Client had a lot of requirements for new version of the system, including hardware and software parts. New version should be much more productive and have possibilities with deployment to remote offices, real time and off line modes with remote buildings. Remote building could be located even in different cities, but all system should work as one unit.

We had different challenges with data interaction speed - all system for all online modules should work in real time. In security system domain package delivery and reactions during a second is equal to ages. Reaction to the hardware events should take milliseconds, even if some door is located 100 km far away from the central database. We can have dozen offices, each of them can have 10-50 controllers, and each controller can have 1-4 doors and 1-4 locks and readers (card, biometric, chip, keypad, etc.).

To meet the customer’s requirements our team prepared couple system architecture designs and then developed prove of concept (POC) versions for each architecture design. After performing the measurement's (speed, response time, emulation of specified amount of clients, sensors, signal sources, etc.) we selected the better solution and added some improvements. After we got the skeleton of the future solution we started split the whole system by subsystems. Of course, we use previous best practices from customer's existing solutions and our team experience.

Access control system has a lot of objects, data entities and sources of the signals, including controllers, repeaters, concentrators, converters, doors, locks, readers and so forth. During a working day system, even with 100-200 controllers can generate up to 500-1000 events per second. Of course, we cannot implement "classic" architecture for such amount of information, when we just have single UART/USB port and one hardware driver, which will write data on the fly directly to DB - we just won't have such available computing resources.

For solving all challenges, listed above, we designed multi-level dedicated data interaction system which works as one unit. All data flows were divided by their importance level and affects to the system logic. System has memory cache subsystems, priority queues and other logical blocks.

On the architecture design stage we take a timeout and made a decision to perform a brainstorming and think about the future. What challenges we can face in 5 or even in 10 years from now? What technologies and trends will dictate the technological rules in the future? - That is the crucial question in case if investments in something new.

Finally we built the architecture, where we can build each subsystem from the blocks - logical and functional bricks. Using this concept we can insert to the logic to some level artificial intelligence blocks, which are not exist today, but can be developed in 2-3 years for now. Today system has preconfigured and custom configured rules and strategies for reactions on system events, but in the future we have plans of development self-learning blocks, which will detect "strange" system behaviors, check system health, detect business rules fail and so forth.

As we mentioned above, we cannot just write and read to database with 10 000 operations per second (imagine, that some queries can take megabytes) - we should use different approach. Classic systems can "DB via" data interaction model, which all modules "speak" with other modules via data in the DB. For instance, after some event from the sensor we inserted data to the DB, DB trigger generated event and notified client apps (operator's working place). In case of heavy loads such type of approach will not work - we will have signal delays, crashes because of memory problems, system overloads and so on.

This is why for the designed system we applied "hybrid" concept, when we use DB read/write operations only when it's really necessary. All modules can communicate between each other in real time using sockets and websockets, share memory and other technologies with data double backup protection (if some module, responsible for some data package sent it and crashed, other module will backup the same package). When data is ready for saving to DB - it will done once without DB overloading. Data can be aggregated, stored in the not core data storage (service local DBs) and so on.

Client's software part should be user-friendly, easy for use, modern and clean. Client part for previous version of software had not obvious structure and access to the system's features and functionality - and that's OK for the previous version, because customer's developers added new modules and features during 15 years - it's a great software even nowadays. In our case we already had list of modules, their features and requirements for the future extensions, for example, as it was mentioned before artificial intelligence modules for decision making and forecasting or something else.

We built modular system, where "host" core app can load necessary administrative and operation modules depending of requirements for the working place, who works with current client (users, roles, permissions), client's available functionality and so on. This includes (not all modules):

If you have tons of information - you should have easy way for accessing this information. For sure, each client's installation will have standard requirements for the data reports and representation, but maybe 30% - it will be custom requirements for data access or data export.

This is why we implemented solution, which combines classic reporting ideology (with possibility to define custom reports) and Business Intelligence (BI) system for better and modern data representation and control. Not all customer can afford to use and pay for standalone or cloud-based BI solutions, but some of them already use the systems as Power BI and connect connect to allowed data entrypoints in data storage.

Modern access control system cannot work in isolation (at least in 90%). Data, provided by access control system can be used in enterprise systems, such as ERP, CRM, HRM, Payroll and many other systems.

Pervious software version had a lot of custom developed integration modules, custom developed for each client. It was too hard to support such "zoo" of custom integrations, especially in case of system modernization and later system update for each client. Sometimes it was real nightmare for our customer.

For solving at least 90% of the problems, listed above, we developed open REST API for the system for any external integrations and extensions. Now, in case of developing new version or changes of DB storage we support API versioning, where process obsolete methods and entrypoints, handle data collisions and so on. After some period clients will migrate their migration to up-to-date API versions and we don't have system collapse after each new version. Of course, for multi-versioning support we have to support data transition structures, regression tests and other principles.