I'm by no means an expert in this, I just follow the list to learn, but
would it not be possible to make ITERATIONS count a runtime configurable
parameter in the /proc filesystem that defaults to 100? That would allow
for the best tuning scenario for a given system.
Just a thought.
John Cavan
"Juan J. Quintela" wrote:
>
> >>>>> "trond" == Trond Myklebust <trond.myklebust@fys.uio.no> writes:
>
> >>>>> " " == Juan J Quintela <quintela@fi.udc.es> writes:
> >> Then you want only invalidate the non_locked pages: do you
>
> trond> That's right. This patch looks much more appropriate.
>
> >> + while (count == ITERATIONS) {
> >> + spin_lock(&pagecache_lock);
> >> + spin_lock(&pagemap_lru_lock);
> >> + head = &inode->i_mapping->pages;
> >> + curr = head->next;
> >> + count = 0;
> >> +
> >> + while ((curr != head) && (count++ < ITERATIONS)) {
>
> trond> Just one question: Isn't it better to do it all in 1 iteration through
> trond> the loop rather than doing it in batches of 100 pages?
> trond> You can argue that you're freeing up the spinlocks for the duration of
> trond> the loop_and_test, but is that really going to make a huge difference
> trond> to SMP performance?
>
> Trond, I have not an SMP machine (yet), and I can not tell you numbers
> now. I put the counter there to show that we *may* want to limit the
> latency there. I am thinking in the write of a big file, that can
> take a lot to free all the pages, but I don't know, *you* are the NFS
> expert, this was one of the reasons that we want feedback from the
> users of the call. (You have been very good giving comments).
>
> My idea to put a limit is to put a limit than normally you do all in
> one iteration, but in the exceptional case of a big amount of pages,
> the latency is limited. There is a limit in the number of pages that
> can be in that list?
>
> 100 is one number that can need tuning, I don't know. SMP experts
> anywhere?
>
> By the way, while we are here, the only difference between
> truncate_inode_pages and invalidate_inode_pages is the one that you
> told here before? I am documenting some of the MM stuff, and your
> comments in that aspect are really wellcome. (You will have noted
> now that I am quite newbie here).
>
> Later, Juan.
>
> --
> In theory, practice and theory are the same, but in practice they
> are different -- Larry McVoy
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.rutgers.edu
> Please read the FAQ at http://www.tux.org/lkml/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 21:00:20 EST