cluster-devel.redhat.com archive mirror
 help / color / mirror / Atom feed
From: Bob Peterson <rpeterso@redhat.com>
To: cluster-devel.redhat.com
Subject: [Cluster-devel] [GFS2 PATCH] GFS2: Average in only non-zero round-trip times for congestion stats
Date: Thu, 23 Apr 2015 12:45:01 -0400 (EDT)	[thread overview]
Message-ID: <918499302.5872367.1429807501247.JavaMail.zimbra@redhat.com> (raw)
In-Reply-To: <5538B62C.4030303@redhat.com>

----- Original Message -----
> Hi,
> 
> That looks better. Do you get better results with it compared with the
> previous version?
> 
> Steve.
> 
> On 22/04/15 18:00, Bob Peterson wrote:
> > Hi,
> >
> > This patch changes function gfs2_rgrp_congested so that it only factors
> > in non-zero values into its average round trip time. If the round-trip
> > time is zero for a particular cpu, that cpu has obviously never dealt
> > with bouncing the resource group in question, so factoring in a zero
> > value will only skew the numbers. It also fixes a compile error on
> > some arches related to division.
> >
> > Regards,
> >
> > Bob Peterson
> > Signed-off-by: Bob Peterson <rpeterso@redhat.com>

It's not straightforward because the "preferred rgrps" patch skews the
performance results greatly. However, based on my performance tests, yes,
the numbers do look better. I temporarily commented out that "preferred
rgrps" patch to get a clearer picture. Here are numbers from my testing:
This test consists of 5 nodes all simultaneously doing this 'dd' command:
dd if=/dev/zero of=/mnt/gfs2/`hostname` bs=1M count=100
(after drop_caches)

Stock (nothing disabled):
            Run 1
         --------
pats     716 MB/s
jets     615 MB/s
bills    669 MB/s
dolphins 605 MB/s
ravens   735 MB/s
         --------
Average: 668 MB/s

Stock (plus preferred rgrps disabled): AKA 'c3'
            Run 1     Run 2
         --------  --------
pats     766 MB/s  675 MB/s  
jets     649 MB/s  716 MB/s  
bills    790 MB/s  735 MB/s  
dolphins 761 MB/s  727 MB/s  
ravens   712 MB/s  728 MB/s  
         --------  --------
Average: 736 MB/s  716 MB/s

Without latest patch (plus preferred rgrps disabled): AKA 'c2'
(In other words, this is the previous patch which was called
"GFS2: Use average srttb value in congestion calculations")
            Run 1     Run 2     Run 3 
         --------  --------  ---------
pats     830 MB/s  688 MB/s  697 MB/s
jets     833 MB/s  622 MB/s  645 MB/s
bills    831 MB/s  796 MB/s  637 MB/s
dolphins 834 MB/s  597 MB/s  690 MB/s
ravens   815 MB/s  731 MB/s  734 MB/s
         --------  --------  ---------
Average: 829 MB/s  687 MB/s  681 MB/s

Latest patch (plus preferred rgrps disabled):
            Run 1     Run 2     Run 3 
         --------  --------  ---------
pats     811 MB/s  829 MB/s  652 MB/s
jets     825 MB/s  863 MB/s  702 MB/s
bills    846 MB/s  825 MB/s  710 MB/s
dolphins 845 MB/s  845 MB/s  683 MB/s
ravens   820 MB/s  818 MB/s  682 MB/s
         --------  --------  ---------
Average: 829 MB/s  836 MB/s  686 MB/s

Latest patch (nothing disabled):
            Run 1     Run 2     Run 3 
         --------  --------  ---------
pats     834 MB/s  817 MB/s  819 MB/s
jets     837 MB/s  835 MB/s  836 MB/s
bills    841 MB/s  837 MB/s  834 MB/s
dolphins 838 MB/s  851 MB/s  842 MB/s
ravens   795 MB/s  808 MB/s  815 MB/s
         --------  --------  ---------
Average: 829 MB/s  830 MB/s  829 MB/s

This test (simultaneous dd) is known to be a worst case scenario,
so I expect it to show the most improvement. For ordinary block
allocations, I don't expect that big of an improvement.

Regards,

Bob Peterson
Red Hat File Systems



  reply	other threads:[~2015-04-23 16:45 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <1202489161.4769650.1429721928815.JavaMail.zimbra@redhat.com>
2015-04-22 17:00 ` [Cluster-devel] [GFS2 PATCH] GFS2: Average in only non-zero round-trip times for congestion stats Bob Peterson
2015-04-23  9:06   ` Steven Whitehouse
2015-04-23 16:45     ` Bob Peterson [this message]
2015-04-23 17:59       ` Steven Whitehouse

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=918499302.5872367.1429807501247.JavaMail.zimbra@redhat.com \
    --to=rpeterso@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).