Data Flow, Packet Loss and Virtualization – Ethernet Summit 2012
February 22, 2012, San Jose, CA - This years ethernet summit focused on data flow in the network, virtualization and also delays/latency/packet loss in the datapath. As the networking world shifts from 10GbE to 40G/100GbE backplanes and interconnects for servers, a number of the data integrity issues have to be addressed. The first day of the event was heavily focused on the PHY aspects and hardware that was to be used for the transistion and what to expect with the new 40G/100G equip. Day two focused on using ethernet as an interface to the cloud and also virtualization plans.
The opening keynote for day two was Larry Roberts of Roberts Consulting. Roberts, who was the principle architect of ARPANET, discussed several protocol changes for reducing delay and packet loss in ethernet networks. Todays North American network traffic is reported to have on the order of 20-25% bad packets being sent. The expansion of companies, people and ISPs using the network has also led to long delay times on TCP/IP. The keynote focused on a newly adopted (Oct 2011) protocol called ITV Q.3313. This flow control signaling method dramatically reduces the TCP/IP retransmission and request processes for both fixed and variable rate packets.
The Flow Signaling Aware (FSA) control reduces jitter for fixed rate bandwidth request. This jitter reduction fixes a number of issues with VOIP transmission and increase the SNR of the connection. The method also helps with satellite transmission and extended latency, in DARPA tests, to improve data throughput 30-50x for TCP/IP traffic.
The Q.3313 protocol also will be adding security to the data stream. It supports 21 levels of encryption in the flow all designed to be implemented based on minimizing delay. The protocol also supports UDP, IPSec and other handshake protocols once the full specification is adopted by 2013.
HP discussed software defined networks (SDN) and hybrid SDNs. Using OpenFlow as a basis, they discussed the implementation of a system when a common cloud network can be easily reconfigured at the software level. The reconfiguration allows for the virtualized servers to dynamically reassigned and reconfigured for multi-tenancy, operational automation, disater recovery, backup, and a convergence of location for applications. This is facilitated by the logical breakout of the control plane from the data plane in the implementation. This cloud centric architecture adds some complexity for small configurations, but it provides a great deal of flexibility and rapid test and adoption time.
The main theme was virtualization for the data center and for ethernet connections. The use of virtualization has surpassed the 50% point for network connections and is now the default solution. This allows the hardware to be utilized at much higher rates to help offset the operating cost for hardware. The operating state of 10GE being the current basis of hardware and multiple VMs being used for applications in this space, is what is driving the aggregation of 10GE into 40GE ports and networks. The overall discussions for the event was that the migration is being optimzed for both high bandwidth light data (web and app based access) which wants the 40GE front end and deep computation which is happy on the traditional 10GE environments.
Article created by Pallab Chatterjee & Norm Towson.