Nowadays, in order to compete in global markets, it is necessary for an organization to improve it’s supply chain performance. To increase efficiency and accountability of supply chains, novel managerial methods like coordination in supply chain management are always presented. Inventory management and traortation are two key issues in supply chains, which have a considerable effect on performance of supply chain management. By coordinating these issues, a proper distribution planning from suppliers to customers is achieved. By solving an inventory routing problem, the quantity of products, delivery time, type of traortation, and routes of product delivery to each retailer (customer) are determined. In the thesis, modelling and presenting solution method for the multi-product inventory routing problem with two-dimensional constraints are studied. In this combinatorial problem, load delivery from depot to customers are managed by means of routing and simultaneously controlling customers' inventory level in a way that in addition to consider weighted constraints of vehicles, classic two-dimensional loading constraints are met. The goal of the presented mathematical model is determining delivery periods of customers' demands and set of routes by which total costs including distribution and inventory costs are minimized along with preventing customers from encountering shortage during planning horizon and meeting weighted capacity, vehicle loading, and LIFO constraints. Inventory routing problem is a type of NP-hard problems. Therefore, in this study, heuristic and metaheuristic algorithms have been proposed to solve the problem. These algorithms can be categorized as follows: three heuristic algorithms for filling two-dimensional containers, a heuristic algorithm for creation of an initial solution, and two metaheuristic algorithms including adaptive large neighborhood search (ALNS) and turbulent population-based adaptive large neighborhood search for improving the proposed initial solution. Finally, to evaluate the performance of the proposed algorithms, several instances are designed. Results indicate that the turbulent population-based adaptive large neighborhood search outperforms ALNS algorithm.