Re: [PATCH] tracer for sys_open() - sreadahead

[Corrado Zoccolo <czoccolo@gmail.com>]

On Thu, Feb 5, 2009 at 4:47 PM, Arjan van de Ven <arjan@infradead.org> wrote:
>
> another issue with the "use strace" approach is selinux...
> (this is in addition to the performance and behavioral changes)

Regarding behavioral changes, we should consider that just introducing
sreadahead in the boot process will change the system behaviour,
especially with distros that try and start multiple processes in
parallel during boot.

This means that, in order to approximate the optimal boot, sreadahead
should monitor every boot (even when it is performing readahead), and
modify the order of the opened files accordingly.
I'm using a modified sreadahead version that implements this (on top
of original ext3 patch), and the number of I/O waits shown by
bootchart reduced noticeably w.r.t. naive sreadahead.

So the assumption that just 1 slow boot is enough doesn't hold. We
should try to have a low overhead tracing system in the kernel, to be
able to implement this more efficiently.

Corrado