next up previous
Next: Conclusion Up: JFFS : The Journalling Previous: JFFS2

Subsections

Future Development

One oft-requested feature which is currently not planned for development in JFFS2 is eXecute In Place (XIP) functionality. When programs are run from JFFS2, the executable code is copied from the flash into RAM before the CPU can execute it. Likewise, even when the mmap() system call is used, data are not accessed directly from the flash but are copied into RAM when required.

XIP functionality in JFFS2 is not currently planned because it is fairly difficult to implement and because the potential benefits of XIP are not clearly sufficient to justify the effort required to do so.

For obvious reasons, XIP and compression are mutually exclusive - if data are compressed, they cannot be used directly in place. Given a prototype platform with sufficient quantities of both RAM and flash that neither XIP or compression are required, and the desire to save money on the hardware, a choice can be made between halving the amount of RAM and using XIP, or halving the amount of flash and using compression.

By choosing the latter option, the cost saving will generally be greater than the former option, because flash is more expensive than RAM. The operating system is able to be more flexible in its use of the available RAM, discarding file buffers during periods of high memory pressure. Furthermore, because write operations to flash chips are so slow, compressing the data may actually be faster for many workloads.

The main problem with XIP, however, is the interaction with memory management hardware. Firstly, for all known memory management units, each page of data must be exactly page-aligned on the flash chip in order for it to be mapped into processes address space - which makes such a file system even more wasteful of space than the mere absence of compression already implies. Secondly, while giving write or erase commands to a flash chip, it may return status words on all read cycles, therefore all currently valid mappings of the pages of the chip would have to be found and invalidated for the duration of the operation. These two limitations make a writable filesystem with XIP functionality extremely difficult to implement, and it is unlikely that JFFS2 could support XIP without fundamental changes to its design.

An read-only XIP filesystem would be a more reasonable request, and an entirely separate file system providing this functionality, based on the existing ROMFS file system, is likely to be developed at some time in the near future.

Improved Fault Tolerance

The main area where JFFS2 still requires development is in fault tolerance. There are still areas where, although designed to be resilient, JFFS2 may exhibit a more serious failure mode than is absolutely necessary given a physical error.

In particular, JFFS2 will need more sophisticated methods of dealing with single-bit errors in flash chips. Currently, the node contains a 32-bit CRC, but this only gives error detection; it does not allow the file system to correct errors. Error correction is an absolute requirement for operation on NAND flash chips, which have lower tolerances. It is desirable even on NOR flash.

JFFS2 already has a primitive method of dealing with blocks for which errors are returned by the hardware driver -- it files them on a separate bad_list and refuses to use them again until the next time the file system is remounted. This should be developed.

Garbage Collection Space Requirements

A major annoyance for users is the amount of space currently required to be left spare for garbage collection. It is hoped that a formal proof of the amount of required space can be produced, but in the meantime a very conservative approach is taken -- five full erase blocks must be available for writing before new writes will be permitted from user space.

It should be possible to reduce this figure significantly -- hopefully to a single block for NOR flash and to two or three blocks in the case of NAND flash, where extra space should always be available to copy away data from bad blocks.

The approach to this problem in JFFS1 was to evaluate and attempt to prove an upper bound on the amount of space required. This appeared to fail because there appeared to be no such upper bound. For JFFS2, it is suspected that a more useful approach may be to define a reasonable upper bound, such as a single erase block, and to modify the code to make it true.

Transaction Support

For storing database information in JFFS2 file systems, it may be desirable to expose transactions to user space. It has been argued that user space can implement transactions itself, using only the file system functionality required by POSIX. This is true -- but implementing a transaction-based system on top of JFFS2 would be far less efficient than using the existing journalling capability of the file system; for the same reason that emulating a block device and then using a standard journalling file system on top of that was considered inadequate.

Little work -- and relatively little thought -- has gone into this subject with respect to JFFS2, yet at first consideration it seems that to implement this in JFFS2 would not be particularly difficult or obtrusive. It is an interesting avenue for future research.


next up previous
Next: Conclusion Up: JFFS : The Journalling Previous: JFFS2
David Woodhouse 2001-10-10