On 07/09/12 23:28, Steven Chamberlain wrote:
[...]
> Now with softupdates, metadata changes might cancel out in-memory and
> they get written out a whole block at a time, so write latency is hardly
> relevant any more.
Yup, that seems decidedly plausible. Turns out that the spinning disk on
the amd64 box has a write cache, which is enabled, so that would have
hidden the problem. (It also has NCQ. I don't know whether the FreeBSD
driver uses NCQ or not.)
TBH I'd rather have the drive's write cache disabled anyway --- a proper
filesystem shouldn't need it...
[...]
> The drawbacks were explained on this page (last paragraph) :
>
> http://www.freebsd.org/doc/handbook/configtuning-disk.html#SOFT-UPDATES
I did actually read that, with some puzzlement, and eventually dismissed
it as an archaism, because I wouldn't expect any other behaviour with a
modern file system!
Anyway, thanks for the assistance; I am actually planning to put this
machine into active service reasonably soon...
--
┌─── ｄｇ＠ｃｏｗｌａｒｋ．ｃｏｍ ───── http://www.cowlark.com ─────
│
│ life←{ ↑1 ⍵∨.^3 4=+/,¯1 0 1∘.⊖¯1 0 1∘.⌽⊂⍵ }
│ --- Conway's Game Of Life, in one line of APL
Attachment:
signature.asc
Description: OpenPGP digital signature