Category 1: Cyclictest Parameter Analysis

1. Effect of -q (Quiet Mode)

The customer's original command did not include -q, so cyclictest printed live status updates to the terminal throughout the run. Test 260522_3 adds -q to suppress those updates. The intent is to determine whether the terminal I/O introduced by those updates contributes measurable latency jitter.

Several parameters differ between the original and 260522_3 beyond -q alone, preventing a clean attribution of any observed differences to -q specifically:

• Duration: The original used -D 0 (run indefinitely) and ran considerably longer than 6 hours. 260522_3 ran for exactly 6 hours (-D6h). A longer run increases the probability of capturing rare worst-case latency events.
• Histogram: The original did not include -h600, so no histogram was captured — only summary statistics are available.
• Thread intervals: Without an explicit -i flag and with -t 2 and -d 500 (default distance), cyclictest staggers threads by 500 µs. Thread 0 ran at I:1000 µs and Thread 1 at I:1500 µs. Test 260522_3 uses -i1000 for both threads.

Customer Original — OoB OP-TEE (no -q)

Thread	Interval	Count	Min	Avg	Max
Core 0	1000 µs	90,242,144	5 µs	15 µs	489 µs
Core 1	1500 µs	60,161,415	6 µs	27 µs	295 µs

Customer Original — OP-TEE PRNG (no -q)

Thread	Interval	Count	Min	Avg	Max
Core 0	1000 µs	75,969,951	4 µs	13 µs	372 µs
Core 1	1500 µs	50,646,620	7 µs	25 µs	252 µs

260522_3 — OoB OP-TEE (with -q)

Thread	Interval	Min	Avg	Max
Core 0	1000 µs	5 µs	10 µs	475 µs
Core 1	1000 µs	7 µs	25 µs	266 µs

2. Effect of -M (--refresh_on_max)

Comparison A: -i1000 — 260522_3 (no -M) vs 260521_1 (with -M)

Condition: Default FS; isolate CPUs; OoB OPTEE

Comparison B: -i200 — 260522_1 (no -M) vs 260520_2 (with -M)

Condition: Default FS; isolate CPUs; OoB OPTEE

3. Effect of Interrupt Interval: -i1000 vs -i200

A shorter interval (200 µs vs 1000 µs) causes the measurement thread to wake up 5× more frequently, collecting proportionally more samples over the same duration. This section tests whether the higher sampling frequency affects the observed latency distribution. All three comparisons share the same hardware caveat as the rest of this report: tests were run across two boards, so inter-test variability may reflect hardware differences in addition to the parameter under test.

Comparison A: No -M — 260522_3 (-i1000) vs 260522_1 (-i200)

Condition: Default FS; isolate CPUs; OoB OPTEE

Comparison B: With -M, OoB OP-TEE — 260521_1 (-i1000) vs 260520_2 (-i200)

Condition: Default FS; isolate CPUs; OoB OPTEE

Comparison C: With -M, OP-TEE PRNG — 260521_3 (-i1000) vs 260521_2 (-i200)

Condition: Default FS; isolate CPUs; OP-TEE PRNG

Consistency Assessment

The direction of max latency change when moving from -i1000 to -i200 is consistent across all three comparison pairs. In all three pairs, -i200 produced lower maximum and average latencies than -i1000. The reduction in average latency is notable — at -i1000 the thread sleeps longer between wakeups, which may allow CPU caches to go cold and increase the cost of resuming the cyclictest code path on wakeup.

4. Combined Effect: Customer Test vs Customer Test & -i200 -M

This section compares the "Customer test" column (-i1000, no -M) against the "Customer test & -i200 -M" column (-i200, with -M) for all available same-condition pairs. Because both -i and -M change simultaneously, the results here should be interpreted in light of the single-variable baselines established in Sections 2 and 3: -i200 consistently reduces max and average latency, while -M alone remains inconclusive due to hardware variability.

260522_3 (CT) vs 260520_2 (CT&-i200-M)

Condition: Default FS; isolate CPUs; OoB OPTEE

Core 0 max: 475 µs → 127 µs (-73%). Core 1 max: 266 µs → 113 µs (-58%).

260525_2 (CT) vs 260525_4 (CT&-i200-M)

Condition: Default FS; isolate CPUs; DDR mods; OoB OPTEE

Core 0 max: 226 µs → 126 µs (-44%). Core 1 max: 133 µs → 98 µs (-26%).

260523_2 (CT) vs 260523_1 (CT&-i200-M)

Condition: Default FS; isolate CPUs; DDR mods; OP-TEE PRNG

Core 0 max: 101 µs → 53 µs (-48%). Core 1 max: 135 µs → 70 µs (-48%).

260525_1 (CT) vs 260525_3 (CT&-i200-M)

Condition: Base FS; isolate CPUs; DDR mods; OoB OPTEE

Core 0 max: 240 µs → 234 µs (-2%). Core 1 max: 113 µs → 221 µs (+96%).

Observations

Three of the four pairs (Comparisons 1–3) show reduced max latency in the combined -i200 -M test, consistent with the -i200 effect seen in Section 3. The magnitude of reduction is larger than Section 3's -i200-only pairs in some cases, which may reflect board-to-board variability or a genuine additive benefit from -M.

Comparison 4 (260525_1 vs 260525_3, Base FS; isolate CPUs; DDR mods; OoB OPTEE) is the exception: Core 0 max is nearly unchanged (240 µs → 234 µs, -2%), and Core 1 max increases substantially (113 µs → 221 µs, +96%). This outlier warrants further investigation — particularly whether the Base FS or DDR modification conditions interact differently with the -i200 or -M changes.