Attributes of HyperAI Native Nodes

This document provides an overview of the key attributes associated with NativeNodes in HyperAI.

interfaceName

Description: The interfaceName attribute refers to the specific network interface used by the node, e.g. eth0, to communicate with the control plane.

Kubernetes Type: Annotation

Implementation: See interface_name_and_type

monetaryCost

Description: The monetaryCost attribute stores the predicted monetary cost of a HyperAI node. The value is inferred from the node CPU and memory using the packaged cost regression model.

Kubernetes Type: Annotation

Kubernetes Key: hyperai.eu/node-monetary-cost

Implementation: See get_monetary_cost_annotation

monetaryCostCategory

Description: The monetaryCostCategory attribute stores the qualitative cost label assigned from the predicted monetary cost value only. Supported labels are very low, low, medium, high and very high.

Kubernetes Type: Annotation

Kubernetes Key: hyperai.eu/node-monetary-cost-category

Implementation: See get_monetary_cost_annotation

geolocation

Description: The geolocation attribute is split in three fields (city, region, country) used for selection from the HyperAI scheduler. It uses the node’s public IP address to determine them.

Kubernetes Type: Labels

Implementation: See geolocation

bandwidth

Description: Bandwidth is the maximum rate of data transfer across a network path. This attribute is split in two fields (download speed, upload speed). Each value is categorized into one of the following classes based on its measured speed:

Class A: ≥ 100 Mbps
Class B: ≥ 50 Mbps and < 100 Mbps
Class C: ≥ 10 Mbps and < 50 Mbps
Class D: < 10 Mbps

Kubernetes Type: Annotations

Implementation : See bandwidth

latency

Description: Latency is the time it takes for a packet of data to travel from its source to its destination.

Kubernetes Type: Annotation

Implementation : See latency

packetLoss

Description: The packetLoss occurs when one or more packets of data travelling across a computer network fail to reach their destination.

Kubernetes Type: Annotation

Implementation : See packetLoss

nodeCategory

Description: The nodeCategory will be a short and representative string of the nodes computing capacity and accelerators.

Kubernetes Type: Annotation

Implementation : See nodeCategory

uptime

Description: Uptime refers to the duration for which the node has been running without interruption.

Kubernetes Type: Label

Implementation : See uptime

TPU

Description: This attribute will describe the total TPUs in the node.

Kubernetes Type: Label

Implementation : See TPU

Accelarators

Description: This attribute will describe the available accelerators for future scheduling.

Kubernetes Type: Label

Implementation : See Accelerators

nodePool

Description: The nodePool is a custom tag, for logical clustering, that will group the nodes into groups. The separation can be done with respact to the node’s purpose (e.g. production/testing) , special features (GPU pool) or any other criterion.

Kubernetes Type: Label

Implementation : See NodePool

networkType

Description: The networkType attribute refers to the specific technology or protocol used for network communication. It can be Ethernet, Wi-Fi or unknown.

Kubernetes Type: Annotation

Implementation: See interface_name_and_type

energyEfficiency

Description: The energyEfficiency attribute stores the estimated energy-efficiency value of the node. HyperTool first tries to obtain this value from the labelled hardware dataset using CPU core count and clock speed. The FLOPs and TDP regressors are used only as a fallback when no usable hardware match is available.

Kubernetes Type: Annotation

Kubernetes Key: hyperai.eu/node-energy-efficiency

Implementation: See energy_efficiency_annotation

energyEfficiencyCategory

Description: The energyEfficiencyCategory attribute stores the qualitative energy-efficiency label assigned from the final energy-efficiency value. Supported labels are very low, low, medium, high and very high.

Kubernetes Type: Annotation

Kubernetes Key: hyperai.eu/node-energy-efficiency-category

Implementation: See energy_efficiency_annotation

energyEfficiencySource

Description: The energyEfficiencySource attribute shows how the energy-efficiency value was obtained. Possible values are hardware_exact, hardware_nearest, model_fallback and unknown.

Kubernetes Type: Annotation

Kubernetes Key: hyperai.eu/node-energy-efficiency-source

Implementation: See energy_efficiency_annotation

flopsPerSec

Description: he flops_per_sec attribute represents the computational performance of the node.

Kubernetes type: Annotation

Kubernetes Key: hyperai.eu/node-flops-per-sec

Implementation: See flops_per_sec

trustScore

Description: The trustScore attribute depicts how trustworthy a node is. It is computed from a weighted set of runtime-health signals that the node observes about itself locally.

Kubernetes type: Annotation

Kubernetes Key: hyperai.eu/node-trust-level

Implementation: See trust_level

securityLevel

Description: The securityLevel attribute describes how secure a node is against malicious outside attacks. It is computed from a weighted set of observable security signals.

Kubernetes type: Annotation

Kubernetes Key: hyperai.eu/node-security-level

Implementation: See security_level

name

Description: The name of the node is a unique identifier for the node within the Kubernetes cluster.

Kubernetes Type: Field