cvs commit: src/usr.sbin/gstat gstat.c

[Poul-Henning Kamp]

In message <4780CC1B.8090708 at freebsd.org>, Colin Percival writes:
>Poul-Henning Kamp wrote:
>>   When the ms/req fields exceed 1 second, drop the fractions to fit more digits.
>>   
>>   This is unfortunately necessary with some flash based devices which can
>>   get hundreds of seconds behind with softupdates enabled.
>
>I'm curious, why is this an issue with flash devices?  Given their famous
>ability to perform large numbers of transactions per second, I would have
>expected flash-based drives to be the least likely to have this problem.

The problem is flash-devices which are built (mostly) for cameras, their
(cheap-ass) flash-adaptation-layer performs great for physically
sequential writes for big files, and they fall totally apart with
random writes to things like UFS/FFS inodes and bitmaps.

Pretty much all of the layout optimizations in UFS/FFS are pessimizations
for Flash devices.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.