How to Replace Multiple Legacy Software With a Telecommunications Billing System?

Billing in Telecom comprises several processes, interacts with different business operation software and requests data from other business modules.

The implementation and replacement of legacy billing systems were divided into several stages to ensure uninterrupted operational activities.

1. The legacy billing software and hardware used to calculate fixed net services were replaced. The legacy system ran on IBM AS/400.
2. To solve the problem of unreadable data and conversion between various formats Product Catalogue was also reorganized.
3. Manual calculations for some repeated business processes were replaced with automation, such as billing of corporate customers.
4. In case a Telecom operator needs to process a unique business case manually, an implemented billing system ensures convenient usage of reference data and offers an opportunity to generate an invoice by a template.
5. A delivered billing system was integrated with systems capturing and rating Internet and mobile services usage.

The built system is scalable and multithreaded. Due to the usage of virtual machines, it could expand almost endlessly. To avoid deadlocks, threads are designed to run independently. As far as calculation flows may significantly differ by the amount of the data to process, we’ve implemented the algorithm of balancing flows for bill cycles. The built billing system runs on 6 streams, which fits a usual operative bill cycle with 300-500 thousand residential accounts. In terms of billing, one of the most resource-consuming operations is downloading reference data for a billing cycle from the database and uploading results back. Multithreaded architecture has also helped to decrease time and resources for these processes, as they are now split into threads. On average, for a standard large bill cycle, each of these processes now takes about 1.5 hours. The calculation process itself is relatively fast and takes about 20 minutes for an average billing cycle. Bill formatting is also a resource-consuming process, as far as the billing system generates large XML files (about 350 GB each).

We offer to refer to the following data for one standard bill cycle processing:

• Generation creation — 15 minutes
• Download reference data from a database — 10 minutes per 100 million of records
• Usage capturing — 6 minutes per 7 million of records
• Calculation — 10 minutes per 400 thousand of accounts
• Formatting — 12 minutes per 400 thousand of resulting invoices
• Upload of the results — 12 minutes.

A linear approach to computing events and processing every event one-by-one slows the speed of operation significantly. Concurrent computation is much more effective. That is why to improve the efficiency of using computing power, the calculation logic for a Telecom rating system is written using parallel / multithreaded algorithms .

The approach allows upscaling the calculation speed almost linearly (depending on the configuration and composition of the input data). According to the requirements, the expected speed of event processing can be calculated as follows: 80,000,000/10/60/60 = 2,222 CDRs/sec. The calculation means the system must process at least 2,222 events (CDR, Call Detail Records) per second to handle the full scope of events within 10 hours.

To speed up the operation, the event service was divided into three parts:

• the logic for downloading reference data from the database to local files (ProxyRefresh),
• the calculation logic that reads events from files (Rating) and
• the logic for loading billed data into the database (Loading).

When implementing a single-threaded logic and direct work with the database, the calculation speed of this configuration does not exceed 500 CDRs/sec, which is four-times less than required. We have used multisequencing and decreased the number of synchronizations to a minimum. Such an approach to the development allows using the full power of computing resources and minimizing time losses on waiting for the necessary data from other modules. To boost system performance, a modular architecture was used for the implementation of the computation logic. The benefit of a modular architecture is the ability to group computational modules and balance the operational load between these groups to ensure better performance and utilization of computing power.