x
NFV Elements

Telefónica Proves Brocade Router Performs for NFV

Tests run by Telefónica in its labs show Brocade's software-based router can achieve 80Gbit/s throughput, matching the performance levels required for carrier applications and setting a benchmark that supports NFV deployment. (See Telefónica, Brocade Team Up on NFV Benchmarks.)

Brocade Communications Systems Inc. (Nasdaq: BRCD) provided Telefónica SA (NYSE: TEF) with a thumb drive containing the latest iteration of its Vyatta 5600 vRouter, and the carrier deployed that on a commercial off-the-shelf Intel Corp. (Nasdaq: INTC)-based x86 server within a Red Hat KVM environment. Deployed as a single virtual machine, the Vyatta 5600 was able to support all of the server's available ports at line rate.

"In less than two hours, we deployed the Brocade Vyatta 5600 vRouter from a memory stick and completed our performance tests in our NFV Reference Lab," notes Francisco-Javier Ramon, head of Telefónica's NFV Reference Lab, in the press release. "These results are allowing us, as network operators, to aggressively change our perspective regarding what is possible with software-driven networking in order to accelerate the adoption and deployment of these revolutionary technologies."

By hitting the 80Gbit/s mark, Brocade actually exceeded its own original goal, which was to prove that a software router can support the 10Gbit/s performance that is mainstream in carrier environments, says Andrew Coward, VP of service provider strategy at Brocade.


Dig deeper into how telecom service providers are evolving their data center strategies at our NFV & the Data Center event.


"The performance of software networking products has been pretty abysmal," Coward concedes. "It's been less than one gig -- sometimes more like a couple hundred megs. For carrier-type applications, it became really important to have a much better performance, otherwise there is a significant disconnect between the 10-gig interfaces on most routers today and the software networking type product."

By over-delivering on this promise, Brocade believes it has created a software router that makes the price-performance curve look far more attractive for NFV, and enables a server to not only fill the 10Gbit/s pipes widely in use today, but also have processing power left over for applications and other network functions, Coward notes.

Intel's Xeon processor-based servers and its Data Plane Development Kit were key to the performance improvements, and Brocade rewrote its architecture around those improvements with drivers that make the most of the Intel process, Coward says. The company provided the new software router to a number of its service provider customers for testing.

Telefónica was clearly pleased enough with the test results to announce them publicly -- an unusual move these days for many telecom service providers.

Coward says the Brocade Vyatta 5600 vRouter is in about 40 proofs-of-concept currently, but the next big step will be determining functions or applications that can go beyond the single-use test and be more repeatable.

— Carol Wilson, Editor-at-Large, Light Reading

schlettie 8/20/2014 | 10:51:04 PM
Table sizes How large a FIB (# of route entries) did they test?  How many ACL entries?  Testing with small tables that fit in L1/L2 cache is not realistic for production deployment.
dwx 8/20/2014 | 10:20:31 PM
Re: Which Functions? My understanding from what I've read is this was just pure packet switching.  
cnwedit 8/20/2014 | 1:54:36 PM
Re: Which Functions? It's my understanding they have virtualized other functions, but those weren't part of this test. One of the things Brocade told me they hope to do is get a set of functions that can be virtualized and used at multiple telecom carriers. Today, they are in 40 proofs of concept but they tend to be company specific. 
Atlantis-dude 8/20/2014 | 1:49:02 PM
Which Functions? Is this just pkt switching? Any other functions that were virtualized?
dwx 8/20/2014 | 12:11:16 PM
Re: Baked-in hardware dependency - interim fix or NFV trend? It is all software, but obviously takes advantage of modern Intel NICs.   There is a wealth of information available on Intel's DPDK which is really really key to achieving those kind of throughput numbers.   I have no affiliation with either company but here is a joint whitepaper from Wind River and Intel describing the changes made to achieve the higher throughput:  

http://embedded.communities.intel.com/servlet/JiveServlet/previewBody/7070-102-1-2281/7785_PerformanceDesignOptions_WP_1111.pdf

They make the handling of packets on the server look much like the handling of packets on an old school software router, but of course processor speeds are much higher now.  

I'd be interested to see details regarding the 80Gbps number and packet sizes used.  Most of the high throughput numbers are based on large frames >1500 bytes since it's the PPS which kills the throughput.   Also high touch things like ACLs, NAT, or other packet mangling usually drastically reduces throughput versus a router with dedicated hardware to perform those tasks.  
slioch 8/20/2014 | 11:42:32 AM
Re: Baked-in hardware dependency - interim fix or NFV trend? It's all software
Yulot 8/20/2014 | 9:51:16 AM
Re: Baked-in hardware dependency - interim fix or NFV trend? Hi Johnwilmes. Is it a mix of SW and network processor HW?? That sounds more right to reach this type of throughput, because SW only architecture usually drops packet all over above a Gig.
johnwilmes_STG 8/19/2014 | 8:04:56 PM
Baked-in hardware dependency - interim fix or NFV trend? "Intel's Xeon processor-based servers and its Data Plane Development Kit were key to the performance improvements, and Brocade rewrote its architecture around those improvements with drivers that make the most of the Intel process ..."

 
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE