Data Plane Challenges for Emerging Data Center Architectures
I’ve seen a lot of articles and blogs about SDN discussing topics related to the network control plane, such as the use of OpenFlow® to improve overall network management and the wide range of emerging solutions for network orchestration. Along the same lines, however, service providers and network equipment manufacturers have also recognized significant challenges in the data plane that have major implications for the overall throughput of data centers. I spend a lot of time thinking about data plane challenges that must be addressed through advanced software solutions, as part of maximizing the performance of data center networking subsystems.
One set of data plane challenges applies to the virtual switches instantiated on application server blades.
First, the number of Virtual Machines (VMs) per server blade is increasing rapidly, leveraging on-going improvements in the performance of the processors used on those blades. Today, a typical server blade hosts around 20 VMs, with that number expected to grow to hundreds within a few years. Because of this growth, the data-center network needs to expand beyond its current limit at the Top-of-Rack, to the point where a virtual switch on each server blade is used to distribute the increasing volume of network traffic to virtualized applications. This implies the need for a high-performance virtual switch solution that is not constrained by the I/O performance limitations of standard hypervisors.
Second, adding to the networking load now placed on server blades, multi-tenant architectures serving applications for multiple users require traffic engineering (ACL, tunneling, QoS, etc.) to be performed at the server edge in order to provide users with individual, differentiated services. This requires advanced packet processing functions running at high performance in a virtualized environment.
A third challenge for virtual switches is that the growing number of VMs per blade makes high-bandwidth VM-to-VM communication (VM2VM) mandatory. However the standard hypervisor virtual switch that allows communications between VMs limits VM-to-VM bandwidth and does not scale well.
Finally, in order to isolate and secure VM2VM in a multi-tenant architecture, VM2VM requires routing, firewalling, and load balancing services that extend beyond the basic Layer 2 features provided by a typical virtual switch.
Other challenges are faced by developers of data-center network appliances.
The data-center aggregation layer, comprising functions such as ADCs, firewalls, IPSs, and UTMs, is migrating from physical implementations to virtualized architectures, instantiated on commodity hardware platforms similar to the application servers in order to achieve minimum cost and maximum flexibility. These virtual network appliances require a very high performance, rich networking software layer able to distribute high bandwidth network traffic to a large number of endpoints (application servers and appliance VMs) with very low latency.
In addition, a single instance of a virtual network appliance must be fully scalable across processors, blades, and racks, avoiding the cost and complexity of instantiating multiple virtual appliances as bandwidth requirements increase. Of course, it is critical that the solutions selected to solve these virtual switch and virtual network appliance challenges be fully compatible with emerging standards such as OpenFlow® and OpenStack, while also providing Carrier Grade reliability for enterprise-class data centers.
The networking data plane is the basic foundation on which advanced data-center architectures are built and high-performance data-plane software is key to the overall throughput and latency of the data center. Well-designed solutions interface efficiently and synchronize seamlessly with advanced control plane architectures, providing operators with the full benefits of an SDN implementation.
--Eric Carmès, CEO of 6WIND
Eric Carmès is a Founder and the Chief Executive Officer of 6WIND. In this capacity, he drives the corporate strategy and growth. Eric’s leadership in the design of innovative networking architectures has been instrumental in the company being awarded eight patents. With more than 20 years’ experience in IP standards and architectures, he is an expert in current and next-generation IP technologies and protocols.
