This research is to design a dynamic data packing and flushing method for buffer architecture to improve write performance of Hybrid Solid State Disk (HSDD), combined with Single Level Cell (SLC) and Multi Level Cell (MLC) flash memory, for multimedia applications. SLC has relatively excellent endurance and performance and is expensive, whereas MLC has relatively bad endurance and performance and is cheap. As short and frequent write requests (random request) are transferred in SLC and long and infrequent write requests (sequential request) are transferred in MLC, HSSD processes write requests efficiently to enhance its overall endurance. To optimize the write performance of HSSD, access pattern is analyzed to devise a decision function to determine random or sequential request, which can be one of major media accessing characteristics. Specifically data must be written in parallel for maximum sequential allocation in case of multimedia applications. But it is difficult to detect precise access pattern of requests, because several requests tend to be mixed by multi-processing. Also data tend to be written intensively in some specific logical blocks. Thus, data being written should be pushed and packed into the buffer to enhance the degree of sequential write. And the data to be flushed is selected by the condition of HSSD. The proposed buffering method shows better performance by 42% on average than basic buffering structure. Although an overhead occurred by additional structure in proposed architecture is considered.