Packet load-balancing is a key technique for parallel packet switch (PPS). It is already known that there have been practical load-balancing techniques with IP routing table lookup under the line rate of OC768 (40GBps) or OC3072 (160Gps). However, according to the technology of today, it is difficult to implement packet processing under such high line rate. Based upon demultiplexing packet processing and packet load-balancing, a demultiplexing distributed architecture for packet load-balancing and its algorithm DDPA (demultiplexing distributed parallel packet switch algorithm) are presented. Moreover, a modified DDPA based on packet-discarding is proposed, which is more practical. The validity of the technique is proved and the key parameters are deterministically guaranteed.