scale, speed and scope: why telcos are turning to gpu-powered analytics

25
Jim McHugh, VP & GM of NVIDIA Todd Mostak, CEO of MapD Scale, Speed and Scope: Why Telcos are Turning to GPU- Powered Analytics February 2017

Upload: nvidia

Post on 12-Apr-2017

259 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Page 1: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

Jim McHugh, VP & GM of NVIDIATodd Mostak, CEO of MapD

Scale, Speed and Scope: Why Telcos are Turning to GPU-Powered Analytics

February 2017

Page 2: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

2 2

AGENDA

The rapidly evolving telco business model and its implications

Why current compute models are struggling

The path forward: GPUs and GPU-powered analytics

How to enable speed at scale

Case Studies: How GPUs go to market in the service provider world

Page 3: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

3

The Shifting Landscape for Service Providers

GROWTH OF BUSINESS SERVICES

NETWORK CONVERGENCE

CONTINUED CONSOLIDATION

DATA EXPLOSION,NOT REVENUE

Page 4: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

4

Not Just Data Volume, Data Types

Satisfaction: network, call,

service

Usage:data, browser, type,

time spent, text

Network: speed, latency, signal

strength, network type

Location:GPS, wifi estimation,

accelerometer

Customer:Age, family size, gender,

brand preference, behavior

Hardware:Handset, chipset, RAM, screen size, SIM card,

software version

Page 5: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

5

Driving Complex, Value-Laden Use Cases

Rationalize and prioritize infrastructure investment

Customer Experience Management (Customer 360)

Operational Analytics

Network Optimization

Data monetization

Page 6: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

6

Issuing iterative queries becomes wearisome.

As little as 500ms reduces interaction and limits the

amount of data covered.2

Analyst creativity is impaired.

For large scale data problems, potential avenues of

exploration are ignored because the time cost is too

high to even consider.3

Slow Compute – The Bottleneck

Long response timeconstrains questions asked.

Over time this behavior hardens.1

1. http://engineroom.ft.com/2016/04/04/a-faster-ft-cåom/ 2. http://go.mapd.com/rs/116-GLR-105/images/2014-Latency-InfoVis.pdf 3. https://www.microsoft.com/en-us/research/publication/trust-me-im-partially-right-incremental-visualization-lets-analysts-explore-large-datasets-faster/

Page 7: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

7

Pre-aggregation struggles at scale

Scale out on CPU infrastructure has

tremendous hidden costs

Sampling misses the whole picture

Workarounds Create Additional Problems

EXPLORE THE OUTLIERS

AND LONG-TAIL EVENTS

RELY ON ACCURATE DATA

SCALE WITH A ROI

Page 8: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

8

2008 2009 2010 2011 2012 2013 2014 20160.0

1.0

2.0

3.0

4.0

5.0

6.0

NVIDIA GPU x86 CPU

TFLO

PS

M2090

M1060

K20

K80

K40

Fast GPU+

Strong CPU

P100

The GPU Accelerated Data Center

Page 9: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

9

GPU Accelerated Analytics

Accelerated analytics, everywhere, every platform

TESLAServers in every shape and size

DGX-1The accelerated

analytics supercomputer for instant productivity

CLOUDEverywhere

Page 10: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

10

NVIDIA DGX-1

Accelerated analytics supercomputer-in-a-box

170 TFLOPS | 8x Tesla P100 16GB | NVLink Hybrid Cube Mesh2x Xeon | 8 TB RAID 0 | Quad IB 100Gbps, Dual 10GbE | 3U — 3200W

Page 11: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

11

NVIDIA and MapD for Accelerated Analytics

IMMERSIVE VISUALIZATIONPETABYTE SCALEUNPARALLELED SPEED

Explore and discover insights in milliseconds with world’s fastest data exploration platform

Dynamically interact and visualize billions of data points in milliseconds

Instantaneously visualize and query multi-billion row datasets across multiple high density nodes

Page 12: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

12

MapD: software optimized for the fastest hardware

SPEED OF THOUGHT VISUALIZATION100X FASTER QUERIES

MapD Core

An in-memory, relational, column store database powered by GPUs

MapD Immerse

A visual analytics engine that leverages the speed + rendering

capabilities of MapD Core

+

Page 13: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

13

Proof Points

Noted DB blogger, Mark Litwintschik has benchmarked MapD vs. major CPU systems and found it to be between 74x to 3,500x faster than CPU-powered databases.

Page 14: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

14

Data Lake/Data Warehouse/SOR

Performance Starts with Memory Management

SSD or NVRAM STORAGE (L3)250GB to 20TB1-2 GB/sec

CPU RAM (L2)32GB to 3TB70-120 GB/sec

GPU RAM (L1)24GB to 384GB3000-5000 GB/sec

Hot Data Speedup = 1500x to 5000xOver Cold Data

Warm DataSpeedup = 35x to 120xOver Cold Data

Cold Data

COMPUTELAYER

STORAGELAYER

SP

EE

D IN

CR

EA

SE

S SIZE

INC

RE

AS

ES

Page 15: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

15

Purpose Built + Highly Optimized

Query Compilation Engine creates one custom function that runs at speeds approaching hand-written functions. LLVM enables generic targeting of different architectures + run simultaneously on CPU/GPU.

Page 16: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

16

Purpose Built + Highly Optimized

Backend Rendering — Data goes from compute (CUDA) to graphics (OpenGL) pipeline without copy and comes back as compressed PNG (~100 KB) rather than raw data (> 1GB).

Page 17: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

17

Purpose Built + Highly Optimized

Streaming — Speed eliminates need to pre-index or aggregate data. Compute resides on GPUs freeing CPUs to parse + ingest. Finally, newest data can be combined with billions of rows of “near historical” data.

Page 18: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

18

Where MapD Sits

Page 19: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

19

Verizon

IMPACTCHALLENGE

Over the air (OTA) technology is the primary way wireless companies manage subscribers (via SIM cards). OTA polling + pushes create massive data files.

Verizon’s legacy CPU powered database did not allow real-time queries – so they down sampled to reduce time…. but they sensed they were not getting the whole picture.

The down sampling required by CPU-era solutions was missing key outliers.

Finding those outliers was worth millions.

Additionally, ease of use drove higher utilization thus more informed decision-making.

All at a fraction of the cost.

Using MapD’s GPU-powered database + visual analytics engine, Verizon was able to execute queries against the entire SIM card population.

Further, Verizon was able to query + visualize streaming data + near historical data – for the entire country or an individual card.

SOLUTION

Page 20: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

20

Major US Cable Operator

IMPACTCHALLENGE

The business services team was stuck with hardware and software that only enabled them to look at 1x1 mile blocks.

Each new block required long wait times — taking minutes to load.

As a result, there was no adjacent discovery and they struggled to optimize marketing around capabilities and infrastructure.

The solution has completely altered the company’s approach for business services, resulting in operational efficiencies (discovered some contractors “inspecting” the same property 10+ times), targeting marketing more effectively (based on capacity utilization based marketing) and campaign analytics.

Using GPU-powered visual analytics from MapD and NVIDIA the operator was able to see their entire footprint – eliminating the need to go section by section.

Furthermore, they retained full grain level detail on every customer for when they zoomed into a building or residence.

SOLUTION

Page 21: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

21

Major Wireless Provider

IMPACTCHALLENGE

Client subscribed to third party performance data to prioritize what to upgrade to create better coverage map claims.

Their current CPU-era infrastructure only allowed them to see 3% of the data in any given region.

Using less HW the telco was able to determine, instantaneously, what infrastructure projects would yield the best ROI from a coverage map perspective – while improving customer experience and reducing dropped calls.

Using NVIDIA GPUs and MapD the telco was able to see and interact with their national footprint – not a neighborhood.

Zoom, cross-filter etc. all work in real time.

SOLUTION

Page 22: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

22

DEMONSTRATION

Page 23: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

23

For More Information

/ Twitter: @NVIDIADC, @JimMcHugh/ DGX for Accelerated Analytics:

www.nvidia.com/analytics/ DGX for Deep Learning:

www.nvidia.com/dgx1

/ Twitter: @MapD, @ToddMostak/ Product Overview:

www.mapd.com/products / Demos: www.mapd.com/demos

Page 24: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

24

May 8 - 11, 2017 | Silicon Valley | #GTC17www.gputechconf.com

CONNECTConnect with technology experts from NVIDIA and other leading organizations

LEARNGain insight and valuable hands-on training through hundreds of sessions and research posters

DISCOVERSee how GPUs are creating amazing breakthroughs in important fields such as deep learning and AI

INNOVATEHear about disruptive innovations from startups

Don’t miss the world’s most important event for GPU developers May 8 – 11, 2017 in Silicon Valley: MapD in booth #621

SAVE ADDITIONAL 20% OFF REGULAR RATES AT WWW.GPUTECHCONF.COM

Page 25: Scale, Speed and Scope: Why Telcos Are Turning to GPU-Powered Analytics

Thank You