The computational environment in the network compute fabric will be heterogeneous, which will increase with emerging computational hardware complementing traditional network equipment. One example is hardware acceleration technologies that provide compute and storage capabilities specialized for certain workload tasks, such as graphics processing units, Field-Programmable Gate Arrays (FPGA), and storage-class persistent memories.
By adopting tools and languages like OneAPI, developers will be able to write a single-source implementation of an algorithm that can run on different hardware choices. Selection of actual processing or memory/storage instance will be based on availability and performance requirements, making it easy for developers to create high-performance applications that can run anywhere.
The Network Compute Fabric also opens up for extensive in-network computation. Modern transport networking equipment will no longer be limited to packet transport but also provide programmable compute capabilities. This will, in turn, open up new divisions of application functionality where some tasks, such as QoS, scheduling, policing, encoding, and recoding can be performed directly in the data plane.
The Network Compute Fabric will enable efficiency gains through transparent shortcuts inside the infrastructure between application components and the network and compute platform. Once applications are collocated with network functions on the same host, rack, or cluster, parts of the network and operating system stacks can be bypassed. Depending on the situation, different technologies can be used, such as modern interconnect technologies and protocols in combination with fast replication services like Derecho.