As the number of processor cores in a server increases, the number of virtual machines and containers making requests to access data exponentially increases. Before too long, network bottlenecks start to emerge as all those virtual machines and containers attempt to share access to the underlying physical network.
To alleviate that issue, Solarflare is now adding XtremeScale Fabric software to a network interface card (NIC) that accelerates performance by allowing applications to bypass an operating system to access the network via a Universal Kernel Bypass (UKB) suite that includes multiple instances of TCP, UDP, DPDK, and NVMe-oF protocols deployed on the NIC.
Ahmet Houssein, vice president of marketing at Solarflare, says XtremeScale Flare software makes it possible for servers to attain network performance rates that otherwise would have required an IT organization to acquire a field programmable gate array (FPGA).
Houssein says the Smart NICs provided by Solarflare make it feasible for IT organizations to embrace multi-core processors and graphical processor units (GPUs) to increase application performance as an alternative to processors featuring faster clock speeds.https://o1.qnsr.com/log/p.gif?;n=203;c=204663295;s=11915;x=7936;f=201904081034270;u=j;z=TIMESTAMP;a=20410779;e=i
“From a clock speed perspective, Moore’s Law is dead,” says Houssein.
Smart NICs, adds Houssein, also make it possible to offload firewalls from a server on to the NIC in addition to allowing Solarflare to develop a Solarflare Capture application, capturing packets at rates of up to 40 Gbps. That data can then be used to drive any number of analytics and compliance applications, says Houssein. Solarflare has also developed a SolarSecure application that monitors packet flows across the NIC.
Houssein says it’s increasingly apparent that enterprise IT organizations are going to need web-scale level network performance in the years ahead, requiring both new classes of servers and NICs. The challenge, says Houssein, will be finding a way to automatically balance the load demands of all the workloads that will be running on servers that continue to become more dense with each new processor advance.