Potential skew in /proc/net/rpc/nfsd

Potential skew in /proc/net/rpc/nfsd

[Christian Robottom Reis]

(Linux anthem 2.6.10-5-686-smp #1 SMP (Ubuntu Hoary))

Hello there,

    I've been studying the nfsd statistics file on our server, a
dual-hyperthreaded-Xeon box (which appears to 2.6 as having 4 CPUs) and
I've noticed an oddits in the thread statistics. Here's the output,
edited to display clearly what I mean:

            th
             8
       2747936
         59079.566
         14835.486
          5952.540
             0.000
          3293.329
          2232.819
          1900.259
          1999.936
             0.000
          6549.737

Do the values of zero in the 4th and 9th percentage there make any
sense, or are they an artifact of how the statistics are calculated
(noting that I have 8 threads being used)? I've shifted the number of
threads to 64 this morning, and this afternoon I now see:

            th
            64
       2748083
         59543.017
         14912.958
          5959.034
             0.854
          3293.654
          2233.022
          1900.395
          1999.936
             0.000
          6549.737

The 0.854 there in the 4th percentage suggests to me that the accounting
is now functioning as it should; I might be mistaken, since it's the
first time I look at these numbers, but that's what it suggests to me.

Take care,
--
Christian Robottom Reis | http://async.com.br/~kiko/ | [+55 16] 3376 0125


-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Potential skew in /proc/net/rpc/nfsd

[Neil Brown]

On Wednesday October 19, kiko@async.com.br wrote:
> 
> (Linux anthem 2.6.10-5-686-smp #1 SMP (Ubuntu Hoary))
> 
> Hello there,
> 
>     I've been studying the nfsd statistics file on our server, a
> dual-hyperthreaded-Xeon box (which appears to 2.6 as having 4 CPUs) and
> I've noticed an oddits in the thread statistics. Here's the output,
> edited to display clearly what I mean:
> 
>             th
>              8
>        2747936
>          59079.566
>          14835.486
>           5952.540
>              0.000
>           3293.329
>           2232.819
>           1900.259
>           1999.936
>              0.000
>           6549.737
> 
> Do the values of zero in the 4th and 9th percentage there make any
> sense, or are they an artifact of how the statistics are calculated
> (noting that I have 8 threads being used)?

An artifact of only having 8 threads.
The percentage of threads that can be in use with only 8 must be one of
0%, 12.5%, 25%, 37.5%, 50%, 62.5%, 75%, 87.5%, 100%

Note that none of those line in the [40-50) or [90-100) ranges.

> I've shifted the number of
> threads to 64 this morning, and this afternoon I now see:
> 
>             th
>             64
>        2748083
>          59543.017
>          14912.958
>           5959.034
>              0.854
>           3293.654
>           2233.022
>           1900.395
>           1999.936
>              0.000
>           6549.737
> 
> The 0.854 there in the 4th percentage suggests to me that the accounting
> is now functioning as it should; I might be mistaken, since it's the
> first time I look at these numbers, but that's what it suggests to
> me.

Yes.  It was always functioning as it should.  If you look at the
differences between the second set of number and the first, you should
see a sensible distributing.
It looks like you never get up to 80% of the 64 threads, so you were
never using more than about 50 threads in that time.

NeilBrown


-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs