#3725 HIGH First D: JFFS2 needs more aggressive garbage collection
Zarro Boogs per Child
bugtracker at laptop.org
Fri Sep 21 19:18:41 EDT 2007
#3725: JFFS2 needs more aggressive garbage collection
-------------------------------+--------------------------------------------
Reporter: wmb at firmworks.com | Owner: dwmw2
Type: defect | Status: new
Priority: high | Milestone: First Deployment, V1.0
Component: kernel | Version: Development build as of this date
Keywords: JFFS2 | Verified: 0
-------------------------------+--------------------------------------------
Pursuant to #3247, Quanta provided us with a NAND FLASH image which
appears not to have any obvious errors, but which takes a long time to
mount. The initial mount under OFW takes about 40 seconds and the
subsequent mount under Linux takes about 50 seconds. (Compare with
typical 5-second mount times for fresh JFFS2 images from OS builds.) dwmw2
and wmb both have copies of this image.
The image resulted from Quanta's long-run testing, which includes
writing/deleting/rewriting several files of various sizes many times.
This image has 800K dirent nodes and 6M data nodes, compared to 25M
dirents and 300K data nodes for a fresh image. All but one of the blocks
have summaries, so lack of summaries cannot account for the slowness. The
image occupies 8003 of the approximately 8192 (minus bad blocks) on the
NAND, but I suspect that the amount of current (non-obsolete) data is far
less. If this image is like another long-run image that I analyzed some
time ago, many of the nodes are obsolete and there are lots of
consecutive small nodes that could be coalesced.
dwmw2 has offered some ideas for optimizing JFFS2's garbage collection
process; it seems like a good time to implement some of those idea.
--
Ticket URL: <https://dev.laptop.org/ticket/3725>
One Laptop Per Child <https://dev.laptop.org>
OLPC bug tracking system
More information about the Bugs
mailing list