Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
ACCELERATING DATA ANALYTICS
Xelera Technologies
Felix Winterstein
O U T L I N E
18.12.2018 2
• Xelera Technologies
• Use case 1 – intelligent voice recognition
• Use case 2 – predictive analytics acceleration
• Xelera’s product portfolio
X E L E R A T E C H N O L O G I E S
18.12.2018 3
ON-PREM / NEAR-EDGE• Private hosting
• Real-time
• Latency-critical
Edge
CLOUD• Big Data processing
• Data warehousing
• …
LAN/WAN
Internet
O U T L I N E
18.12.2018 4
• Xelera Technologies
• Use case 1 – intelligent voice recognition
• Use case 2 – predictive analytics acceleration
• Xelera’s product portfolio
N E A R - E D G E A C C E L E R A T I O N S E R V I C E S
18.12.2018 5
Compute server
(near-edge)
Low latency
connection
Edge device
(limited compute
capabilities)
I N T E L L I G E N T V O I C E R E C O G N I T I O N
18.12.2018 6
Speech segmentation
text-independent
speaker identification
60 ms
e2e latency constraint
Classification / clustering
Ethernet
Feature extraction(Deep Learning)
I N T E L L I G E N T V O I C E R E C O G N I T I O N
18.12.2018 7
CPU
CPU+FPGA
Latency
[ms]
O U T L I N E
18.12.2018 8
• Xelera Technologies
• Use case 1 – intelligent voice recognition
• Use case 2 – predictive analytics acceleration
• Xelera’s product portfolio
A C C E L E R A T E D A N A L Y T I C S
18.12.2018 9
• Popular Big Data software framework
• Open source; strong commercial use
• Business intelligence
• Data warehousing
• Real-time / streaming apps
• Recommender systems
• Log processing
• Fraud detection
• In-memory, distributed
• Combines real-time capabilities and
advanced analytics
Spark
User app
Spark Streaming
SparkMLLib
GraphX
SparkSQL
Xelera middleware
FPGA accelerator
On-premise
Libraries
Standard
Spark
Spark
Acceleration
A C C E L E R A T E D A N A L Y T I C S
18.12.2018 10
Training data
Training the machine learning
model (slow)
Model
Input data
(current data)
Machine
learning
inference
Predict
Use case: Real-time predictive analytics
• Target: use existing data to make
predictions about a system
• Algorithm: Spark MLLib’s Random Forest
• Applied in:
• Banking / Finance
• Healthcare
• Ecommerce
• High-frequency model updates as BI
applications move to real-time
R A N D O M F O R E S T A C C E L E R A T I O N
18.12.2018 11
Vote
Tree 2Tree 1
kernel 0
kernel 1
shell
R A N D O M F O R E S T U S E C A S E
18.12.2018 12
Processing time
[minutes]
41x
speed-up
57x
speed-up
67x
speed-up
Work load size
38.5 min
0.9 min
93.7 min
1.6 min
202.4 min
2.9 min
CPUFPGA
O U T L I N E
18.12.2018 13
• Xelera Technologies
• Use case 1 – intelligent voice recognition
• Use case 2 – predictive analytics acceleration
• Xelera’s product portfolio
F R O M U S E C A S E S T O A N A C C E L E R A T I O N S T A C K
18.12.2018 14
Platform x Functions x Integration
Use case 2Use case 1
X E L E R A T E C H N O L O G I E S
18.12.2018 15
Random
Forest
XG
Boost
Cluster-
ing
Regress-
ionSVM
Deep
Learning
Fuzzy
Search
Com-
pression
Java/
C++
Proprietary
SW interfaces
Network
protocols
Integration
hooks
End user
application
Plugins
Platforms On-premise
B A C K U P
18.12.2018 17
• Performance acceleration for all business critical
processes (value contributors)
• Multiple processor architecture Application- and load-based switching
• Seamless integration into business processes,
analytics, networks, storage and data bases
with
• Cockpit and dashboards for control center,
online monitoring and reports
• Significant energy savings – lower TCO
• Smooth transition – no disruption
X E L E R A ‘ S R E A L T I M E A C C E L E R A T I O N P L A T F O R M 2 0 2 0
18.12.2018 18
Virtualization
On-Premise
Cloud
Edge ComputingIoT
5G
P R O D U C T F A M I L Y
18.12.2018 19
A powerful solution portfolio aligned with the market maturity and expectations
Predict
Pre-purchase assessment
SuiteSeamless stack integration
PluginsPlug-and-Play Acceleration
Assess performance gains, costs and savings
when migrating business applications to FPGAs
Connect business applications with FPGA
technology (baseline infrastructure)
Accelerate complex data analytics applications
Accelerate performance-critical operations in
databases
Handle large-scale IoT infrastructure with FPGAs
at the network edge