Re: [ECOS] jffs2 garbage collection

[Eric Donnat <e dot donnat at silicomp dot fr>]

On Thu, 26 Jun 2003 18:30:57 +0200
"Koeller, T." <Thomas.Koeller@baslerweb.com> wrote:

> I am planning to use the jff2 flash file system
> for data storage in my target, but I am a bit
> worried about one topic: garbage collection. So
> if anyone with prior experience in this field could
> shed some light on the topics below, I'd be
> gratefuel.
> 
> My understanding is that gc can be triggered
> by any operation that increases the amount of data
> stored in the file system, such as writing data to
> a file. Since gc may cause erasing of flash sectors,
> which is a slow operation and may even involve busy
> waiting within the flash driver, I am afraid of
> the impact on the entire system if files are written
> by high-priority threads.
> 
> Should I really be concerned with this, or does
> jffs2 already take measures to avoid such problems?
> Is there a way to offload gc to a low-priority background
> thread? Should I constrain jffs2 file writing to a
> 'file server' thread with suitably chosen priority?
> 

You hit the right point. The jffs2 mailing list may give you
some other insights about locking issues. ...

You may learn other performance issues by just examining
the "gueing" code. There is also a explicit readme.txt's sentence
that ends with
"... this code is still in an early stage of development".

In practice, this result is such line of code:
[...]
FSTAB_ENTRY(jffs2_fste, "jffs2", 0,
            CYG_SYNCMODE_FILE_FILESYSTEM | CYG_SYNCMODE_IO_FILESYSTEM,
            jffs2_mount,
[...]

And maybe some other bottlenecks, I just hadn't time
to enumerate.

IMHO, the jffs2 port is still in a stage that had focused on
critical objectives such as "stability", "compatibility with linux", ...
but "hyperperformance on eCos".


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss