White Papers

Understanding OPNFV

Issue link: https://read.uberflip.com/i/1344850

Contents of this Issue


Page 37 of 144

Understanding OPNFV 37 experienced if the workload is completely decoupled from the underlying hardware and can request resources through APIs. This, of course, is achieved through virtualization. Decoupling doesn't mean divorcing a VNF from hardware, since in NFV there will always be some VNFs that need servers with specific features, for instance data plane acceleration. Over time, new paradigms such as Platform-as-as-Service (PaaS) and "serverless apps" (where the code is executed in response to events while completely abstracting the underlying server, operating system and related infrastructure) will further promote VNF-hardware decoupling. An additional dependence on hardware is that traditional apps expect availability to be an attribute of the underlying infrastructure. So, it is the infrastructure that has to guarantee the required 9's of availability. In a cloud native application, the burden shifts to the VNF. The hardware is assumed to fail all the time. The current paradigm is often referred to as 'Pets' (i.e. each node has to be cared for) vs. the new paradigm that is referred to as 'Cattle' (i.e. instances are dealt with as an aggregate, and individual instances may come and go). Example 1: Old Contract Between VNF & 'Pets' Infrastructure ● CPU brand, type, frequency, number of sockets ● Memory size ● NVMe flash brand, size ● Hard drive brand, type, speed, size, quantity ● NIC brand, performance and quantity ● Power supply and fan specifications and quantity Example 2: New Contract Between VNF & 'Cattle' Infrastructure ● vCPU quantity ● Memory size ● HDD and/or SDD block storage size ● I/O with specific performance characteristics ● Data plane acceleration or other special features as required ● State isolation: In classic cloud native applications, the state for the application is isolated to database (SQL or NoSQL) services. All other functionality is in stateless services. Again, since networking is essentially all about distributed state management, state isolation applies only within the context of a microservices in VNFs. Each microservice should separate transaction processing logic from its state. This promotes

Articles in this issue

view archives of White Papers - Understanding OPNFV