UDM-PRO SE
10G LAN/WAN over SFP+ · IDS / IPS up to 3.5+ Gbps
- Uplinks
- 2x ISP path
- Threat mgmt
- Built-in
Appendix · Technical overview / Datasheet · v2026.05-v3
The infrastructure
your validator runs on.
Chef Node operates EarthNodes on enterprise hardware we own and run end-to-end: Supermicro A+ servers, AMD EPYC compute, dual ISP uplinks, sentry-gated validators, ZFS storage, battery-backed power, and no public-cloud custody.
Compute192cores per server
Memory1.1TB DDR5 ECC
NVMe61.4TB raw Gen4 hot tier
Archive108TB populated backup spine
Fabric1.34Tbps vendor-rated switching
A five-layer network segmented like a bank's.
North-south traffic enters through a redundant edge, passes IDS/IPS inspection, aggregates over 25-Gigabit SFP28 uplinks, and lands on isolated VLANs at the distribution layer. Two independent ISP paths keep the node reachable when one route disappears.
USW-Pro-Aggregation
4x 25G SFP28 uplinks · 760 Gbps switching capacity
- Redundancy
- 2 up / 2 down
- VLAN
- Honey-pot per VLAN
USW-EnterpriseXG-24
2x 25G SFP28 uplinks · 580 Gbps switching capacity
- Layer
- L3
- Identity
- Private addressing
Supermicro A+ AS-1125HS-TNR
Bare metal · Type-1 hypervisor · isolated VMs
- NICs
- Mellanox 25GbE + Intel 10GbE
- Fabric
- Quad path
45-Drives Storinator
High-density SATA archive · ZFS replication target · hot spare
- Drives
- 6x Exos X20 18 TB
- Raw
- 108 TB
Design principles
- 01Two ISPs, different last-mile technologies
One outage, one fiber cut, or one provider change window does not take the node offline.
- 02No validator port exposed to the internet
Validators are not directly routable from the public network. Sentries are the only north-south gateway.
- 03MAC-level allowlist on Layer 2
External physical devices are blocked by default. Anything new must be pre-registered.
- 04Application-layer load balancer, single port
All traffic is gated through one port. Internal services resolve through private DNS.
- 05Public identity abstracted behind proxies
The validator's private network identity is not advertised directly to internet scanners.
Supermicro A+ AS-1125HS-TNR.
The bare-metal platform Tier 1, Tier 2, and Tier 3 EarthNodes run on. Dual 96-core EPYC Turin, 1.1 TB DDR5 ECC, enterprise Gen4 NVMe, and independent network controllers in a single chassis with headroom for validators, sentries, AI inference VaaS, and storage services.
Hyper A+ Server
AS-1125HS-TNR · 1U · dual-socket
2x EPYC 9655 Turin
- Cores per CPU
- 96
- Threads per CPU
- 192
- Total threads
- 384
- Architecture
- Zen 5 · SP5 platform
- Use case
- Validator + sentry + VaaS, isolated
1.1 TB registered ECC
- Speed
- 4800 MT/s
- Population
- 12x 96 GB RDIMM
- Raw capacity
- 1,152 GB / 1.125 TB
- ECC
- Yes, error-correcting
- Use case
- In-memory VM working sets
Mellanox ConnectX-4 Lx + Intel X550-AT2
- 25 GbE
- Mellanox · dual SFP28
- 10 GbE
- Intel · dual RJ-45
- Offload
- RoCE, TCP/UDP, LSO/LRO, RSS
- Virt
- SR-IOV up to 256 VFs
- Overlay
- NVGRE / VXLAN hardware offload
Roadmap Mellanox ConnectX-6 Dx adds dual-port 100GbE or single-port 200GbE headroom, PCIe 4.0, RoCE, SR-IOV/OVS offload, and inline IPsec/TLS crypto acceleration for lower CPU overhead and cleaner secure fabric scaling.
8x Micron 7450 PRO · 7.68 TB U.3
- Total raw
- 61.44 TB
- Capacity class
- 64 TB nominal
- Interface
- PCIe Gen4 x4 · U.3
- Read seq
- 6,800 MB/s
- Write seq
- 5,600 MB/s
- QoS 99.99%
- ≤ 2 ms latency
Roadmap Next chassis qualifies PCIe Gen5 enterprise NVMe without changing the ZFS topology.
Three ZFS pools. One backup spine.
The NVMe array is partitioned into three high-availability ZFS pools: one for the EarthNode itself, one for snapshots and active backups, and one for value-added services. Every pool snapshots to a separate bare-metal backup server.
EN
Validator working setHot path for chain state, mempool, and validator keys. Encrypted at rest and hardware-isolated.
SNAP
Snapshots & active backupsContinuous ZFS snapshots, point-in-time recovery, and live replication for off-site backup.
VaaS
Value-added servicesDedicated capacity for AI inference, storage VaaS, VPN, and future World Mobile service tiers.
45-Drives Storinator
High-density SATA archive · backup and replication target
Direct-wire high-density storage server
- Form factor
- Rack-mount JBOD-class
- Bay capacity
- Up to 15 hot-swap bays
- Topology
- Direct-wired SATA, no expander bottleneck
- Filesystem
- OpenZFS · RAIDZ2
- Role
- ZFS replication target + warm archive
6x Seagate Exos X20 · 18 TB CMR
- Populated raw
- 108 TB
- Capacity class
- 120 TB archive tier
- Interface
- SATA 6 Gb/s
- Spindle speed
- 7,200 RPM
- Recording
- CMR conventional
- MTBF
- 2.5M hours
Roadmap Chassis is provisioned for expansion to 15x 44 TB drives, supporting a 660 TB raw archive tier.
Replication target + observability host
- Replication
- Continuous ZFS send/recv
- Snapshots
- Point-in-time, immutable
- Monitoring
- Prometheus + Grafana stack
- Power
- Independent Smart-UPS
- Failover role
- Promotable to primary
Direct fabric peering to primary compute
- Uplink
- 2x 10 GbE · redundant replication VLAN
- Housing
- Separately racked · isolated power
- PSU
- 2x 900 W hot-swap redundant
- UPS
- 2x APC SMT3000RMT2U · 2,700 W each
- Recovery
- RPO ≤ minutes · RTO ≤ minutes
Six points of failure, each one doubled.
Uptime is not a number we publish. It is a list of decisions made before a failure happens. Field-replaceable components, storage paths, network paths, and power inputs are designed with mirrored or redundant partners where the platform supports it.
Dual redundant PSUs
Hot-swappable power supplies keep the host online through a single PSU replacement or PSU fault.
Dual CPUs
Two EPYC 9655 Turin sockets provide compute headroom. Host-level faults are covered by backup, restore, and failover runbooks.
Storage path resilience
ZFS redundancy, mirrored boot media, snapshots, and the backup server prevent one storage fault from becoming one recovery path.
Quadruple network paths
Two NICs, two technologies, and four physical links keep the node talking.
RAID-1 boot drives
Dual M.2 NVMe boot drives are mirrored so a boot-disk fault can be isolated without data loss.
Mission-critical UPS
Smart UPS on primary and backup server gives the operator time to ride through or cleanly handle power events.
The validator is not directly routable.
Public traffic terminates before it reaches the validator host. It hits IDS/IPS at the edge, traverses isolated VLANs, passes through a single-port load balancer, and is relayed by a sentry on a private network.
Perimeter · IDS / IPS
UDM-PRO SE · 3.5+ Gbps inspectionEvery packet entering the network is inspected for known attack signatures and behavioral anomalies before it reaches a switch.
VLAN isolation · honey-pot per VLAN
L3 switching at aggregation + distributionNetwork segments are isolated by VLAN, and each VLAN ships with a honey-pot used to detect lateral movement.
MAC allowlist · L2
Pre-registered devices onlyUnknown physical devices cannot communicate on the network even if connected directly to a switch port.
Single-port load balancer
Application layer · private DNSNorth-south traffic enters through one software-controlled gate. Internal services are not publicly addressable.
Sentry-gated validator
Validator on a private isolated networkOnly the sentry talks to the validator over a single private port. The EarthNode is not directly advertised or routed publicly.
Encryption · proxies · zero-trust admin
End-to-end · self-hosted ZT planeAdministrative access flows through a self-hosted zero-trust platform that controls and monitors every connection.
OS hardening · 2FA · fail2ban
Root forbidden · brute-force preventionRoot login is forbidden, two-factor authentication is required, and fail2ban shuts brute-force attempts down at source.
✓
Third-party penetration tested.
Independent penetration testing is scoped against the exposed network, access controls, and administrative plane. Current scope, cadence, and remediation status can be shared during technical diligence.
Type-1 hypervisor. One VM per node.
Every EarthNode lives inside its own virtual machine on a Type-1 hypervisor, not a container and not a shared OS. Process isolation is enforced underneath the guest operating system.
Why Type-1, not containers.
Containers share a kernel. A bug in the kernel or a misconfigured namespace becomes everyone's problem. A Type-1 hypervisor gives every VM its own kernel and isolates it at the hardware level.
Practically, that means an EarthNode, sentry, AI inference VaaS, and storage VaaS can run on the same chassis while materially reducing cross-service risk.
Future WM services · auto-enabled at the contracted tierEN · Validatoryour node
Sentrypublic gateway
VaaSAI / storage / VPN
▲ Type-1 Hypervisor isolated VMs
Ubuntu 24.04 LTS most-supported
Supermicro A+ AS-1125HS-TNR bare metal compute
45-Drives Storinator bare metal storage
The whole picture, one diagram.
Public traffic on the left, your EarthNode on the right. Between them: five network layers, three ZFS pools, two ISPs, sentry isolation, and a dedicated storage server with on-site and off-site replication.
public traffic pathoff-site replication / sentry pathinternal isolated
Whole-site power, battery-backed.
The validator is buffered from grid events. An EG4 GridBOSS sits at service entrance, a FlexBOSS21 hybrid inverter handles transfer in milliseconds, and a 14.3 kWh LiFePO4 wall battery carries the load.
Power flow - how the stack hangs together
EG4 Power Backup Stack
GridBOSS MID v3.1 · FlexBOSS21 hybrid · PowerPro WallMount 14.3 kWh
EG4 GridBOSS MID v3.1
- Service rating
- 200 A · 22 kAIC main breaker
- Hybrid ports
- 3x · up to 90 A each
- Smart-load ports
- 4x programmable · 125 / 80 / 60 / 60 A
- Generator input
- Up to 125 A · 2-wire start
- Transfer
- 200 A automatic · seamless to off-grid
- Enclosure
- NEMA 3R outdoor · -40 F to 113 F
- Connectivity
- Wi-Fi · EG4 monitoring portal · OTA firmware
- Compliance
- UL 1741 · UL 67 · UL 869A · NEC 690.12 RSD
Wired role The APC Smart-UPS and DPU sit on a dedicated smart-load port so they stay energized first when non-critical loads are shed.
EG4 FlexBOSS21 IV-16000-HYB
- Continuous output
- 12 kW battery-only · 16 kW with PV
- Surge / LRA
- 156 A · starts a 5-ton AC compressor
- Grid pass-through
- 90 A @ 120/240 VAC
- PV input
- 21 kW usable · 24 kW max · 600 V DC
- MPPTs
- 3x independent · 26 / 26 / 15 A · up to 5 strings
- Battery comms
- RS-485 + CAN bus · closed-loop
- Parallel
- Up to 10 units · 160 kW total
- Enclosure
- NEMA 4X · coastal-rated
- Compliance
- UL 1741 · UL 9540 · CSA 22.2 #330 · NEC 690.12
EG4 PowerPro WallMount · 14.3 kWh
- Chemistry
- Lithium iron phosphate (LFP)
- Capacity
- 280 Ah · 51.2 V · 14.3 kWh total · 12.87 kWh usable
- Continuous power
- 10.24 kW · 11.52 kW peak · 300 A surge (3 s)
- Cycle life
- > 8,000 cycles @ 80% DoD
- Lifetime energy
- 82.6 MWh production envelope
- Round-trip efficiency
- 99%
- Self-heating
- Integrated · cold-weather ready
- Enclosure
- IP65 · indoor / outdoor
- Safety
- UL 1973 · UL 9540A · dual fire arrestors · E-Stop
- Warranty
- 10 years
Runtime At a steady ~1.2 kW draw for chassis, storage, networking, APC, and DPU, the WallMount holds the site for roughly 10-11 hours off-grid before reserve policy.
APC Smart-UPS + DPU network plane
- UPS role
- Line-interactive buffer after the inverter
- Network plane
- DPU stays online during transfer and load shedding
- Validator path
- Primary compute and storage remain on protected load
- Transfer layer
- GridBOSS + FlexBOSS + APC buffer
- Load priority
- Critical node services before discretionary loads
- Operations
- Power events are monitored and handled as runbook events
Design intent Solid lines are power flow. The smart-load port keeps UPS and network-control gear alive while the inverter and battery carry the validator stack.
The proof layer, not just the stack.
For technical reviews, the infrastructure conversation should include operational evidence, not only hardware diagrams. These are the controls we expect sophisticated operators to ask about.
Operator Evidence Pack
Available during technical diligence and tier negotiation
Key handling is scoped separately
- Validator keys
- Encrypted at rest
- Access
- Role-limited operator actions
- Recovery
- Documented backup and revocation path
- Disclosure
- Shared under diligence
Replication needs restore proof
- Snapshots
- Point-in-time ZFS snapshots
- Off-site
- Tier-dependent replication cadence
- Validation
- Restore drills tracked by runbook
- Objective
- RPO/RTO stated as tested targets
Perimeter controls before the sentry
- Ingress
- IDS/IPS, proxies, sentry gateway
- Exposure
- No direct validator route
- Escalation
- Upstream mitigation path documented
- Monitoring
- Alert routing and incident review
Baseline versus add-ons
- Baseline
- Dedicated VM, private routing, monitored host
- Tier 2
- Expanded backup, review cadence, and VaaS enablement
- Tier 3
- Premium SLA, deeper operational coverage, and full VaaS access
- Evidence
- Mapped into the operator agreement
Ready to talk specifics
Bring your EarthNode home.
Every line on this page is what we operate today for nodes already running. Want a walkthrough, a rack tour, or a copy of the latest pen-test report? Just ask.
Talk to a node operator →