xfs
[Top] [All Lists]

Re: definitions for /proc/fs/xfs/stat

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: definitions for /proc/fs/xfs/stat
From: Mark Seger <mjseger@xxxxxxxxx>
Date: Fri, 14 Jun 2013 21:55:17 -0400
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; bh=8E4KrLXL7l+BkxDxGb4vjPxke1rn3LmdpEzw5s9nKeQ=; b=Dj5PBj5NRAYJisiTy8UG4gjC3gytrOaQFlwg9QkkFghEZMs8gl/Z5xklPvM9SeLss+ d6ewndlxz2c+XRsIDRyH2yfjkSwc1JF/ON9bGsbjHf0M+W8ZVpQJb/1IsQg8dCUGDiCL 1fQFrTpLjScxpbRkPPBRxhvzgqdrhkbt0qy9muxtGKctDTS4ShegPLWt1+f3E9q7atg+ hg5FyOVhflZ/f5EOAlUWs3FlZwaORm4i5cY7rv5WN+1t+O4wxoKhYYfL3qfhPlgWUQUe bx19LLzabJPkp8+q8CfC8iWKD0kEb0OdzV85RKzdkTDz8izNBVZysxGK3vw8B5NsX5xl KboQ==
In-reply-to: <504625587.1365681.1371255450937.JavaMail.root@xxxxxxxxxx>
References: <CAC2B=ZFP_Fg34aFpk857stgB7MGcrYs9tybRS-ttw1CXNeU41Q@xxxxxxxxxxxxxx> <91017249.1356192.1371248207334.JavaMail.root@xxxxxxxxxx> <CAC2B=ZHYV6d-1PO_=-jXsQidZnYPHVwcVAaQh2mxJt=5K03AEA@xxxxxxxxxxxxxx> <504625587.1365681.1371255450937.JavaMail.root@xxxxxxxxxx>
I'm doing 1 second samples and the rates are very steady.  The reason I ended up at this level of testing was I had done a sustained test for 2 minutes at about 5MB/sec and was seeing over 500MB/sec going to the disk, again sampling at 1-second intervals.  I'd be happy to provide detailed output and can even sample more frequently if you like.

from my shorter test I was experimenting looking at some of the XFS data with collectl and recorded this, it if help at all:

segerm@az1-sw-object-0006:~$ collectl --import xfs
waiting for 1 second sample...
#<--XFS Ops--><-----------XFS Logging----------><------Extents------><------DirOps-------><----Trans---><-
# Write Reads  Writes WrtKBs NoRoom Force Sleep  ExtA BlkA ExtF ExtF  Look Cre8 Remv Gdnt  Sync Asyn Empt
      0     0       0      0      0     0     0     0    0    0    0     0    0    0    0     0    0    0
      0     0       3    768      0     3     3     0    0    0    0     0    0    0    0     3    0    0
      0     0       0      0      0     0     0     0    0    0    0     0    0    0    0     0    0    0
      0     0       0      0      0     0     0     0    0    0    0     0    0    0    0     0    0    0
      0     0       0      0      0     0     0     0    0    0    0     0    0    0    0     0    0    0
      0     0       1    256      0     1     1     0    0    0    0     0    0    0    0     1    0    0
     38    88      95  24320      0    95    96    54   54   54   54   115   76   76  154    95  473    0
    339   776     968 247816      0   968   978   484  484  479  479  1011  675  671 1351   967 4087    0
    321   748     929 237806      0   929   935   450  450  453  453   967  645  647 1287   930 3847    0
    279   637     810 207360      0   810   811   391  391  390  390   838  559  558 1118   810 3324    0
    209   482     610 156160      0   610   610   286  286  289  289   627  417  420  834   610 2451    0
      0     0       3    768      0     3     3     0    0    0    0     0    0    0    0     3    0    0

I can say for a fact I was doing about 300 wtrites/sec whcih the write numbers seem to support, though I don't know what the read numbers are measuring.  you can also see from the logging data that was 250MB/sec going to disk.

Are there other numbers that are meaningful that you'd like to see?  All it takes is adding a couple of print statement as what you're seeing above only took a hour or so to throw to together.

I can collect as much (or little as you like) and actually can save the complete contents of /proc/fs/xfs/stat every second in a file for later playback.

-mark



On Fri, Jun 14, 2013 at 8:17 PM, Nathan Scott <nathans@xxxxxxxxxx> wrote:


----- Original Message -----
> actually I have since found a decoder ring here -
> http://xfs.org/index.php/Runtime_Stats and have been incorporating a lot of
> the data so I can look at things in real time.  I'd still love to know why

Ah, good stuff.

> writing 1000 1K files results in 200MB/sec of disk I/O though.  clearly

For how many seconds?  (or fractions thereof?)  You may need the level of
detail that only tracing will provide to explore further, although off the
top of my head I'm not sure exactly which events you should focus on (log,
inode creation, or space allocation at a guess).

cheers.

--
Nathan

<Prev in Thread] Current Thread [Next in Thread>