SQream Platform
GPU Powered Data & Analytics Acceleration
Enterprise (Private Deployment) SQL on GPU for Large & Complex Queries
Public Cloud (GCP, AWS) GPU Powered Data Lakehouse
No Code Data Solution for Small & Medium Business
By Inbal Aharoni
Every telecom operator wants its users to stay connected. Planning a mobile network so it can maintain access for millions of customers is a great challenge for operators. Mobile network coverage calculation requires a great deal of manual work, which delays improvement and slows down implementation in network quality of service.
SQream was challenged with this use case by a European Tier-0 telecom customer and had to demonstrate the ability to calculate all existing and potential cells, collect all relevant GEO information, and select the existing and potential cells for network planning scenarios.
SQream (currently running only on private cloud), Google BigQuery, Amazon Redshift, Azure Synapse Analytics, Snowflake.
The use case ran over a ~11.5TB dataset, containing 4 tables and representing the data being collected from each cell’s coordinates in Germany.
The main consideration for customizing the hardware stack for each one of the competing vendors was the right balance between cost and performance. We took into account each vendor’s recommendation depending on the size of the chosen dataset (11.5TB) and maintained an equal number of nodes for all participants.
Environment
Configuration
Compute cost (hour)
Storage cost (TB)
Amazon Redshift
AWS
8X ra3.4xlarge
$26.08
$24
Snowflake
Large
$16.00
$40 (on-demand)
SQream
8X g4dn.8xlarge
$17.4
$23
Google BigQuery
GCP
Flat-rate 400 slots
$20
$46 (on-demand)
4x nl-standard-32 (with additional 2x GPU each)
$16.88
Azure Synapse Analytics
Azure
DW1500c (Dedicated SQL pool)
$18.00
8X Standard_NC16as_T4_v3
$9.6
$21
After configuring the chosen cloud environment for the use case, we were ready to begin. First, we had to ingest the 11.5TB into the analytical platform. Second, we started running the 8 queries designed for data processing, which generate the raster and cell polygon in the required resolution. Third, we analyzed the data to create the optimized network planning scenarios for the operator, using 5 queries.
As we were running the use case, we focused on two metrics for comparison:
The results revealed several performance differentiators between the competing platforms. Overall, in all cloud environments, SQream presented the best TTTI, between X1.8 to X3.2 faster. As for the average execution time of the 13 queries, SQream presented between 1.1X to 2.25X faster results (862 seconds on GCP, 1038 seconds on AWS, and 1993 on Azure).
Even though the compute cost of machines with GPUs (which is SQream’s case) is usually much higher, the outstanding performance of SQream during the field test staging showed it also to be the most cost-effective option.
Learn more about how SQream performed in other benchmarks, such as TPC-H (10TB), TPCx-BB (300TB and 30TB), and other industry use cases.