Managing the network and data traffic of a major Internet service provider in the Czech Republic is not an easy task; it’s based on real data stored in real time. The volumes of transmitted data are enormous – a perfect use case for Big Data technologies. But what if you don’t have any experience with these?
Solution to a problem
This is precisely what the Internet service provider realized, thus facing a big challenge. The company opened a tender to find a solution for storing large volumes of metadata about network traffic, placing a strong emphasis on the new solution’s smooth operation and high accessibility. The ISP approached previous and new suppliers, including Adastra, which has extensive experience in Big Data from a wide range of projects implemented successfully in various sectors.
In 3 months, Adastra built the customer a versatile Big Data platform that is both highly accessible and easy to extend.
Business description of the solution
Adastra came up with a purely generic data platform that delivers excellent computing power and whose disk capacity can easily be extended according to the client’s future needs. Taking operating costs into account, Adastra suggested a smaller cluster, which fully complies with current requirements, provides high computing power, and has sufficient storage capacity. The solution delivers close to 1PB (petabyte) of space, processes 300 compute threads, has 2.5TB of RAM, and uses Hortonworks distribution. The network traffic metadata is processed on the Spark framework, which uses stable technologies such as Apache Kafka, Apache Hadoop and Apache HBase.
The new Big Data platform processes all the data related to the client’s Internet traffic. It enables flexible cluster-resource allocation depending on the required data flow. Real data flows are in the tens of billions every day. Spark will calculate the basic daily aggregation within 12 minutes when allocating 100 compute threads.
The solution also includes a third-party component for the high-performance conversion of network metadata from probes, which was successfully tested on the client’s system during the PoC (proof of concept).
Adastra’s Big Data platform processes tens of billions of records daily.
The installation and implementation of the Hadoop platform took 3 months, during which time testing was also carried out. Adastra’s suggestion worked well from the beginning, so there was no need to make any significant changes, and the new Big Data platform was put into operation a month earlier than originally planned.
Thanks to problem-free testing, Adastra completed the project a month before the contractual deadline.
According to the client, the key benefits of the Big Data platform are:
- Maximum performance – the client can handle yearly increases in data volume simply by adding several additional cores in the streaming application on YARN. Increasing the number of cores makes it possible for storage flow to reach close to 1 million records per second. Thus, the client can flexibly respond to current needs in terms of both data traffic and computing power required for Advanced Analytics and connected Machine Learning.
- Versatility – the solution isn’t tied to a specific hardware manufacturer or server type. The cluster can be supplemented by any type of server depending on availability or by specialized servers with GPU for accelerating Machine Learning algorithms.
- 24/7 support – Adastra provides 24/7 support for operations and troubleshooting.
- Reporting – the Big Data platform is an excellent basis for follow-up activities and development: a complete reporting layer can effectively be built on top of the stored data, and information can be accessible and visualized for end users.
Approximate throughput during complete data processing, including storage, is approximately 600,000 records per second when 80 cores are allocated for the application (about 28% utilization of cluster resources).
Are you interested in a similar solution? Contact us.
We will contact you as soon as possible.