This document specifies minimum and recommended hardware requirements
for ArcherDB deployments, along with cloud instance mappings and sizing
formulas.
Use hardware planning together with capacity quotas:
Tier
Runtime posture
Capacity posture
lite
Same high-performance runtime as all tiers
Smallest RAM/disk quota
standard
Same high-performance runtime as all tiers
Larger RAM/disk quota
pro
Same high-performance runtime as all tiers
Mid-tier RAM/disk quota
enterprise
Same high-performance runtime as all tiers
Large RAM/disk quota
ultra
Same high-performance runtime as all tiers
Highest RAM/disk quota
See Tier Profiles for canonical
runtime/capacity intent.
Entities
Min RAM
Rec RAM
Min Disk
Rec Disk
Min CPU
1M
2 GB
4 GB
10 GB
50 GB
2 cores
10M
4 GB
8 GB
50 GB
200 GB
4 cores
100M
16 GB
32 GB
200 GB
1 TB
8 cores
500M
64 GB
96 GB
1 TB
5 TB
16 cores
1B
96 GB
128 GB
2 TB
10 TB
32 cores
Events/sec
Min RAM
Rec RAM
Rec CPU
Rec Disk
10K
4 GB
8 GB
4 cores
SATA SSD
100K
8 GB
16 GB
8 cores
NVMe
500K
16 GB
32 GB
16 cores
NVMe Gen3+
1M
32 GB
64 GB
32 cores
NVMe Gen4
These are the absolute minimum specifications for ArcherDB to
function. Performance may be limited.
For development, testing, and small-scale deployments (<1M
entities):
Component
Minimum
Notes
CPU
2 cores, x86-64
Must support AES-NI
RAM
2 GB
Limits entity count to ~20M
Disk
10 GB SSD
HDD not supported
Disk Speed
100 MB/s seq
SATA SSD sufficient
Network
100 Mbps
For replication
For production single-node deployments (<100M entities):
Component
Minimum
Notes
CPU
4 cores, x86-64
AVX2 recommended
RAM
8 GB
Limits entity count to ~70M
Disk
100 GB NVMe
Direct I/O requires NVMe
Disk Speed
1 GB/s seq
NVMe Gen3 minimum
Network
1 Gbps
10 Gbps for replication
For production cluster deployments:
Component
Minimum per Node
Notes
CPU
8 cores, x86-64
AVX2 highly recommended
RAM
16 GB
All nodes must match
Disk
200 GB NVMe
Replicas need identical storage
Disk Speed
2 GB/s seq
NVMe Gen3+
Network
10 Gbps
Between replicas
For typical production workloads (100M-500M entities, <500K
events/sec):
Component
Specification
Rationale
CPU
16 cores, Intel Xeon or AMD EPYC
Handles compaction + queries
RAM
64 GB ECC
500M entities + cache
Disk
1 TB NVMe Gen4
3+ GB/s for low latency
Network
25 Gbps
Replication headroom
For demanding workloads (>500M entities, >500K events/sec):
Component
Specification
Rationale
CPU
32+ cores, latest gen
Max throughput
RAM
128-256 GB ECC
Billions of entities
Disk
2+ TB NVMe Gen5
7+ GB/s peak
Network
100 Gbps
Multi-region replication
Required CPU features:
Feature
Required
Why
x86-64
Yes
Instruction set
AES-NI
Yes
Aegis-128L checksums and crypto primitives used by core
internals
AVX2
Recommended
SIMD operations
RDTSCP
Recommended
High-resolution timing
ARM64 (Apple Silicon, Graviton) is also supported.
Metric
Minimum
Recommended
High Perf
Sequential Read
1 GB/s
3 GB/s
7 GB/s
Sequential Write
500 MB/s
1.5 GB/s
5 GB/s
Random Read IOPS
50K
200K
500K
Random Write IOPS
30K
100K
300K
Latency (p99)
<1ms
<0.5ms
<0.2ms
Important: ArcherDB uses Direct I/O. Ensure:
File system supports O_DIRECT (ext4, xfs, btrfs)
No network-attached storage for data files
NVMe strongly preferred over SATA
ArcherDB maintains a RAM index for O(1) entity lookups:
Index Memory = (Entity Count / Load Factor) x Entry Size
Where:
- Load Factor = 0.70 (target 70% utilization)
- Entry Size = 96 bytes per runtime slot (64-byte entry plus scan-helper arrays)
Recommended RAM = Index Memory x 1.4 (headroom for cache, buffers)
Entity Count
Index Size
Min RAM
Rec RAM
1 Million
137 MB
2 GB
4 GB
10 Million
1.37 GB
4 GB
8 GB
50 Million
6.86 GB
12 GB
16 GB
100 Million
13.7 GB
24 GB
32 GB
250 Million
34.3 GB
48 GB
64 GB
500 Million
68.6 GB
96 GB
128 GB
1 Billion
137 GB
160 GB
192 GB
For a production node with 128 GB RAM:
Component
Allocation
Purpose
Index
~69 GB
Entity lookup and scan-helper arrays (500M entities)
Block Cache
16 GB
LSM tree read caching
Query Buffers
2 GB
Result assembly
VSR Buffers
2 GB
Replication pipeline
OS/Overhead
16 GB
Kernel, page cache
Base Storage = Entity Count x Event Size x History Depth
Where:
- Event Size = 128 bytes per GeoEvent
- History Depth = average events per entity
With Safety Margin:
Recommended Storage = Base Storage x 1.5
Entities
Latest Only
10x History
50x History
1M
128 MB
1.3 GB
6.4 GB
10M
1.3 GB
13 GB
64 GB
100M
13 GB
130 GB
640 GB
500M
64 GB
640 GB
3.2 TB
1B
128 GB
1.3 TB
6.4 TB
Workload
Storage Type
Why
Development
SATA SSD
Cost-effective
Standard Production
NVMe Gen3
Good balance
High Throughput
NVMe Gen4/Gen5
Maximum IOPS
High Capacity
NVMe + S3 tiering
Cost-effective scale
Replication Bandwidth = Events/sec x Event Size x Replicas
Example: 100K events/sec with 3 replicas
= 100,000 x 128 x 2 (primary to backups)
= 25.6 MB/s = 205 Mbps
Events/sec
Min Bandwidth
Rec Bandwidth
10K
100 Mbps
1 Gbps
100K
1 Gbps
10 Gbps
500K
5 Gbps
25 Gbps
1M
10 Gbps
50 Gbps
Scenario
Max Latency
Recommended
Same rack
1ms
<0.1ms
Same datacenter
5ms
<1ms
Same region
20ms
<5ms
Cross-region
100ms
<50ms
Use Case
Instance
vCPUs
RAM
Storage
Cost/mo
Dev/Test
t3.large
2
8 GB
gp3
~$60
Small Prod
m6i.xlarge
4
16 GB
gp3
~$140
Standard Prod
m6i.4xlarge
16
64 GB
io2
~$560
High Perf
m6i.8xlarge
32
128 GB
io2
~$1,120
Extreme
i4i.4xlarge
16
128 GB
local NVMe
~$1,000
Storage Notes:
Use io2 for production (up to 64K IOPS)
i4i instances have local NVMe (best performance)
gp3 sufficient for dev/test
Use Case
Instance
vCPUs
RAM
Storage
Dev/Test
e2-standard-2
2
8 GB
pd-ssd
Small Prod
n2-standard-4
4
16 GB
pd-ssd
Standard Prod
n2-standard-16
16
64 GB
pd-extreme
High Perf
n2-standard-32
32
128 GB
local SSD
Extreme
c3-standard-44
44
176 GB
local SSD
Use Case
Instance
vCPUs
RAM
Storage
Dev/Test
Standard_D2s_v5
2
8 GB
Premium SSD
Small Prod
Standard_D4s_v5
4
16 GB
Premium SSD
Standard Prod
Standard_D16s_v5
16
64 GB
Premium SSD v2
High Perf
Standard_D32s_v5
32
128 GB
Ultra Disk
Extreme
Standard_L16s_v3
16
128 GB
local NVMe
Provider
Configuration
Approx Cost
Hetzner AX102
32 cores, 128 GB, 2x NVMe
~$180/mo
OVH Advance-2
16 cores, 64 GB, 2x NVMe
~$120/mo
Vultr Bare Metal
24 cores, 256 GB, NVMe
~$350/mo
Equinix m3.large
24 cores, 64 GB, NVMe
~$500/mo
Required RAM (GB) = ceil(entities / 10_000_000) * 1.5 + 4
Required Disk (GB) = (entities * 128 * history_depth) / (1024^3) * 1.5
Required Cores = max(4, ceil(events_per_sec / 100_000) * 4)
RAM:
Index = (100M / 0.7) * 96 = 13.7 GB
Recommended = 13.7 * 1.4 = 19.2 GB
With headroom = 24 GB minimum, 32 GB recommended
Disk (with 20x history):
Base = 100M * 128 * 20 = 256 GB
Recommended = 256 * 1.5 = 384 GB
Actual = 512 GB NVMe
CPU:
Base = 50K / 100K * 4 = 2 cores
Recommended = 8 cores (for compaction headroom)
Network:
Replication = 50K * 128 * 2 = 12.8 MB/s
Recommended = 1 Gbps
1. Entity Estimation
Current entities: ____________
Annual growth rate: ____________%
Planning horizon: ____________ years
Projected entities: ____________
2. Throughput Estimation
Peak events/sec: ____________
Average events/sec: ____________
Batch size: ____________
3. Memory Calculation
Index memory: ____________ GB (entities / 0.7 * 96 / 1GB)
With headroom: ____________ GB (index * 1.4)
Actual RAM: ____________ GB (round up to available)
4. Storage Calculation
Base storage: ____________ GB (entities * 128 / 1GB)
With history: ____________ GB (base * history_depth)
With margin: ____________ GB (with_history * 1.5)
Actual disk: ____________ GB (round up)
5. Instance Selection
Cloud provider: ____________
Instance type: ____________
Monthly cost: $____________
Edit this page