Upload
arnon-shimoni
View
163
Download
0
Embed Size (px)
Citation preview
THE FASTEST BIG DATA ANALYTICS
Arnon ShimoniBig Data Solutions Architect
March 2016
700,000 – 1,300,000 points per second
Autonomous car sensors generate a lot of data
4x the sensors means millions of transactions every second
How do you handle that kind of data?
(Without storing it for later analysis)
The question is…
Stick it in Oracle? Exasol? SAP HANA?
Will they fit in a car?
Perhaps you can handle it with Hadoop?
10,000 machines is... A lot... Of power, heat, floorspace, maintenance
How they do it How we do it
Super fast analytics with SQream DB software
inside commodity server
A GPU-Based SQL Database
100x faster results in a small footprint
Designed from the
ground up around GPU capabilities
Easily analyseup to petabytes in a standard 2U enclosure
On the fly fast
compression
Extreme scalability
Very Fast
Faster querying = more questions answered per hour
SQream was 37% faster thanTeradata, at Orange Telecom
when analyzing 4.3 billion CDRs, for 30 million subscribers
SQream was 16x faster thanMongoDB at a Homeland Security company
when querying 1.6 billion geospatial entitiespoint in polygon, distance calculation, line-crossings
SQream Teradata
SQream MongoDB
VS
VS
Scale in, up, and out
SQream DB scales easily both in and out
Add more GPUs to add more compute power
Add more nodes when you need higher availability
Size
The first of it’s kind low-power, embedded analytics platform
SQream can also run on an embedded platform with an Nvidia GPU like the Jeston TX1 or the Drive platform
which is certified for automotive use
NVIDIA Tegra X164 bit ARMv8
256 CUDA coresprocessing power
Multiple LIDAR / radar sensors
Ultrasonic range sensors
Cameras
Radar rangefinders
Other CANBUS sensors, GPS, …
A big data supercomputer in the car
• Available as Cloud (on SoftLayer )or on-premise solution
• Pay per TB, not per compute or traffic
• SQream’s no hassle design will haveyou up and running quickly
110,000,000 entries representing 4 years of data
Seamless integration
Most common tools supported out-of-the-box
SQream DB server SQream DB Compiler
SQL parser
Optimizations
Parallel query graph plan
GPU/CPU Runtime
Columnar storage engine
GPU CUDA kernels for physical operators
SQL Query Data set
Query queue manager
connectors
JDBC ODBC ADO.NET Python Driver CLI client
Filesystemext4/XFS/ZFS
SQream Technologies Ltd - Confidential
Use SQream to analyze your telemetry
110,000,000 entries representing 4 years of data