Intel Aims to Cut NFV Latency
The Open Virtualization Profile (OVP), as it's called, is part of a dedicated push by Intel into network functions virtualization (NFV) and software-defined networking (SDN).
It's a virtualization add-on to Wind River's embedded Linux software. OVP does things like provision and migrate virtual machines -- but it's also got a twist aimed at carriers.
So far, virtualization software has been developed with the data center in mind, aiming to maximize throughput. But running the traffic through hypervisors and virtual switches -- extra elements of software -- adds latency that's not present in a hardware-centric network, says Davide Ricci, Wind River's product line manager.
The difference in timing can be about 700 microseconds, worst-case. For many data-center applications, that's as negligible as adding one more brown M&M to the bowl. But carriers, just like certain rock bands, get really bothered knowing that extra latency is in there, Ricci says.
"That's the big deal. In the physical world, there's no context switch, but in the virtual world, the context switch is a part of the system that introduces latency," Ricci says, with "context switch" being the phrase for the traversing of software elements such as the hypervisor or virtual switch.
This can be fixed by letting software pre-empt whatever the hypervisor is doing. So, that's what Wind River did: OVP adds a pre-empting capability to KVM, an open-source hypervisor.
OVP can shrink the worst-case delay to "well below 15" microseconds, with average delays being more like 3 or 4 microseconds, Ricci says.
Why this matters
OVP adds to Intel's NFV-related arsenal, which includes reference designs for switches and virtual switches, introduced in April. NFV provides a rich opportunity for processor vendors, and Intel seems determined not to blow its chance here.
— Craig Matsumoto, Managing Editor, Light Reading