> That looks promising. Would you like to do some more empirical tests for
> tuning? Make the allocation page-cluster size a global variable, and then
> use gdb to adjust its value in between test runs. If you'd like to do
> that, I'd be happy to put that form of your changes in right away whil
Conversely, you could just do spin_try_lock and have any thread that fails
to get the lock simply mmap a single page itself rather than waiting for
the other thread.
___
Bug-hurd mailing list
[EMAIL PROTECTED]
http://mail.gnu.org/mailman/listinfo/bug-hu
> Sure. Do you have a suggestions on how to gather statistics,
> i.e. other than compiling.
Not particularly. Pick some task that does the same thing every time and
does a lot of disk i/o. Compiles are such things.
> Additionally, was changing the spinlock to a mutex the right thing to do?
A
> That looks promising. Would you like to do some more empirical tests for
> tuning? Make the allocation page-cluster size a global variable, and then
> use gdb to adjust its value in between test runs. If you'd like to do
> that, I'd be happy to put that form of your changes in right away whil
That looks promising. Would you like to do some more empirical tests for
tuning? Make the allocation page-cluster size a global variable, and then
use gdb to adjust its value in between test runs. If you'd like to do
that, I'd be happy to put that form of your changes in right away while you
(a
> Either way, I think that correcting this false
> optimization is the Rigth Thing to do and that this is a reasonable
> implementation.
I decided that I wanted to see if there really was a measurable
speedup, so I ran some trivial non-scientific tests. I compiled the
Hurd four times (each time
> Have you observed or measured a particular benefit to this change?
Admittedly, when I wrote the patch, I had not actually gathered any
statistics: I was basing all of my changes on what I saw in the code
and the assumptions that I made about the behavior of ext2fs.
However, based on your prompt
Have you observed or measured a particular benefit to this change?
___
Bug-hurd mailing list
[EMAIL PROTECTED]
http://mail.gnu.org/mailman/listinfo/bug-hurd
I have the modified the allocation and deallocation scheme in the
ext2fs pager code to be a bit smarter. Now, when we need to allocate
a single page, we preallocate several more at the same time. This
code also reduces the lock contention by waiting until as late as
possible to take a lock.
I a
