Performance measurements › arm_1 › kem-32 · system: x86_1 · arm_1

kem-32 Quasi-Cyclic Twisted McEliece Key Encapsulation Mechanism — performance on AArch64 (system arm_1)

Independent measurement following the structure of the NICCS ARM self-assessment guide, §3.5 (1)–(7). Not a submitter self-assessment.

1. Basic information

Category: public-key algorithm; function: key encapsulation
Algorithm: Quasi-Cyclic Twisted McEliece Key Encapsulation Mechanism
Implementation versions measured: reference
Parameter sets: QCTM128, QCTM256, QCTM512
Security evaluation: kem-32 report
Measurement method: arm_1 method
Submission: NICCS page

2. Assessment environment

item	value
processor	Qualcomm Oryon (CPU 2, one core)
machine	ASUS Vivobook S 15
clock	fixed 2.71 GHz (governor performance, minimum = maximum), boost off, SMT none
memory	30562 MiB
OS / kernel	Ubuntu 26.04.1 LTS / 7.0.0-34-generic
compiler / build tool	gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0 / cmake version 4.2.3
campaign start / end (UTC)	2026-09-28T15:05:46 / 2026-09-29T08:43:52

3. Functional testing (KAT)

Each library was checked against the submitted KAT vectors (SHA-256 manifest kem-32/kat.sha256 in the harness) before timing.

instance	build flags	KAT
`QCTM128`	guide	PASS
`QCTM256`	guide	PASS
`QCTM512`	guide	PASS

4. Performance

instance	operation	mean cycles	mean time	ops/s	median trial	n (trials × iters)
`QCTM128`	keygen	55.69 G	20.7 s	0.0484	20.8 s	40 (5 × 8)
`QCTM128`	enc	13.43 M	4.98 ms	201	4.97 ms	850 (5 × 170)
`QCTM128`	dec	4.98 G	1.85 s	0.542	1.85 s	100 (5 × 20)
`QCTM256`	keygen	119.50 G	44.4 s	0.0225	44.7 s	5 (5 × 1)
`QCTM256`	enc	43.27 M	16.1 ms	62.3	16.1 ms	205 (5 × 41)
`QCTM256`	dec	29.85 G	11.1 s	0.0902	11.1 s	30 (5 × 6)
`QCTM512`	keygen	697.86 G	259 s	0.00386	259 s	1 (1 × 1)
`QCTM512`	enc	163.01 M	60.5 ms	16.5	60.3 ms	100 (5 × 20)
`QCTM512`	dec	201.48 G	74.8 s	0.0134	75.2 s	4 (4 × 1)

5. Resource consumption

Static memory is approximated by the library's loadable ELF segments; peak memory is the process high-water mark (VmHWM) including the benchmark driver and libc, so both are upper-bound proxies rather than isolated algorithm memory.

instance	operation	static: ELF image (bytes)	baseline RSS	peak RSS
`QCTM128`	keygen	92252	3612 KiB	148700 KiB
`QCTM128`	enc	92252	11392 KiB	148560 KiB
`QCTM128`	dec	92252	14620 KiB	148848 KiB
`QCTM256`	keygen	90508	5352 KiB	474076 KiB
`QCTM256`	enc	90508	13580 KiB	474248 KiB
`QCTM256`	dec	90508	23852 KiB	474344 KiB
`QCTM512`	keygen	91716	3272 KiB	1827828 KiB
`QCTM512`	enc	91716	15324 KiB	1827592 KiB
`QCTM512`	dec	91716	15896 KiB	1826112 KiB

6. Transmission and storage overhead

External public-key, ciphertext and signature sizes follow the curated performance/external_sizes.csv catalog; secret-key and shared-secret lengths remain API figures. See the size audit for disagreements.

instance	public key	secret key	ciphertext	shared secret
`QCTM128`	167987	192545	640	32
`QCTM256`	595662	641942	1153	32
`QCTM512`	2374727	2467243	2264	64

Symmetric primitives

Survey verdict: bypass — own SHAKE256 for session key/keygen expansion; OpenSSL AES-256-CTR-DRBG seeded by the ICCS DRNG; auxfunc not linked

Share of each operation spent in the ICCS placeholder hash functions and in the ICCS DRNG (reference build, measured in the same run with link-time wrappers):

instance	operation	pseudohash/XOF/sm3	DRNG	calls per operation
`QCTM128`	keygen	0.0%	0.0%	drng 1
`QCTM128`	enc	0.0%	0.0%	–
`QCTM128`	dec	0.0%	0.0%	–
`QCTM256`	keygen	0.0%	0.0%	drng 1
`QCTM256`	enc	0.0%	0.0%	–
`QCTM256`	dec	0.0%	0.0%	–
`QCTM512`	keygen	0.0%	0.0%	drng 1
`QCTM512`	enc	0.0%	0.0%	–
`QCTM512`	dec	0.0%	0.0%	–

7. Raw evidence index

Paths are relative to performance/data/arm_1/ in the harness; each JSON file records its commands, environment and trials.

instance	item	file
`QCTM128`	KAT log (sha256 `2a8169ca8ae22343…`)	`kat/kem-32/QCTM128.log`
`QCTM128`	timing dec	`records/kem-32/QCTM128__dec.json`
`QCTM128`	timing enc	`records/kem-32/QCTM128__enc.json`
`QCTM128`	timing keygen	`records/kem-32/QCTM128__keygen.json`
`QCTM128`	hash profile dec	`profile/kem-32/QCTM128__dec.json`
`QCTM128`	hash profile enc	`profile/kem-32/QCTM128__enc.json`
`QCTM128`	hash profile keygen	`profile/kem-32/QCTM128__keygen.json`
`QCTM256`	KAT log (sha256 `aa2c8443858ae876…`)	`kat/kem-32/QCTM256.log`
`QCTM256`	timing dec	`records/kem-32/QCTM256__dec.json`
`QCTM256`	timing enc	`records/kem-32/QCTM256__enc.json`
`QCTM256`	timing keygen	`records/kem-32/QCTM256__keygen.json`
`QCTM256`	hash profile dec	`profile/kem-32/QCTM256__dec.json`
`QCTM256`	hash profile enc	`profile/kem-32/QCTM256__enc.json`
`QCTM256`	hash profile keygen	`profile/kem-32/QCTM256__keygen.json`
`QCTM512`	KAT log (sha256 `8162573769a321d8…`)	`kat/kem-32/QCTM512.log`
`QCTM512`	timing dec	`records/kem-32/QCTM512__dec.json`
`QCTM512`	timing enc	`records/kem-32/QCTM512__enc.json`
`QCTM512`	timing keygen	`records/kem-32/QCTM512__keygen.json`
`QCTM512`	hash profile dec	`profile/kem-32/QCTM512__dec.json`
`QCTM512`	hash profile enc	`profile/kem-32/QCTM512__enc.json`
`QCTM512`	hash profile keygen	`profile/kem-32/QCTM512__keygen.json`

Scripts: performance/campaign.py, performance/ngcc_perf.c, performance/hashprof/ in the harness.