The Linux Foundation’s Fast Data (FD.io, or Fido) collaborative software project is touting a set of new performance gains that it says will reach terabit levels to accommodate multiple deployment environments, including bare metal, virtual machine (VM) and container.
Similar to the way business users can order compute and storage on demand, FD.io is designed to enable service providers to offer on-demand network services. These services could include everything from dialing up bandwidth for a specific period, routing or firewall resources.
Fido is an open-source project that’s focused on establishing a high-performance I/O (input/output) services framework for dynamic computing environments.
Emerging improvement increases in Intel’s new Xeon processor family, including increased PCIe bandwidth, allowed Fido to double its performance at scale without software modifications.
Building on its recent 17.04 release, Fido claims it is the only vSwitch for which performance scaling is input/output-bound rather than central processing unit (CPU) bound.
“The increase in PCI bandwidth per socket in Intel’s latest product offering has doubled our performance to 1 Tbps with our existing unmodified 17.04 release, once again pegging the limit of what the PCI bus can handle,” said David Ward, FD.io board chair and SVP, CTO and chief architect of engineering for Cisco, in a release. “We look forward to being able to make similar gains as PCI bandwidth per socket continues to grow.”
A key component of the Fido project is vector packet processing (VPP), which handles server-class optimization such as vector instruction, cache optimizations, and packet pre-fetching. VPP reads the largest available vector of packets from the network I/O layer.
The VPP framework gives developers the potential to build any number of packet processing solutions by varying the underlying forwarding graphs while also easily accommodating new graph nodes to be plugged in.
Cisco has open sourced the Vector Packet Processor, a virtual switch and router, to the FD.io community to provide network operators with a modular, high-performance infrastructure using existing Intel-based systems.
The routing giant is hardly alone in its desire to enable on-demand services. AT&T already provides on-demand capabilities via its FlexWare service. These capabilities allow business users to request bandwidth and other resources such as routing and firewalls via an online portal.
AT&T's FlexWare currently includes four main elements: Juniper Networks' virtual routing, Cisco's virtual router, Fortinet's virtual security and Riverbed's virtual WAN optimization.
Additionally, AT&T now offers three other virtual security options: Palo Alto Networks Next-Generation Security Platform (AT&T- or self-managed); Juniper Networks vSRX Virtual Firewall (self-managed); and Check Point vSEC (self-managed).
A series of recent tests of FD.io’s release 17.04 showed performance gains on Intel’s newest platform when switching and routing layer 2/3 traffic.
In Intel’s previous generation of Xeon Processor E7-8890v3, FD.io testing showed aggregate forwarding rate of 480 Gbps (200 Mpps) for 4-Socket machine (using 4 of E7-8890v3 CPU configuration).
However, project participants noted that the same FD.io tests that run on two 2-socket blades (e.g. a modern 2RU server) with the new Intel Xeon Platinum 8168 CPUs within the same power budget show an increase in the forwarding rate to 948 Gbps (400 maximum packets per second) benefiting from the PCIe bandwidth increase of the new CPUs. Additionally, the tests revealed an overall decrease in cycles-per-packet due to CPU microarchitecture improvements.
For customers already building solutions using FD.io release 17.04, these types of performance gains are realized without further development effort as the software is optimized to take full advantage of the additional PCIe bandwidth and other CPU capabilities as the hardware becomes available.
Since being launched February 2016, Fido has continued to make progress with its projects. The group now leverages contributions from more than 173 contributors from over 45 different organizations including network operators, service providers, chip vendors, and integrators who are collaborating to enhance and innovate around VPP technology.