Objectives: The paper deals with the Online Multipath Routing of the Traffic flows in MPLS networks. A good multipath routing algorithm fulfills all the online traffic demands at real time. Many conventional multipath routing algorithms were considered good in respect of many parameters but still suffered from their highly rejected online traffic demands rate. Methods/Statistical Analysis: The algorithm presented classifies the traffic demands into various profile classes (which are generated based on the SLAs signed by the Internet users). These profile classes work as a rough estimator of future traffic demands and are used to solve a multi commodity network flow problem. In the proposed algorithm Multicommodity network flow formulation groups the packet flows as per profile classes with the aim to minimize the cost, to prevent network hotspots i.e. bottleneck links and to ensure minimum rejected requests/ traffic demands. The algorithm works in two phases- In the preprocessing phase SLAs, Global and Local quasi static knowledge about the network are used to generate Multi-commodities based “Profile Classes” in the first phase and then these Multi-commodities based packet flows are distributed over widest residual bandwidth Disjoint Paths with respect to the bottleneck links in the second phase. The transmission of Multi-commodity based flows through disjoint paths w.r.t. bottleneck edges prevent the network from critical deadlock point which helps in minimizing the congestion, delays, rejected requests and maximizing the throughput, i.e. improving the overall performance of the network. Findings: To measure the performance of proposed algorithm to the Profile Based Algorithm a series of experiments were measured. Randomly generated request sequence is used for which a random sequence of individual flow request were generated and measured the performance both in terms of number of flows routed and total bandwidth request that was satisfied. Blocking probability for each edge is also considered random.In the findings is shown that out of total 1200 requests, performance of PBR is very poor as it rejects many requests due to overly strict admission control and minimum hop algorithm. The proposed algorithm cleverly controls traffic flow admission by considering congested edges in a disjoint way, so that one LSP does not interfere other LSPs. Application/Improvements: The network information is utilized to its extreme and most of the online path requests are accepted only a few requests are rejected.