Sasha Sirotkin currently works on the FemtoLinux Project, improving Linux latency and real-time capabilities. FemtoLinux's design goal is to bring Linux performance closer to that of an RTOS, such as VxWorks, to allow complex real-time application development and easy porting.
As Linux has become more widely used in embedded systems, the number of file systems which work directly with the flash storage (i.e. via MTD device as opposed to some block device hardware emulation layer such as the one present on most DiskOnKey devices) has grew substantially.
First embedded Linux systems used read-only CramFS and SquashFS file systems. These are still very much in use today, as many embedded devices such as routers do not need a real read-write file system. Such devices typically store only small amount of configuration information that can change, which occupies only a few 1KB, i.e. less than a single flash block and is usually written directly to the flash while root file system resides on CramFS or SquashFS.
As flash sizes increased and Linux moved into more embedded niches, the need for read-write flash file systems was answered by JFFS2 which for a long time was de-facto standard Linux flash file system. As flash sizes grew even more and devices such as cellular phones that store large amounts of information (pictures, mp3 files) started using Linux, JFFS2 reached its scalability limits. As a result, new file systems specifically designed for large NAND flash devices were developed -- UBIFS, LogFS, and YAFFS. For a long time only UBIFS was part of the mainline kernel and both YAFFS2 and LogFS where available as a patches. At some point in time it looked like LogFS developed was stagnated, with the latest patch available for kernel version 2.6.24. However, LogFS suddenly resurfaced and rather surprisingly was quickly merged into kernel 2.6.34 indicating that its developers kept working on this project, albeit with little publicity. YAFFS2, which contrary to LogFS was widely used, undergoes a similar process with respect to inclusion into mainlaine Linux kernel. It looks like even though in the past YAFFS2 developers did not make any significant effort to put it into the mainline kernel, it is going to change now.
When I started my embedded file system evaluation, I was almost certain that eventually I will choose UBIFS simply because it is part of the mainline kernel and YAFFS2 is an external patch. However, it turned out that YAFFS2 is actually easier to configure -- I kept getting errors while mounting UBIFS partition until I disabled "Verify NAND page writes" kernel parameter. Apparently this is a rather old and wellknown bug, which is still present in kernel 2.6.32 that I use on my systems. This is pretty subjective, but I had zero issues with YAFFS2 even though I had to patch the kernel. The patch works smoothly as all YAFFS2 files reside in a single directory "fs/yaffs2" and the only files that need to be modified are these related to the build system.
