CloudArc Large-Package Benchmark

Generated: 2026-09-09T16:40:14.766191+00:00
Run status: complete
Profile: compressible-text
Workload: v2 / 59ac9b284e1c4a7c
Overall SLO: PASS

Environment

• Platform: Windows-11-10.0.26200-SP0
• Machine: AMD64
• Python: 3.14.0
• Sampling interval: 0.05 seconds

Bounded-Memory SLO

Requirement	Limit
Peak RSS per pack/unpack process	256.0 MiB
Peak RSS spread across tested sizes	64.0 MiB
Pack staging disk	2.10x payload + 64.0 MiB
Unpack staging disk	1.10x payload + 64.0 MiB

Results

Size GiB	Operation	Seconds	MiB/s	Peak RSS MiB	RSS delta MiB	Temp GiB	Temp/input	SLO
1.000	pack	33.496	30.6	26.0	3.0	1.002	1.002	PASS
1.000	unpack	1.421	720.6	29.7	6.8	1.000	1.000	PASS
5.000	pack	183.054	28.0	25.8	2.8	5.008	1.002	PASS
5.000	unpack	75.702	67.6	31.0	8.1	5.000	1.000	PASS
10.000	pack	488.938	20.9	25.9	2.8	10.015	1.002	PASS
10.000	unpack	494.140	20.7	31.2	8.2	10.000	1.000	PASS

Scaling Evaluation

• Requested/completed sizes: 3/3 (PASS).
• Pack peak-RSS spread: 0.2 MiB (PASS).
• Unpack peak-RSS spread: 1.5 MiB (PASS).

Method

• Each operation runs in a fresh child process.
• Peak RSS combines parent sampling with the worker’s final OS high-water mark.
• Temporary disk counts logical bytes only inside .cloudarc-pack-* or .cloudarc-unpack-* staging directories.
• Sampling is combined with an operation-derived floor so a short atomic publication cannot be missed.
• Dataset generation and durable source/archive/output files are excluded.
• The SLO covers payload-size scaling with one file and bounded lexical vocabulary; manifest/index cardinality has a separate memory cost.