**Design:**
+Change `static CSV_FILE: Mutex<Option<String>>` to `Mutex<Option<(String, std::fs::File)>>`, or open once during `init()` and store `Mutex<Option<File>>`.
+
+**Acceptance Criteria:**
+- [ ] CSV tests in `full_integration_test.rs` pass.
+- [ ] File is created with headers on `init()`.
+- [ ] Multiple `write_result` calls append correctly.
+
+**Effort:** 30 min  
+**Risk:** None  
+**Performance Impact:** Low — removes one `open()` syscall per test.
+
+---
+
+### PRD-005: Remove Allocating Error String Matching
+
+**Background:**
+`src/server_pro/enforcer.rs:157-161` does:
+```rust
+match format!("{}", e).as_str() {
+    s if s.contains("daily") => ...
+}
+```
+This allocates a `String` from the error just for substring matching.
+
+**MikroTik Compatibility:**
+- **100% safe.** Server-pro internal logic only.
+
+**Objective:**
+Match without allocation.
+
+**Design:**
+Use `e.to_string().contains("daily")` (still allocates but clearer) or, better, downcast the `rusqlite::Error` or match on structured error variants. If the error is `anyhow::Error`, use `.downcast_ref::<rusqlite::Error>()`.
+
+**Acceptance Criteria:**
+- [ ] Quota enforcement behavior unchanged.
+- [ ] Enforcer tests pass.
+
+**Effort:** 30 min  
+**Risk:** None  
+**Performance Impact:** Low — removes one allocation per enforcer tick.
+
+---
+
+### PRD-006: Replace `chrono_date_today()` with `chrono` Crate
+
+**Background:**
+`user_db.rs:617-638` contains a hand-rolled Gregorian calendar converter that loops from 1970 to compute today's date. Called before almost every DB write. The `chrono` crate is already pulled in transitively by `rusqlite`.
+
+**MikroTik Compatibility:**
+- **100% safe.** No protocol involvement.
+
+**Objective:**
+Replace 30 lines of error-prone manual date math with one `chrono` call.
+
+**Design:**
+Add `chrono = { version = "0.4", optional = true }` gated behind `pro` feature (or use the transitive dep directly). Replace `chrono_date_today()` with:
+```rust
+chrono::Local::now().format("%Y-%m-%d").to_string()
+```
+
+**Acceptance Criteria:**
+- [ ] `pro` feature compiles.
+- [ ] Date strings match format `YYYY-MM-DD`.
+- [ ] DB write tests pass.
+
+**Effort:** 1 hr  
+**Risk:** Low — adds explicit dep that already exists transitively  
+**Performance Impact:** Low — eliminates loop overhead, but called infrequently.
+
+---
+
+### PRD-007: Optimize Syslog Mutex and Timestamp Formatting
+
+**Background:**
+`syslog_logger.rs` holds a global `std::sync::Mutex` while formatting a timestamp (manual calendar math) and sending UDP. `std::sync::Mutex` is relatively slow, and the timestamp logic duplicates `chrono_date_today()` issues.
+
+**MikroTik Compatibility:**
+- **100% safe.** No protocol involvement.
+
+**Objective:**
+Reduce lock contention and allocation in logging path.
+
+**Design:**
+1. Use `parking_lot::Mutex` (faster, no poisoning) OR switch to `std::sync::Mutex` but clone the `SyslogSender` config outside the lock.
+2. Replace `bsd_timestamp()` with `chrono::Local::now().format("%b %e %H:%M:%S")`.
+3. Pre-allocate the `String` with `with_capacity(256)`.
+
+**Acceptance Criteria:**
+- [ ] Syslog output format remains RFC 3164 compliant.
+- [ ] `test_syslog_events` in `full_integration_test.rs` passes.
+
+**Effort:** 1 hr  
+**Risk:** Low  
+**Performance Impact:** Low — logging is not a hot path, but reduces global lock hold time.
+
+---
+
+### PRD-008: Cache `ip.to_string()` in Quota Checks
+
+**Background:**
+`quota.rs:389` calls `ip.to_string()` and then passes `&ip_str` to multiple DB methods, allocating a new `String` on every `remaining_budget()` call.
+
+**MikroTik Compatibility:**
+- **100% safe.** Server-pro internal logic.
+
+**Objective:**
+Eliminate redundant IP stringification.
+
+**Design:**
+Change DB methods to accept `&std::net::IpAddr` directly and stringify inside only when needed for SQL parameter binding (which `rusqlite` may already handle via `ToSql`). Alternatively, pass `ip_str: &str` from a single `to_string()` call and avoid re-stringifying in sub-calls.
+
+**Acceptance Criteria:**
+- [ ] Quota checks return identical results.
+- [ ] `pro` feature tests pass.
+
+**Effort:** 1 hr  
+**Risk:** Low  
+**Performance Impact:** Low — one allocation removed per quota check.
+
+---
+
+## Tier 2: Moderate Fixes (Platform & Latency)
+
+---
+
+### PRD-009: FreeBSD CPU Sampling via `libc::sysctl` FFI
+
+**Background:**
+On FreeBSD, `cpu.rs` spawns `sysctl -n kern.cp_time` as a child process every second. `fork()` + `exec()` is orders of magnitude slower than a direct syscall.
+
+**MikroTik Compatibility:**
+- **100% safe.** No protocol involvement. Platform-specific internal code.
+
+**Objective:**
+Replace subprocess with direct `sysctl(3)` syscall.
+
+**Design:**
+```rust
+#[cfg(target_os = "freebsd")]
+fn get_cpu_times() -> (u64, u64) {
+    let mut mib = [libc::CTL_KERN, libc::KERN_CP_TIME];
+    let mut cp_time: [libc::c_ulong; 5] = [0; 5];
+    let mut len = std::mem::size_of_val(&cp_time);
+    unsafe {
+        if libc::sysctl(
+            mib.as_mut_ptr(),
+            mib.len() as u32,
+            &mut cp_time as *mut _ as *mut libc::c_void,
+            &mut len,
+            std::ptr::null_mut(),
+            0,
+        ) == 0 {
+            let total = cp_time[0] + cp_time[1] + cp_time[2] + cp_time[3] + cp_time[4];
+            return (total as u64, cp_time[4] as u64);
+        }
+    }
+    (0, 0)
+}
+```
+
+**Acceptance Criteria:**
+- [ ] Compiles on FreeBSD.
+- [ ] Returns same values as previous `sysctl` command approach.
+- [ ] No child process spawned (verify with `ktrace` or `ps`).
+
+**Effort:** 3 hrs  
+**Risk:** Medium — requires FreeBSD test environment; FFI is unsafe  
+**Performance Impact:** Medium — eliminates 1 fork/exec per second on FreeBSD.
+
+---
+
+### PRD-010: Replace 100ms Poll with `tokio::sync::Notify`
+
+**Background:**
+In `server.rs:313-332`, the primary connection of a multi-connection TCP test busy-polls the session map every 100ms waiting for secondary connections to join.
+
+**MikroTik Compatibility:**
+- **100% safe.** This is internal server-side coordination. The wire behavior (waiting for connections, then starting the test) is unchanged. MikroTik clients will not observe a difference except potentially faster test startup.
+
+**Objective:**
+Eliminate polling latency and unnecessary mutex acquisitions.
+
+**Design:**
+1. Add a `tokio::sync::Notify` to `TcpSession`:
+```rust
+struct TcpSession {
+    peer_ip: IpAddr,
+    streams: Vec<OwnedTcpStream>,
+    expected: u8,
+    notify: tokio::sync::Notify,
+}
+```
+2. In the secondary connection handler, after pushing to `streams`, call `session.notify.notify_one()`.
+3. In the primary wait loop, replace the sleep loop with:
+```rust
+let count = { /* lock, get count, drop lock */ };
+if count + 1 >= conn_count { break; }
+
+// Wait for notification or 10s deadline
+let timeout = tokio::time::sleep(Duration::from_secs(10));
+tokio::pin!(timeout);
+
+loop {
+    tokio::select! {
+        _ = session.notify.notified() => {
+            let count = { /* lock, get count */ };
+            if count + 1 >= conn_count { break; }
+        }
+        _ = &mut timeout => { break; }
+    }
+}
+```
+
+**Acceptance Criteria:**
+- [ ] Multi-connection TCP tests pass.
+- [ ] Test startup latency is ≤ 1ms after last connection joins (was up to 100ms).
+- [ ] No deadlock under concurrent multi-connection tests.
+
+**Effort:** 2 hrs  
+**Risk:** Medium — concurrency change; must carefully manage lock/notify ordering to avoid races  
+**Performance Impact:** Medium — improves multi-conn test startup latency by up to 100ms per test.
+
+---
+
+### PRD-011: Reuse UDP RX Timer Instead of Per-Call Timeout
+
+**Background:**
+Both client and server UDP RX loops create a new `tokio::time::timeout` on every `recv`/`recv_from` call:
+```rust
+tokio::time::timeout(Duration::from_secs(5), socket.recv(&mut buf)).await
+```
+At high packet rates, this registers and cancels timers on Tokio's timer wheel constantly.
+
+**MikroTik Compatibility:**
+- **100% safe.** Internal async timing only. UDP packet processing is unchanged.
+
+**Objective:**
+Reduce timer wheel churn in high-rate UDP RX loops.
+
+**Design:**
+Option A — `tokio::select!` with a pinned sleep future:
+```rust
+let mut timeout = tokio::time::sleep(Duration::from_secs(5));
+tokio::pin!(timeout);
+
+loop {
+    tokio::select! {
+        biased; // prioritize recv
+        res = socket.recv(&mut buf) => { /* handle */ timeout.as_mut().reset(Instant::now() + Duration::from_secs(5)); }
+        _ = &mut timeout => { tracing::debug!("UDP RX timeout"); }
+    }
+}
+```
+
+Option B — Use `socket2` to set `SO_RCVTIMEO` on the underlying socket, then use blocking/async recv without Tokio timeouts. This moves timeout handling into the kernel, which is even cheaper.
+
+**Recommendation:** Start with Option A (pure Tokio, no platform risk). Option B can be a follow-up.
+
+**Acceptance Criteria:**
+- [ ] UDP send/receive/both tests pass.
+- [ ] UDP RX still times out correctly when no packets arrive.
+- [ ] No change to packet parsing or sequence tracking.
+
+**Effort:** 2 hrs  
+**Risk:** Medium — changes timeout behavior; must ensure test abortion still works correctly  
+**Performance Impact:** Medium — reduces timer wheel registration overhead, noticeable at >50K pps.
+
+---
+
+## Tier 3: High Impact (Do These With Full Focus)
+
+---
+
+### PRD-012: Optimize TCP Client RX Status Message Scan
+
+**Background:**
+`tcp_client_rx_loop` (`client.rs:210-216`) scans up to 256KB byte-by-byte on every `read()` call looking for a 12-byte status marker (`0x07` + `0x80|cpu`). Since data is all zeros, this is almost always a full scan.
+
+**MikroTik Compatibility Consideration:**
+- **High confidence of safety.** The protocol is: MikroTik injects 12-byte status messages into the TCP stream. Our client must detect them. Changing *how* we detect them (faster scan) does not change:
+  - What bytes are sent on the wire
+  - What bytes we expect
+  - How we respond to status messages
+- **One edge case to handle:** TCP is a stream. A status message may be split across two `read()` calls. The current code does **not** handle this correctly (it scans each buffer independently). The optimized version *should* handle split messages to be strictly more correct than the current implementation.
+
+**Objective:**
+Replace O(n) byte-by-byte scan with SIMD-accelerated or state-machine-based detection, while correctly handling split messages.
+
+**Design — Recommended: Ring Buffer Approach**
+
+Since status messages are 12 bytes and all other bytes are zeros, maintain a 12-byte ring buffer across reads:
+
+```rust
+const STATUS_MSG_SIZE: usize = 12;
+
+async fn tcp_client_rx_loop(mut reader: OwnedReadHalf, state: Arc<BandwidthState>) {
+    let mut buf = vec![0u8; 256 * 1024];
+    let mut carry = [0u8; STATUS_MSG_SIZE - 1]; // up to 11 bytes from previous read
+    let mut carry_len = 0usize;
+
+    while state.running.load(Ordering::Relaxed) {
+        match reader.read(&mut buf).await {
+            Ok(0) | Err(_) => break,
+            Ok(n) => {
+                state.rx_bytes.fetch_add(n as u64, Ordering::Relaxed);
+
+                // Check if a status message spans the carry + start of buf
+                if carry_len > 0 {
+                    let needed = STATUS_MSG_SIZE - carry_len;
+                    if n >= needed {
+                        let mut candidate = [0u8; STATUS_MSG_SIZE];
+                        candidate[..carry_len].copy_from_slice(&carry[..carry_len]);
+                        candidate[carry_len..].copy_from_slice(&buf[..needed]);
+                        if candidate[0] == STATUS_MSG_TYPE && candidate[1] >= 0x80 {
+                            state.remote_cpu.store(candidate[1] & 0x7F, Ordering::Relaxed);
+                        }
+                    }
+                }
+
+                // Scan within buf for status messages
+                // Since data is zeros, use memchr to find 0x07 candidates
+                if n >= STATUS_MSG_SIZE {
+                    let search_end = n - STATUS_MSG_SIZE + 1;
+                    let mut offset = 0;
+                    while let Some(pos) = memchr::memchr(STATUS_MSG_TYPE, &buf[offset..search_end]) {
+                        let i = offset + pos;
+                        if buf[i + 1] >= 0x80 {
+                            state.remote_cpu.store(buf[i + 1] & 0x7F, Ordering::Relaxed);
+                            break;
+                        }
+                        offset = i + 1;
+                        if offset >= search_end { break; }
+                    }
+                }
+
+                // Save trailing bytes for next read
+                carry_len = (n).min(STATUS_MSG_SIZE - 1);
+                if n >= carry_len {
+                    carry[..carry_len].copy_from_slice(&buf[n - carry_len..n]);
+                }
+            }
+        }
+    }
+}
+```
+
+**Alternative: `memchr` crate only**
+If we determine split messages are extremely rare and the current behavior is "good enough," simply replace the `for` loop with:
+```rust
+if let Some(pos) = memchr::memchr(STATUS_MSG_TYPE, &buf[..n - STATUS_MSG_SIZE + 1]) {
+    if buf[pos + 1] >= 0x80 { /* ... */ }
+}
+```
+This is a 5-line change with massive speedup (SIMD scan). However, the ring buffer approach is strictly more correct and not much more complex.
+
+**Acceptance Criteria:**
+- [ ] TCP bidirectional tests pass.
+- [ ] Remote CPU reporting still works.
+- [ ] Status messages split across reads are correctly detected (unit test for this).
+- [ ] `memchr` crate added to deps (very lightweight).
+- [ ] No change to wire bytes or server behavior.
+
+**Effort:** 4 hrs  
+**Risk:** Medium — hot path change; must be carefully reviewed and tested  
+**Performance Impact:** **High** — eliminates 256KB byte scan per read. At 10K reads/sec, saves ~2.5GB of memory scanning per second.
+
+---
+
+### PRD-013: SQLite Connection Pool / Channel-Based Writer
+
+**Background:**
+`server_pro` uses a single `Arc<Mutex<Connection>>`. All quota checks, usage recordings, and auth lookups serialize through one lock. `remaining_budget()` issues 15 queries, locking 15+ times. This is the primary scalability bottleneck for the pro server.
+
+**MikroTik Compatibility:**
+- **100% safe.** Server-side infrastructure only. No protocol change.
+
+**Objective:**
+Enable concurrent quota checks and usage recording without mutex contention.
+
+**Design — Option A: Connection Pool (Recommended for reads)**
+Use `r2d2_sqlite` or `deadpool-sqlite`:
+1. Open a pool of ~4-8 connections to the same SQLite file (WAL mode supports this).
+2. Read-only operations (`remaining_budget`, `get_user`, `check_user`) borrow a connection from the pool.
+3. Write operations (`record_usage`, `record_session`) also borrow from the pool (WAL allows concurrent readers + one writer).
+
+**Design — Option B: Channel-Based Writer (Recommended for writes)**
+1. Keep one dedicated `Connection` owned by a single Tokio task.
+2. Expose an `mpsc::channel` where other tasks send write requests (`RecordUsage { user, tx, rx }`).
+3. The writer task batches or sequentially executes writes without any mutex.
+4. Reads use a separate read-only connection or pool.
+
+**Hybrid Recommendation:**
+- **Reads:** Small connection pool (4 connections) for quota checks and auth lookups.
+- **Writes:** Single dedicated async task with an `mpsc::unbounded_channel` for usage recording.
+- **Cache:** Add a 5-second TTL cache for `remaining_budget()` results per user+IP to avoid redundant DB hits during test setup.
+
+**Acceptance Criteria:**
+- [ ] `pro` feature compiles and all tests pass.
+- [ ] Concurrent test launches scale linearly up to at least 50 concurrent sessions.
+- [ ] Quota enforcement remains correct (no over-quota usage).
+- [ ] Session logging and interval recording remain accurate.
+- [ ] No SQLite "database is locked" errors under load.
+
+**Effort:** 1–2 days  
+**Risk:** High — touches every DB interaction in `server_pro`; potential for data races, quota leaks, or connection exhaustion  
+**Performance Impact:** **High** — enables horizontal scaling of concurrent tests; removes the primary pro server bottleneck.
+
+---
+
+## Execution Roadmap
+
+### Sprint 1: Quick Wins + Foundation (1 day)
+- [ ] PRD-001: WCurve cache
+- [ ] PRD-002: `Instant::now()` dedup
+- [ ] PRD-003: `hash_password` hex fix
+- [ ] PRD-004: CSV file handle cache
+- [ ] PRD-005: Error string matching
+- [ ] PRD-006: `chrono` date replacement
+- [ ] PRD-007: Syslog optimization
+- [ ] PRD-008: `ip.to_string()` cache
+
+**Deliverable:** Low-risk PR with 8 clean commits. Run full integration tests.
+
+### Sprint 2: Platform & Async Fixes (1 day)
+- [ ] PRD-009: FreeBSD CPU FFI
+- [ ] PRD-010: Multi-conn Notify wake
+- [ ] PRD-011: UDP timer reuse
+
+**Deliverable:** PR with platform + latency improvements.
+
+### Sprint 3: Hot Path Optimization (1–2 days)
+- [ ] PRD-012: TCP RX scan optimization
+- [ ] Add unit test for split status messages
+- [ ] Benchmark before/after with `criterion` (or manual throughput test)
+
+**Deliverable:** PR with benchmark numbers proving improvement.
+
+### Sprint 4: Scalability (2–3 days)
+- [ ] PRD-013: SQLite connection pool / channel writer
+- [ ] Load test: 50 concurrent tests, verify no DB lock contention
+- [ ] Add `remaining_budget` cache
+
+**Deliverable:** PR with load test results.
+
+---
+
+## Testing Requirements for All PRDs
+
+Since **no wire protocol changes** are made, the existing integration test suite is the primary validation tool. However, for PRD-012 and PRD-013, additional tests are required:
+
+### New Tests to Add
+
+1. **Split Status Message Unit Test (for PRD-012)**
+   ```rust
+   #[test]
+   fn test_status_message_split_across_reads() {
+       // Feed first 5 bytes, then remaining 7 bytes
+       // Assert CPU value is extracted correctly
+   }
+   ```
+
+2. **Concurrent Quota Load Test (for PRD-013)**
+   ```rust
+   #[tokio::test]
+   async fn test_concurrent_quota_checks() {
+       // Spawn 50 tasks doing remaining_budget() + record_usage()
+       // Assert no panics, no SQLite locked errors
+   }
+   ```
+
+3. **FreeBSD CPU Parity Test (for PRD-009)**
+   Manual verification on FreeBSD that FFI `sysctl` returns same values as command.
+
+---
+
+## Appendix: MikroTik Compatibility Checklist
+
+For every PRD, verify:
+- [ ] No change to `Command` or `StatusMessage` struct layouts or serialization
+- [ ] No change to MD5 challenge-response handshake order
+- [ ] No change to EC-SRP5 handshake order or byte values
+- [ ] No change to TCP packet sizes or UDP payload format
+- [ ] No change to status injection timing (1-second interval)
+- [ ] No change to NAT probe behavior
+- [ ] Client can still authenticate against stock RouterOS `btest` server
+- [ ] Server can still accept connections from stock RouterOS `btest` client
+
+All PRDs in this document satisfy the above checklist by construction.