Session Border Control in the Cloud Accelerating Virtual Network - - PowerPoint PPT Presentation

Session Border Control in the Cloud

Accelerating Virtual Network Functions with GPUs

Kevin Riley –CTO & EVP of Advanced R&D

2 Ribbon Communications Confidential and Proprietary

Our Application and the GPU Opportunity

SBCs…What are They?

Secure and Interwork Unified Communications Deployed in Service Provider Core, Edge and Customer Premise Application Decomposes into Control and Media Plane Transcoding Inefficiencies Inhibiting Cloud Migration at Scale Enhanced Security Capabilities Ill-suited to CPU

Challenges Evolution

Historically Implemented on Purpose Built HW Migration to CPU and Cloud Infrastructure is current State of the Art

Control Plane Media Transcoding Media Security & Forwarding

SBC Application Components

A Break-Out Strategy is Needed for Cloud SBC

How to Unlock Cloud Performance On-Par with Purpose Built Hardware? How to Unlock Cloud Performance On-Par with Purpose Built Hardware?

4 Ribbon Communications Confidential and Proprietary

Observation 1: GPUs are Pervasive in the Data Center

5 Ribbon Communications Confidential and Proprietary

Observation 2: GPU Performance is Gapping CPU

GB/s

Peak Memory Bandwidth

Source: Nvidia’s Presentation

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

X86 CPU NVIDIA GPU

GFLOPS 8000 7000 6000 5000 4000 3000 2000 1000 1400 1200 1000 800 600 400 200

Peak Double Precision FLOPS

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

X86 CPU NVIDIA GPU

Market Realist

6 Ribbon Communications Confidential and Proprietary

Mapping GPUs to the SBC Application

Challenges

• Speech codecs generally employ

various types of recursive filters which are ill-suited for parallelization.

• Even if we parallelize parts of the
• peration, over all speed-up will

be limited by Amdahl’s law.

• A GPU core is relatively less

powerful than a CPU core.

For a stable transcoding system, it’s imperative that processing of all channels is completed within the codec frame time. CPUs and DSPs process channels sequentially and hence need to ensure per channel processing time is low.

TIME Channel N Channel 0

Codec Frame Time

7 Ribbon Communications Confidential and Proprietary

Mapping GPUs to the SBC Application (cont.)

New Approach

• GPU cores are less powerful, but they are plentiful.
• We get better performance when adjacent GPU threads perform

similar jobs.

• Offload entire encode/decode operation for a single channel
• n to a single GPU thread and ensure processing time is

less than frame time. G729A transcoding (encode + decode), with a 10ms frame-time, takes approximately 35us for one channel on an E5-2690v2 processor. On a CPU we can achieve approximately 285 transcodes. When we offload per channel processing to a single GTX970 thread, it takes approximately 6ms (initial prototype). However in this 6ms we can process 1664 channels (GTX970 has 1664 cores).

Channel 0 Channel 1 Channel N

TRADITIONAL APPROACH NEW APPROACH

TIME Channel N Channel 0

Codec Frame Time

8 Ribbon Communications Confidential and Proprietary

• High scale on COTS

hardware

• Codecs ported and
• ptimized in-house using

reference source code.

• Virtualization on COTS

hardware

• Codecs sourced from Intel

IPP and third-party vendors.

• High scale on customized

hardware.

• Codecs sourced from

third-party vendors.

Observation 3: GPUs are the Ideally Suited for SBC Media

SBC Stall Point

9 Ribbon Communications Confidential and Proprietary

GPU Technology Delivers Disruptive Performance Gains

Transcoding Compared to CPU

9x

Transcoding Compared to DSP

3.5x

Power Consumption

<2x

10 Ribbon Communications Confidential and Proprietary

GPU vs CPU

518% 1136% 534% 320% 407% 605% 1458% 1066% 519% 732% G729A EVRC-9.3 EVRCB-9.3 AMR-12.2 AMRW B-6.6

SESSIONS MULTIPLIER

GPU VS CPU - SESSIONS

M60 V-100

193% 333% 133% 81% 111% 356% 543% 314% 172% 209% G729A EVRC-9.3 EVRCB-9.3 AMR-12.2 AMRW B-6.6

SESSIONS /WATT

GPU VS CPU -SESSIONS/WATT

M-60 V-100

11 Ribbon Communications Confidential and Proprietary

Ribbon & Nvidia Thought Leadership with Transcoding on GPUs

• Nvidia GPUs provide massive parallel processing for calculation-intensive tasks,

perfectly suited for simultaneous media transcoding for multiple codec types

• Nvidia is delivering improved performance with reduced power requirements at

unmatched velocity

• Ribbon software framework leverages CPU for media handling and overhead,

with transcoding moved to GPU

Continuing to Disrupt with GPU

Media Transcoding Media Security & Forwarding

Co-Processor Model

• Media Transcoding Ideally Maps
• Most Compute Intensive Component

Front-End Processor Model

• Direct NIC/GPU Memory Copy Unlocks Further Disruption
• Enables Further SBC Media Plane Acceleration
• Unlocks Enhanced DPI/Pattern Matching Security Capability

Control Plane