[Gllug] Partioning advice needed

[Nix]

On 19 Feb 2007, Tethys told this:
> Nix writes:
>
>>(Thankfully some ex-ZFS hackers, e.g. the inimitable and irrepressible
>>Val Henson, are working on the problem :) )
>
> Indeed. chunkfs looks interesting on that front. Val has that rare
> combination of talents: being able to design and implement clever
> new ideas *and* to be able to explain them coherently to the rest
> of us.

Quite so. Personally chunkfs looks like a awful kludge to me, but given
that I can't think of anything better and that it *does* allow use of
fairly conventional fsen on vast disks without major problems with fsck
time (and high-speed incremental fscks and the like), I can't really
complain too much.

(The only thing I worry about is that the chunk-superblock updates to
mark the things as clean and dirty all the time might end up increasing
write inefficiency quite a lot, since they're necessarily write barriers
with big seeks on either side of them. But md has the same problem and I
can't actually detect a slowdown --- even in benchmarks it's barely
detectable --- so again I think I'm worrying about nothing here.)

>          But the more I looked into it, I realised that systemtap had
> the better design,

I dunno. systemtap's `write a kernel module and compile it' feels icky
at first sight, until you consider that without that you'd need a
kernel-side interpreter, and one of those (ACPI) is quite enough. The
write-and-compile thing is just moving that interpreter to userspace :)

>                    and could potentially do a lot more. It was just

Oh yes :)

> a matter of time before it caught up and overtook dtrace. It still
> has some useability issues (particularly when, for example, trying to
> trace a program on a production box without a C compiler),

That's never going to work. The solution's easy, thankfully (although
some idiot security staffs seem to think that C compiler -> security
hole, perhaps thinking that the days when worms and rootkits sent C code
over the net and compiled themselves on the target were still with us:
hint, they're not).

>                                                             but it's
> getting better all the time, and it's now an essential debugging tool
> for me.

I'll find it much more useful when it works in userspace too: nearly
all my work is on userspace apps.

-- 
`In the future, company names will be a 32-character hex string.'
  --- Bruce Schneier on the shortage of company names
-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug