Hardware Sizing & Performance Benchmarks

This reference guide provides hardware sizing profiles and performance benchmark measurements obtained on standard enterprise server hardware.

Hardware Sizing Profiles

Fleet Tier	Usable Storage	Recommended Node Count	CPU per Node	RAM per Node	Drive Specification	Network Interface
Departmental Pilot	250 TB – 1 PB	6 – 8 Nodes	16 Cores (AVX2)	64 GB ECC	24TB SAS-4 HDDs + 2x NVMe Write Cache	2x 25 GbE LACP
Enterprise Standard	2 PB – 10 PB	12 – 24 Nodes	32 Cores (AVX-512)	128 GB ECC	24TB SAS-4 HDDs + NVMe Write Ahead Log	2x 100 GbE RoCEv2
Hyperscale Archive	10 PB – 50+ PB	32 – 96 Nodes	64 Cores (AVX-512)	256 GB ECC	High-Density JBODs (102-bay)	4x 100 GbE Mellanox

Ingestion & Read Throughput Benchmarks

Tested on a 12-node cluster with dual Intel Xeon Platinum 8480+ (56 cores, AVX-512) and 100 Gbps network interfaces:

[Throughput Performance by Operation]
Sequential Read (Profile 8+3):  ████████████████████ 42.8 GB/s
Sequential Write (Encrypted):    ████████████████ 34.2 GB/s
Point Lookup Latency (Kdouja):  ██ 0.38 ms (P99)
Abodi Telemetry CPU Overhead:    ▌ <0.8% of 1 Core

Tip

Zero Heap GC Stalls: Due to Kdouja's arena skip-lists and zero-allocation memory spans, P99.9 write latency remains flat under sustained 100 Gbps ingestion without garbage collection jitter.