6. Compute Domain – General Architecture and Associated
Interfaces
In the above figure following interfaces are important:
Interface 11 : Compute domain and management and
orchestration
Interface 14 : Interconnecting to infrastructure network
8. Compute Domain - Interest
Compute Domain related interfaces:
VNF-NFVI : Reference point between NFVI and VNF
VI-Ha : Reference point interfacing the virtualization layer to the
hardware resources
Nf-Vi : Reference point used to assign virtualized resources in
response to resource allocation requests
9. Compute Domain – Functional Elements
Compute Domain includes following functional elements :
Processor and Accelator
x86 or ARM based
Network interfaces ( NIC’s) card
Storage
Large scale and non-volatile like hard disks, solid state disks and so on.
11. Compute Domain – External Interfaces
Physical network Interfaces
Ethernet
CAT 5/ 6 cable with speed up to 1 Gbps
IEEE802.3 specs compliant
Fibre-Channel
Fibre channel with speeds : 1, 2, 4, 8, 10, 16 and 20 Gbps
4G and 8G standard compliant
Infniband
Serial style of interface
QDR, FDR, EDR speeds
12. Compute Domain – External Interfaces
Internal and External domain interfaces
NFVI to VIM (Nf-Vi)
Nf-Vi/C
Used by VIM to manage the compute and storage portion of NFVI
Vi-Ha/csr
Interface between the compute domain and hypervisor domain
Used by hypervisor/OS to monitor the available physical resources of the
compute domain
Management and Orchestration interfaces
13. Compute Domain – E2E Requirements
Below attachment defines the E2E requirement for compute and
storage
14. Compute Domain – Functional Blocks
Definition
“An NFV compute node is a coherent domain with an instruction set
architecture that is managed as a single compute execution environment. A
compute node includes a NIC to communicate with other compute nodes,
network elements and storage elements”
Various blocks of the compute Domain are detailed below :
CPU Complexity
Level of integration
H/W and S/W interfaces
Hypervisor interfaces
Instruction-sets and programmability
Number of CPU sockets
Clock speed and form factor
15. Compute Domain – Functional Blocks
Network Interfaces and Accelerator
NIC to include H/W acceleration engines for encryption, encapsulation,
forwarding and switching
Hardware accelerators such as encryption, digital signal processing
(DSP), packet header processing, packet buffering and scheduling etc
support
Additions to the instruction set architecture which implement new s/w
acceleration features
16. Compute Domain – Functional Blocks
H/W abstraction layer implemented by the hypervisor and guest O.S
17. Compute Domain – Interfaces within domain
PCIe
Interconnects the NIC and/or acceleration cards to the host CPU
SR-IOV ( Single Root I/O Virtualisation)
Used to virtualize the PCIe and the attached NIC
One physical NIC card can support up to 128 virtual functions (VF’s)
Assigning each PCI express virtual function can be directly assigned to
a virtual machine
I/O overhead in the software emulation layer is diminished
18. Compute Domain – Interfaces within domain
RDMA ( Remote direct memory access)
Allows server-to-server data movement directly between applications
memory
Provides low latency, improved resource utilization, flexible resource
allocation, scalability and unified fabric
RDMA over Converged Ethernet (RoCE)
New RDMA protocol over Ethernet
Provide data center convergence over reliable Ethernet and Data center
bridging (DCB)
RoCE communications can be as low as 1/10th the latency of any other
standards
19. Compute Domain – Interfaces within domain
Infiband
Direct access to a Network Interface (NIC) straight from application
space
Ability for apps to exchange data directly between their respect buffers
across a network
Creation of a virtual channel connecting two applications which exist in
entirely separate address space
Provides messaging service that application can access
Used for storage, IPC, message exchanging
Deliver higher bandwidth and low latency
21. Compute Domain – NFVI Modularity
Composite - NFVI
NFVI from COTS elements like servers, disk, switches, power supplies,
fans and so on
Another approach is use of an OCP pooled resource structure
Field replaceable units are components of the NFVI
NFVI-Pod
NFVI from COTS elements but with no field replaceable units
Deployed with various capacities and form factor
Failed elements will be marked as unavailable for use
NFVI-Plugin
COTS deployable unit of NFVI H/W resources that is field replaceable
unit within some other network element
Managed by the NFV management and orchestration entities
22. Compute Domain – NFVI Modularity
Composite - NFVI
NFVI from COTS elements like servers, disk, switches, power supplies,
fans and so on
Another approach is use of an OCP pooled resource structure
Field replaceable units are components of the NFVI
NFVI-Pod
NFVI from COTS elements but with no field replaceable units
Deployed with various capacities and form factor
Failed elements will be marked as unavailable for use
NFVI-Plugin
COTS deployable unit of NFVI H/W resources that is field replaceable
unit within some other network element
Managed by the NFV management and orchestration entities