Backend#
Fabrics → BBackends
Documentation coming soon!
Custom Resource Definition#
To browse the Custom Resource Definition go to crd.eda.dev.
Backend
SPEC
BackendSpec defines the desired state of Backend
-
Address allocation profile for GPU endpoints.
default: {"edaManagedIPv6": {"leafIndexPoolScope": "Global", "prefixLength": "64"}, "type": "EDAManagedIPv6"}-
EDA managed IPv6 allocation configuration.
-
Leaf Index Pool Allocation scope (used for IP Address allocation). Global scope means that the leaf index will be allocated from a common pool for the entire namespace, Fabric scope means that the leaf index will be allocated from a pool for the entire fabric, Stripe scope means that the leaf index will be allocated from a pool for the stripe.
default: "Global"enum: "Global", "Fabric", "Stripe" -
IPv6 Prefix Length.
default: "64"enum: "64", "96"
-
-
Type of address allocation strategy.
default: "EDAManagedIPv6"enum: "EDAManagedIPv6", "GlobalIPv6Pool", "PerTenantIPv6Pool"
-
-
Reference to an IndexAllocationPool pool to use for Autonomous System Number allocations. Used when eBGP is configured as an underlay protocol.
-
GPU Isolation Groups are used to isolate GPU traffic over the network, GPUs in different GPU isolation groups will not be able to communicate with each other. If all GPUs across all stripes need to be able to communicate with each other, create a single GPUIsolationGroup selecting all GPU facing interfaces.
-
-
Name of the IsolationGroup.
-
-
IP MTU for this fabric. Default is 4136 bytes.
default: 4200range: 1500 to 9000 -
Set of properties to configure the RoCEv2 QoS.
-
If the queue depth is between min and max threshold then this the probability with which packets are dropped or marked.
default: 100 -
The maximum threshold parameter for a RED-managed queue in percent. When the average queue length exceeds the max value, all packets are dropped (or marked if ECN is enabled). Mutually exclusive with min-threshold and max-threshold.
default: 80 -
The minimum threshold parameter for a RED-managed queue in percent. When the average queue length is less than min, all packets are admitted to the queue. Mutually exclusive with min-threshold and max-threshold.
default: 5 -
Number of milliseconds during which outgoing interface is receiving pfc-pause-frames before triggering recovery-timer.
default: 750 -
Number of milliseconds during which the pfc-pause-frames will be ignored.
default: 750 -
Maximum amount of shared buffer memory available to the queue in bytes.
default: 1024000format: int32range: 0 to 4294967295
-
-
StripeConnector is the spine layer interconnecting multiple stripes.
-
Reference to an IndexAllocationPool pool to use for Autonomous System Number allocations.
-
Selects TopoLinks to include in this AI Fabric, the selected TopoLinks will be used to create ISLs between the stripe connector devices and the leaf devices.
-
The name of the Stripe Connector.
-
Node selector to select the nodes to be used for this stripe connector.
-
Reference to an IPAllocationPool used to dynamically allocate an IPv4 address to system/lo0 interfaces for the stripe connector devices. If not specified, the system will use the default IPAllocationPool.
-
-
A list of stripes, stripes contain a set of nodes (rails).
-
-
Optional reference to an IndexAllocationPool pool to use for Autonomous System Number allocations. If left blank, ASN allocation will be done from the ASNAllocationRange.
-
The VLAN used on interfaces facing the GPU servers.
format: int32range: 1 to 4094 -
The name of the Stripe.
-
Node selector to select the nodes to be used for this stripe.
-
Unique ID for a stripe
-
Optional reference to an IPAllocationPool used to dynamically allocate an IPv4 address to system/lo0 interfaces. If left blank, system IP allocation will be done from the SystemIPV4Subnet.
-
-
-
Reference to an IPAllocationPool used to dynamically allocate an IPv4 address to system/lo0 interfaces. If specified under the Leaf/Spine/Superspine/Borderleaf those will take precedence.
STATUS
BackendStatus defines the observed state of Backend
-
Indicates the health score of the Fabric. The health score of the Fabric is determined by the aggregate health score of the resources emitted by the Fabric such as ISL, DefaultRouteReflectors etc.
-
Indicates the reason for the health score.
-
The time when the state of the resource last changed.
-
Operational state of the Fabric. The operational state of the fabric is determined by monitoring the operational state of the following resources (if applicable): DefaultRouters, ISLs.
enum: "Up", "Down", "Degraded", "Unknown"