[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-kernel
Subject:    Re: [patch] voluntary-preempt-2.6.8.1-P5
From:       Ingo Molnar <mingo () elte ! hu>
Date:       2004-08-20 18:56:33
Message-ID: 20040820185633.GE21956 () elte ! hu
[Download RAW message or body]


* K.R. Foley <kr@cybsft.com> wrote:

> Here is a latency trace generated by the NFS compile in the
> stress-kernel. It is ~1.7 ms which doesn't really surprise me for the
> NFS compile. I am going to take the NFS compile out of the test now
> and see what else falls out.

NFS uses lock_kernel() for most of its locking. This means that the
sharing and locking rules and assumptions are not clear (it's all within
this opaque lock). I tried to figure out whether it would be safe to do
a lock-break in nfs_cached_lookup() but it's quite hard ... Please bug
the NFS-client maintainer(s) about this. To break this latency we'd have
to do a lock-break in fs/nfs/dir.c:nfs_cached_lookup()'s main loop:

        for(;(page = find_get_page(dir->i_mapping, desc.page_index)); desc.page_index++) {

                res = -EIO;

you could try this at the end of the loop:

                if (res == 0)
                        goto out_found;
                if (res != -EAGAIN)
                        break;
+               cond_resched();
        }

but i cannot guarantee that this is correct ... It might seem to work 
now but break later under load, etc. It's the NFS maintainer's call.

> If it would be useful for me to provide additional information with
> this, please let me know.

reports like this are perfect for most latencies. I tried to include
everything that is needed normally in /proc/latency_trace itself.

	Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
[prev in list] [next in list] [prev in thread] [next in thread]