[Yaffs] Re: yaffs2 status

Top Page
Attachments:
Message as email
+ (text/plain)
Delete this message
Reply to this message
Author: Sergei Sharonov
Date:  
To: yaffs
New-Topics: Re: [Yaffs] Re: yaffs2 status --> mount speed up
Subject: [Yaffs] Re: yaffs2 status
Hi,
> > Can anybody post any reference numbers? For example it takes me 20+
> > minutes to mount 256 MByte NAND partion with a 200 MByte file on it
> > when using jffs2.
>
> That does sound remarkably slow. But obviously it does depend a lot of the
> hardware speed - what is the platform?


at91rm9200 (arm9 @ 180 MHz)
Clarification:
mount returns after 1.5 minutes but then any (first) write or even ls
takes 10..20 minutes. AFAIK a lot of work is left for gc thread after
mount is done. I've also seen much worse numbers when the large file
was saved via ftp. Node size? I have not investigated.


> Some ballpark figures reckon YAFFS(1) is 4 times as fast to boot per MB of
> flash, or twice as fast per meg of data (assuming the JFFS2 FS is
> compressed at 2:1, which is farily typical). (from
> http://www.aleph1.co.uk/talks/yaffs/mgp00025.html which came from collated
> posts to this list).


Hmm.. 8s/77MB seconds in yaffs case and 20m/200MB in jffs2 case is
more then 4X ;-)

> YAFFS2 is up to twice as fast on read (with 2K pages and 16bit bus) -
> on 512byte
> pages and 8 bit bus it is same speed as YAFFS1.


That is good to know. I have a 100 Mb/s Ethernet and I need to be able
to dump log data as fast as possible.

> ecc checking can really slow things down if it is done in software not
> hardware - although that's primarily on writing, so not important for
> initial scan time.


No hardware ECC here, but one would think 180 MIPS should mitigate that.

> The best way to get some reliable numbers would of course be to try it - it
> should definately be at least twice as fast. We'd be interested to hear what
> you find.


I definetly would like to try yaffs2 out. Where can I find howto?

> Ulitimately the solution to the problem that all log-structured
> filesystems have
> to scan the whole media on boot, and this takes time is checkpointing
> (storing intermediate, or shutdown, status), and smarter partitioning so
> you
> can start before scanning the whole media. The latter is available today -
> the former needs sponsoring.


Cannot do much with partitioning since must be able to suck the data out
fast even immediatelly after reset.

Sergei Sharonov