High Performing Distributed Order Processing System

Abstract

Transactional processing systems are one of the key areas in modern software engineering products. Order management systems (OMS) with online trading facility is one major area where transaction processing systems are heavily used.In online trading system, the key requirement is to send the client orders to stock exchange for processing as quickly as possible. All OMS developers try to keep the round trip time of the order as small as possible. Approximately around few millisecond level. Ability to handle several hundred of transaction per second, automatic fail over, recovery and intelligent load balancing also are the other important characteristics of a better Order Management System.Input and output operations such as direct Database readings and writings during the order processing heavily affects to the performance of the system. Costly, IO operations can be removed completely by introducing a distributed caching framework in between business logic and DB layer.With the caching layer all read and write operations deal with caches those are key value stores in the memory. Updated data in caches write back to the Database by separate threads running at outside the transaction path. Real time distributed characteristic of the caching framework is used to overcome the limitation such as data losses at node crashes.Automatic fail over and recovery is introduced by using the node cluster running the multiple copy of the same OMS with the replicated caching layer. Load balancing strategies such as client count base load distribution, message type base load distributions etc. further improve the throughput and scalability of the system.With the help of distributed transactional caching framework and careful performance tuning, an OMS capable of handling several hundred of orders can be developed successfully. Average round trip time per order even in most complex case can be kept around 3ms to 4ms. System can be scaled horizontally seamless manner to deals with growing customer base with the help of various load balancing strategies. Automatic fail recovery features hide the internal server crashes from the external users of the OMS.