In the past, two main storage technologies, a hard disk drive(HDD) and flash memory chips, had its own distinct uses. The hard disk drive technology was used in machines requiring much storage spaces, such as servers, PCs, laptops while flash memory was used in devices that needed less storage capacity, such as digital cameras, cell phones, and embedded controllers. But recent users’ desires converge on both areas; small platform but large storage capacity. Accordingly, considering the capacity, HDD was thought of as the solution for such devices but HDDs are not fit to small hand-held devices for many reasons. In contrast, NAND has been thought as the main storage replacing HDDs according to the recent advanced technology. Moreover, with the increasing storage capacity, the main features of NAND, for example, light weight, shock-resistance, no seek time are very attractive to the manufacturers and developers. Considering many advantages of NAND, it will take the many places of HDDs.
YAFFS is a de facto standard flash memory file system for NAND flash-memory but it has problems to be an alternative for multimedia file system;it do not have non-linear editing support and it may takes too much time to mount if the NAND is not clean. Moreover, it consumes pages for meta-data which cause the performance degradation by frequent updates and garbage collection.
Another advanced flash file system, PFFS was developed to solve the problems of YAFFS. However, PFFS also is not suitable to be used as a multimedia file system with following reasons. First, PFFS is based on the disk-based EXT2 file system, which means it spends too much storage for metadata because the PFFS uses pointers for data pagess. Second, because PFFS only uses 13 direct, 2 indirect, 2 doubly indirect entries, the largest file size that PFFS can point is 1GB which is not enough size for multimedia file system. Third, PFFS do not consider the non-linear editing desire. Lastly, it also spends NAND pages for updating metadata of large files because its indirect pointers are stored in NAND pages. This additionally causes performance degradation.
In this paper, we propose a Multimedia Non-linear editing File System(MNFS) which works for mobile multimedia devices providing non-linear editing editing with PRAM-NAND flash memory. For this goal, MNFS introduces the extent indexing which prevents wasting of PRAM unlike PFFS. In addition to this, MNFS provides non-linear editing system calls for time and space efficiency. Lastly, MNFS uses extent-aware garbage collection to merge the split extents.
The experiments shows that MNFS outperforms YAFFS for non-linear editing jobs. Moreover, MNFS performs faster metadata updates than YAFFS without wasting pages. In TIO benchmark test after aging, YAFFS has 88.6 performance degradation in the worst case while MNFS does not show that compared to the initial performance.
플래시를 사용한 멀티미디어 장치들이 많아지고 있으나, 이러한 장치를 위한 멀티미디어 플래시 파일 시스템은 존재 하지 않았다. 따라서, 우리는 이러한 멀티미디어 장치를 위해 Non-Linear Editing(NLE) 기능을 제공하며, PRAM에 구현된 멀티미디어 파일 시스템을 제안하였다. MNFS는 extent indexing을 사용하여, PFFS와 같이 인덱싱을 위하여 많은 스토리지 공간을 쓰지 않으며, NLE 기능을 사용하면 YAFFS에서 발생하는 re-write cost와 erase cost 낭비를 줄이고, 효율적으로 편집하게 해준다. 또한 편집으로 인한 잦은 메타데이터 업데이트로 발생하는 오버헤드는 PRAM을 이용하여 해결함으로써 성능을 더욱 향상 시켰다. 실험 결과 NLE 사용한 편집은 YAFFS에서 보다 월등한 성능을 보였으며, page낭비, re-write, erase cost를 없앴음을 나타낸다. TIO 벤치마크를 통한 IO 테스트에서도 YAFFS보다 뛰어난 성능을 보였다.