blktrace.git
14 years agoAdded in running stats for btt
Alan D. Brunelle [Fri, 14 Aug 2009 17:00:27 +0000 (13:00 -0400)]
Added in running stats for btt

Create an overall system and per-device statistics file containing
MB-per-second and I/Os-per-second values. Format for each file is first
column contains an (integer) time stamp (seconds since start of run)
and a (double) value.

File names are:

sys_mbps_fp.dat - system-wide mbps (for all devices watched, of course)
sys_iops_fp.dat - I/Os per sec

Each device watched will have a file with the device preceeding the
_mbps or _iops section of the above file names.

15 years agoVersion 1.0.1 blktrace-1.0.1
Jens Axboe [Mon, 11 May 2009 12:00:10 +0000 (14:00 +0200)]
Version 1.0.1

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkrawverify: warn and return error if no traces are found
Eric Sandeen [Thu, 7 May 2009 16:08:25 +0000 (11:08 -0500)]
blkrawverify: warn and return error if no traces are found

blkrawverify is prints no errors and returns success if the
requested tracefiles aren't found:

# blkrawverify foobar
Verifying foobar
# echo $?
0

With this change it's a bit more informative:

# ./blkrawverify foobar
Verifying foobar
No tracefiles found for foobar
# echo $?
1

Resolves Red Hat Bugzilla #499581

Reported-by: Milos Malik <mmalik@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon manpage and usage reference invalid "msg-queue-name" option
Eric Sandeen [Mon, 4 May 2009 21:04:51 +0000 (16:04 -0500)]
blkiomon manpage and usage reference invalid "msg-queue-name" option

the blkiomon usage text and man page reference a
"msg-queue-name" option, but getopts is only looking
for "msg-qeueue" - fix the docs to match the code.

Reported-by: Milos Malik <mmalik@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agofix up btrace options & manpage
Eric Sandeen [Mon, 4 May 2009 20:40:53 +0000 (15:40 -0500)]
fix up btrace options & manpage

The btrace script & man page didn't quite match for options,
and the btrace script was missing a few options in the getopts
specification (b&n).  Also, there seems to be no such thing as
a "summarize" option anywhere, so remove it.

Reported-by: /Milos Malik <mmalik@redhat.com <mailto:mmalik@redhat.com>>/
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agomore manpage fixups
Eric Sandeen [Mon, 4 May 2009 20:16:31 +0000 (15:16 -0500)]
more manpage fixups

Fix various typos & inconsistencies in man pages.

I think the manpages could use a general tidy-up, but this
mostly fixes things which I'd consider "errors" vs. style
issues.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agofix max-pkts option inconsistencies
Eric Sandeen [Thu, 30 Apr 2009 15:56:04 +0000 (10:56 -0500)]
fix max-pkts option inconsistencies

This is for RH bug 498426,
btrecord doesn't recognize --max-pkts and --max-packets cmd line options

Usage text and some documentation references "--max-pkts" and
the man page references "--max-packets" but the getopts code
is looking for --max_pkts.

I'm not sure if it's best to make the code match the majority
of the docs, or fix the docs to match the code?  This patch does
the former, but I'm not picky if you want to go the other way :)

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoConverted to using the correct remap entries
Alan D. Brunelle [Thu, 30 Apr 2009 17:10:31 +0000 (13:10 -0400)]
Converted to using the correct remap entries

This follows the kernel changes to the blk_io_trace_remap structure to
better align the names of the structure elements with the real intent of
"from" and "to" (devices & sectors).

See the kernel patches @

http://lkml.org/lkml/2009/4/30/340
http://lkml.org/lkml/2009/4/30/341

(Note: since the ABI order didn't change, old user code will work with
the new kernel code & vice versa.)

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: fix unaligned accesses on ia64
Martin Peschke [Mon, 11 May 2009 06:41:33 +0000 (08:41 +0200)]
blkiomon: fix unaligned accesses on ia64

commit 7aa3ebcec011bfe9cc60d6476252c03376a37551 packed
the blkiomon_stat structure so that traces from one
arch could be analyzed on another (in truth only x86
is different, at least from x86_64/ia64/ppc/ppc64/s390/s390x)

Moving the __u32 device member instead of a new padding field should be
fine.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agofix off-by-one issues in blkiomon.h
Martin Peschke [Mon, 20 Apr 2009 14:07:11 +0000 (16:07 +0200)]
fix off-by-one issues in blkiomon.h

Fix two off-by-one issues. Last bucket of histogram was ommitted by mistake
when being converted to big-endian or when being merged with another bucket.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agofix include statement in stats.h
Martin Peschke [Mon, 20 Apr 2009 14:05:18 +0000 (16:05 +0200)]
fix include statement in stats.h

Some endianess conversion macros have been moved and the corresponding
header file is gone. Need to adapt an include statement in stats.h. The
compiler did not complain because blkiomon.c accidently included blktrace.h
prior to stats.h. Introduced by blktrace rewrite which became version 2.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agohandle race to mkdir at startup
Jeff Moyer [Fri, 17 Apr 2009 06:50:22 +0000 (08:50 +0200)]
handle race to mkdir at startup

I ran into a problem when specifying -D dirname-that-doesnt-yet-exist.
Blktrace would fail, spewing the following messages:

[root@megadeth blktrace]# ./blktrace -d /dev/cciss/c0d1 -D ./2.6.30-rc2-cfq-local
Destination dir ./2.6.30-rc2-cfq-local/ can't be made: 17/File exists
Destination dir ./2.6.30-rc2-cfq-local/ can't be made: 17/File exists
Destination dir ./2.6.30-rc2-cfq-local/ can't be made: 17/File exists
Destination dir ./2.6.30-rc2-cfq-local/ can't be made: 17/File exists
Destination dir ./2.6.30-rc2-cfq-local/ can't be made: 17/File exists
FAILED to start thread on CPU 0: 1/Operation not permitted
FAILED to start thread on CPU 4: 1/Operation not permitted
FAILED to start thread on CPU 5: 1/Operation not permitted
FAILED to start thread on CPU 6: 1/Operation not permitted
FAILED to start thread on CPU 7: 1/Operation not permitted

I tracked it down to the fact that there is no synchronization between
threads when trying to create the output directory.  The fix is simple,
just allow the race to happen and detect it.  It's not really worth
putting in any extra synchronization.  It looks like no place else in
that startup path needs synchronization either.

This patch fixes the issue for me.  I tested it by running the very
command that caused me headaches 100% of the time before.  I also did a
chattr +i on the directory and verified that it would really fail in the
case where it couldn't create the directory.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoFixed plug/unplug logic in btt
Alan D. Brunelle [Mon, 6 Apr 2009 11:30:16 +0000 (07:30 -0400)]
Fixed plug/unplug logic in btt

Was not accounting for unplugged time due to timeout unplugs.

15 years agoWorking on fixing % time q plugged
Alan D. Brunelle [Thu, 2 Apr 2009 16:09:08 +0000 (12:09 -0400)]
Working on fixing % time q plugged

15 years agofix trivial typo in manpage
Eric Sandeen [Thu, 26 Mar 2009 19:44:06 +0000 (20:44 +0100)]
fix trivial typo in manpage

Fix small typo as reported by Kent Baxley <kbaxley@redhat.com>
in Red Hat Bug 489941 - tiny typo in blktrace manpage

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jens Axboe <axboe@carl.(none)>
15 years agoAdd NOTIFY to activity mask
Carl Henrik Lunde [Fri, 6 Feb 2009 11:17:25 +0000 (12:17 +0100)]
Add NOTIFY to activity mask

Allow masking in messages by using "-a notify".

Signed-off-by: Carl Henrik Lunde <chlunde@ping.uio.no>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoBlktrace failed to lock reader threads on the cpu used by the corresponding
Tom Zanussi [Mon, 23 Mar 2009 18:41:00 +0000 (19:41 +0100)]
Blktrace failed to lock reader threads on the cpu used by the corresponding
writer. This resulted in stale data being consumed when blktrace accidently
read at a position that was being written to at the same time. This issue
surfaced as "bad trace magic" warnings emitted by blktrace tools.

The problem occured on an SMP System z machine. The patch fixes the issue.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <axboe@carl.(none)>
15 years agoGenerate matplotlib plots for btt generated data
Alan D. Brunelle [Thu, 12 Mar 2009 14:29:40 +0000 (10:29 -0400)]
Generate matplotlib plots for btt generated data

btt_plot.py: Generate matplotlib plots for BTT generated data files

Files handled:
  AQD   - Average Queue Depth           Running average of queue depths

  BNOS  - Block numbers accessed        Markers for each block

  Q2D   - Queue to Issue latencies      Running averages
  D2C   - Issue to Complete latencies   Running averages
  Q2C   - Queue to Complete latencies   Running averages

Usage:
  btt_plot_aqd.py       equivalent to: btt_plot.py -t aqd
  btt_plot_bnos.py      equivalent to: btt_plot.py -t bnos
  btt_plot_q2d.py       equivalent to: btt_plot.py -t q2d
  btt_plot_d2c.py       equivalent to: btt_plot.py -t d2c
  btt_plot_q2c.py       equivalent to: btt_plot.py -t q2c

Arguments:
  [ -A          | --generate-all   ] Default: False
  [ -L          | --no-legend      ] Default: Legend table produced
  [ -o <file>   | --output=<file>  ] Default: <type>.png
  [ -T <string> | --title=<string> ] Default: Based upon <type>
  [ -v          | --verbose        ] Default: False
  <data-files...>

  The -A (--generate-all) argument is different: when this is specified,
  an attempt is made to generate default plots for all 5 types (aqd, bnos,
  q2d, d2c and q2c). It will find files with the appropriate suffix for
  each type ('aqd.dat' for example). If such files are found, a plot for
  that type will be made. The output file name will be the default for
  each type. The -L (--no-legend) option will be obeyed for all plots,
  but the -o (--output) and -T (--title) options will be ignored.

15 years agoMerge branch 'master' of ssh://axboe@router.home.kernel.dk/data/git/blktrace
Jens Axboe [Wed, 18 Feb 2009 12:11:48 +0000 (13:11 +0100)]
Merge branch 'master' of ssh://axboe@router.home.kernel.dk/data/git/blktrace

15 years agoUpdate Jenkins hash to lookup3() variant
Jens Axboe [Wed, 18 Feb 2009 12:08:08 +0000 (13:08 +0100)]
Update Jenkins hash to lookup3() variant

It both mixes and performs better than lookup2().

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoFixed EAGAIN handling in blktrace.c
Alan D. Brunelle [Tue, 17 Feb 2009 13:48:40 +0000 (08:48 -0500)]
Fixed EAGAIN handling in blktrace.c

EAGAIN was causing header failures in network mode. Added in a usleep
and retried the recv().

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoO_NOATIME isn't always present
Jens Axboe [Tue, 17 Feb 2009 12:39:40 +0000 (13:39 +0100)]
O_NOATIME isn't always present

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agobtt: Added no remap option
Alan D. Brunelle [Fri, 13 Feb 2009 17:38:54 +0000 (12:38 -0500)]
btt: Added no remap option

Trying to run btt on pre-2.6.19 kernels has problems handling the
previous remap PDU - it did not include a proper device-from field.
(Probably should have bumped the blktrace version when we did that.)

This option just tosses those out as it just results in lots of crazy
stuff being handled.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agobtt general cleanup plus valgrind clean
Alan D. Brunelle [Fri, 13 Feb 2009 17:43:45 +0000 (12:43 -0500)]
btt general cleanup plus valgrind clean

Lots of general clean up of code, getting interfaces across different
files to be similar (all are no alloc/free), and made it valgrind clean.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com
15 years agobtt: Missed fopen conversion to my_fopen
Alan D. Brunelle [Thu, 12 Feb 2009 19:28:51 +0000 (14:28 -0500)]
btt: Missed fopen conversion to my_fopen

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com
15 years agoCode review updates
Alan D. Brunelle [Thu, 12 Feb 2009 18:30:47 +0000 (13:30 -0500)]
Code review updates

Re-coding large functions, re-arranging some stuff.

15 years agoReworked blktrace master/thread interface
Alan D. Brunelle [Thu, 12 Feb 2009 16:13:20 +0000 (11:13 -0500)]
Reworked blktrace master/thread interface

Allows parallel initializations.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoCleaned up devs that have no data
Alan D. Brunelle [Thu, 12 Feb 2009 13:01:42 +0000 (08:01 -0500)]
Cleaned up devs that have no data

Working around an issue with older kernels (pre-2.6.19): remaps did not
include device-from, so the pad field is being used for a device which
never has any Q or Ds done to it (it's an invalid ID). This code removes
all such devices before output processing.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoMoved starting of tracing after tracers are going
Alan D. Brunelle [Wed, 11 Feb 2009 21:16:12 +0000 (16:16 -0500)]
Moved starting of tracing after tracers are going

Hold off BLKTRACESTART to threads are ready to consume tracers.

15 years agobtt: fixed open in setup_ifile
Alan D. Brunelle [Wed, 11 Feb 2009 18:40:09 +0000 (13:40 -0500)]
btt: fixed open in setup_ifile

Took my_open & my_fopen code from blktrace 2.0: needed to add in open
resource limit increasing stuff.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoSynchronized trace gathering
Alan D. Brunelle [Wed, 11 Feb 2009 18:23:21 +0000 (13:23 -0500)]
Synchronized trace gathering

Previously, each tracer thread would start gathering traces as soon as
it got going - which might slow down later thread start ups. This change
allows each thread to be ready to gather traces, and then the main
thread starts all the threads gathering at the same time.

15 years agoInvoke gethostbyname once, handle errors better
Alan D. Brunelle [Wed, 11 Feb 2009 18:10:13 +0000 (13:10 -0500)]
Invoke gethostbyname once, handle errors better

Instead of invoking gethostbyname once per client, we only need to do it
once at initialization time. Plus: gethostbyname has a non-standard
errno reporting mechanism, handle this better.

15 years agoAdded accept as a system call needing resource increases
Alan D. Brunelle [Wed, 11 Feb 2009 16:42:09 +0000 (11:42 -0500)]
Added accept as a system call needing resource increases

accept(2) opens a socket, and thus needs to handle EMFILE/ENFILE errors
like other system calls.

15 years agoRewrote blktrace to have a single thread per CPU
Alan D. Brunelle [Mon, 9 Feb 2009 20:11:49 +0000 (15:11 -0500)]
Rewrote blktrace to have a single thread per CPU

Massive changes: mostly around the notion of having much fewer threads
(instead of N(devs) X N(cpus) threads, we'll have just N(cpus)). This
is very important for larger systems (with lots of devices to
trace). A lot of the code was stolen from the original blktrace code,
major changes include:

o  On the client side we only have a single thread per client CPU. Each
thread will then open all device files for that CPU, and use poll to
determine which file needs processing.

o  For network client mode w/ sendfile, this means that a single socket
will carry all data to the remote network server. The network server
side will then distribute its reads off that one socket onto different
trace files.

o  For network client mode w/out sendfile, we fall back to doing things
like piped mode: keep buffers of tracers read in, and then the main
thread will issue these on sockets to the server. In this case, the main
thread will still have a single socket per CPU.

o  For networked mode we added an OPEN concept on the client side: as
soon as the connection to the server is set up, a "header" is sent
signifying that this connection will handle a <cpu, device> tuple. For
each socket opened on the client side, it will send a header per device
being managed. The server side will handle utilize opens to set up
appropriate data structures to handle incoming data streams.

o  For both the OPEN and CLOSE headers the server will acknowledge with
a short write back to the client. This allows the client & server sides
to gracefully close socket connections.

o  I also re-did the resource limitiation issue a bit differently: for
open calls (including socket) or for memory map/lock calls I have
provided a wrapper function that will try to increase specific limits as
needed. The previous method (attempting to do it at the beginning of the
run) fails for network server mode - you don't know at initialization
how many devices and CPUs will be handled.

o  The standard output is slightly different in a few places, if this is
a problem w/ compatibility we can work to rectify that. The command line
argument handling is identical though.

o  Using code stolen from Linux to manipulate doubly-linked lists. I've
found that this makes the code easier to read/write (but may be a bit of
overkill here...)

o  The code passes valgrind quite well (at least for my tests so far).
The only nit has to do with inet_ntoa - but that is out of our control.

Thanks to Stefan Raspl <raspl@linux.vnet.ibm.com> for testing and
finding some issues and for providing suggestions.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoFix btt to handle large numbers of output files
Alan D. Brunelle [Fri, 23 Jan 2009 14:48:21 +0000 (09:48 -0500)]
Fix btt to handle large numbers of output files

Simply bump resource limits if file opens fail, and retry.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
15 years agoIncreased limits to allow for large system runs
Alan D. Brunelle [Fri, 23 Jan 2009 10:04:42 +0000 (11:04 +0100)]
Increased limits to allow for large system runs

On 16-way w/ 104 disks and a 32-way w/ 96 disks, I was getting:

$ sudo blktrace -b 1024 -n 8 -I ../files
./cciss_c1d6.blktrace.10: Too many open files
Failed to start worker threads

Due to the nature of our N(cpus) X N(devices) order of file opens, and
our N(cpus) X N(devices) X N(buffers) X (buffer size) amount of mmaps()
going on we're exceeding both the RLIMIT_NOFILE and RLIMIT_MEMLOCK
limits.

This patch raises limits for RLIMIT_NOFILE and RLIMIT_MEMLOCK to
"infinity", and allows blktrace to handle the large(ish) systems. (If
these settings fail, we "guestimate" about how much we really need.)

There is still an underlying blktrace and/or kernel problem: The
directory /sys/kernel/debug/block/<DSF> where <DSF> is the device that
encountered the limit is left behind (not cleaned up correctly). This
stops blktrace from running a second time (even on another device):

$  ls /sys/kernel/debug/block
cciss_c1d6
$ sudo blktrace /dev/sda
BLKTRACESETUP: No such file or directory
Failed to start trace on /dev/sda

and requires a reboot. (Looking into that next, as this patch - whilst
stopping the original problem from happening - does not address the
secondary problem. And there may be some other ways for the secondary
problem to still occur...)

I also fixed a warning concerning ftruncate's return value being
ignored.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
Signed-off-by: Jens Axboe <axboe@carl.(none)>
15 years agoA couple of min-counters weren't initialised correctly (thrput_r,
Martin Peschke [Wed, 21 Jan 2009 13:58:35 +0000 (14:58 +0100)]
A couple of min-counters weren't initialised correctly (thrput_r,
thrput_w).
We have got a perfectly working init function for this purpose.
Removing partially duplicated code.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoThe git commit 11914a53d2ec2974a565311af327b8983d8c820d added __BLK_TA_ABORT
Martin Peschke [Wed, 21 Jan 2009 13:58:35 +0000 (14:58 +0100)]
The git commit 11914a53d2ec2974a565311af327b8983d8c820d added __BLK_TA_ABORT
to blktrace_api.h. A corresponding addition to the blktrace tools repository
has been missing, breaking the API. Blkparse complained:
"Bad fs action 40010011"

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoAdded no messages option to blkparse.c
Alan D. Brunelle [Mon, 12 Jan 2009 17:29:53 +0000 (18:29 +0100)]
Added no messages option to blkparse.c

Added a new option (-M, --no-msgs) option to blkparse: I have found that
the CFQ I/O scheduler sends a *tremendous* amount of messages, that
bloat the .bin file generated when using the -d option. The file sizes
can shrink by >50% when using the -M option in those case.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agogcc 4.3.2 has started to warn about:
Alan D. Brunelle [Wed, 12 Nov 2008 12:11:22 +0000 (07:11 -0500)]
gcc 4.3.2 has started to warn about:

gcc -Wall -W -O2 -g -I. -I.. -D_GNU_SOURCE -D_LARGEFILE_SOURCE
-D_FILE_OFFSET_BITS=64 -c -o output.o output.c
output.c: In function â€˜output_section_hdr’:
output.c:57: warning: format not a string literal and no format
arguments
output.c: In function â€˜__output_pip_avg’:
output.c:496: warning: format not a string literal and no format
arguments
output.c:496: warning: format not a string literal and no format
arguments

so this patch cleans this up.

15 years agoMerge branch 'add-P'
Alan D. Brunelle [Tue, 11 Nov 2008 18:51:33 +0000 (13:51 -0500)]
Merge branch 'add-P'

15 years agoMerge branch 'fix-m'
Alan D. Brunelle [Tue, 11 Nov 2008 18:50:23 +0000 (13:50 -0500)]
Merge branch 'fix-m'

15 years agoAdded -P to create a data file w/ Q, D and C per line
Alan D. Brunelle [Tue, 11 Nov 2008 18:46:13 +0000 (13:46 -0500)]
Added -P to create a data file w/ Q, D and C per line

Easy parsing for graph creation

15 years agoMerge branch 'fix-m' into add-P
Alan D. Brunelle [Tue, 11 Nov 2008 18:42:55 +0000 (13:42 -0500)]
Merge branch 'fix-m' into add-P

15 years agoFixed 'M' displays on per-io output and added in I/O separator
Alan D. Brunelle [Tue, 11 Nov 2008 18:41:05 +0000 (13:41 -0500)]
Fixed 'M' displays on per-io output and added in I/O separator

15 years ago Fixed segfault in aqd.c : need to check for NULL (not requested)
Alan D. Brunelle [Tue, 11 Nov 2008 18:40:10 +0000 (13:40 -0500)]
 Fixed segfault in aqd.c : need to check for NULL (not requested)

15 years agoAdded in -z to provide running waiting-for-issue latencies
Alan D. Brunelle [Mon, 10 Nov 2008 15:35:44 +0000 (10:35 -0500)]
Added in -z to provide running waiting-for-issue latencies

15 years agoMerge branch 'master' of ssh://alanbrunelle@git.kernel.dk/data/git/blktrace
Alan D. Brunelle [Mon, 10 Nov 2008 15:34:02 +0000 (10:34 -0500)]
Merge branch 'master' of ssh://alanbrunelle@git.kernel.dk/data/git/blktrace

15 years agoSet release version 1.0.0 blktrace-1.0.0
Jens Axboe [Thu, 30 Oct 2008 14:06:01 +0000 (15:06 +0100)]
Set release version 1.0.0

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoUpdate rbtree to version with unified parent + color
Jens Axboe [Thu, 30 Oct 2008 14:04:20 +0000 (15:04 +0100)]
Update rbtree to version with unified parent + color

This saves 4-8 bytes per node, which can be quite a bit of memory
with blktrace.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoMoved btrecord/btreplay to version 1.0.0
Alan D. Brunelle [Wed, 29 Oct 2008 19:04:03 +0000 (15:04 -0400)]
Moved btrecord/btreplay to version 1.0.0

15 years agoblkiomon: add through-put statistics
Martin Peschke [Tue, 28 Oct 2008 16:08:12 +0000 (17:08 +0100)]
blkiomon: add through-put statistics

Add accounting of per-request throughput in bytes per millisecond
both for read ad write I/O.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: separate statistics for read and write requests
Martin Peschke [Tue, 28 Oct 2008 16:08:10 +0000 (17:08 +0100)]
blkiomon: separate statistics for read and write requests

Split min/max/avg statistics for request sizes and dispatch-to-completion
latencies into separate statistics for read and write requests.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: fix some debug messages
Martin Peschke [Tue, 28 Oct 2008 16:08:07 +0000 (17:08 +0100)]
blkiomon: fix some debug messages

Cleaning up error messages. Some perror()'s didn't make sense.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: fix trace debug output
Martin Peschke [Tue, 28 Oct 2008 16:08:05 +0000 (17:08 +0100)]
blkiomon: fix trace debug output

Removed leftovers of trace tree and made debug code work by using trace
hash instead of trace tree.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: fix unit in histogram output
Martin Peschke [Tue, 28 Oct 2008 16:08:02 +0000 (17:08 +0100)]
blkiomon: fix unit in histogram output

Fix unit of request sizes as printed in histogram (it's bytes not kilobytes).

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: fix cross-arch data analysis issue
Martin Peschke [Tue, 28 Oct 2008 16:07:57 +0000 (17:07 +0100)]
blkiomon: fix cross-arch data analysis issue

This fixes cross-arch issues. Binary data gathered on System z could not
be read on x86.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: drv_data traces pass-through
Martin Peschke [Mon, 20 Oct 2008 16:14:21 +0000 (18:14 +0200)]
blkiomon: drv_data traces pass-through

This patch adds pass-through support for device driver specific traces
to blkiomon. This way we can aggregate block I/O statistics and device
driver specific statistics at the same time.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkparse: add hint for discarded drv_data traces
Martin Peschke [Fri, 17 Oct 2008 13:09:05 +0000 (15:09 +0200)]
blkparse: add hint for discarded drv_data traces

Display an informational message on blkparse exit to notify users that
additional data was available which would require to be dumped to binary
output.

Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoAdded in -L option - output periodic latency information
Alan D. Brunelle [Thu, 16 Oct 2008 16:03:45 +0000 (12:03 -0400)]
Added in -L option - output periodic latency information

15 years agoAdded in -Q / --active-queue-depth option
Alan D. Brunelle [Thu, 16 Oct 2008 14:53:07 +0000 (10:53 -0400)]
Added in -Q / --active-queue-depth option

This will output a data file containing the time stamp and number of
I/Os issued to underlying drivers per device. It will give you an idea
as to how many I/Os are being actively worked per device at any time
during the run.

15 years agoAdd driver data support
Stefan Raspl [Thu, 16 Oct 2008 06:14:17 +0000 (08:14 +0200)]
Add driver data support

Adds a new type of action 'drv_data' for blktrace to handle binary
driver-specific data. Since the data is binary, blkparse will only put it in
a binary file, not in the regular human-readable output.

Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Martin Peschke <mpeschke@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblktrace: accept -v (lower case) for version info as well
Jens Axboe [Thu, 16 Oct 2008 06:05:53 +0000 (08:05 +0200)]
blktrace: accept -v (lower case) for version info as well

Christof Schmitt <christof.schmitt@de.ibm.com> points out that the
documentation uses -v but blktrace supports only -V, so change
blktrace to accept both cases.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblkiomon: I/O monitor
Martin Peschke [Thu, 16 Oct 2008 05:48:25 +0000 (07:48 +0200)]
blkiomon: I/O monitor

blkiomon periodicaly generates per devive request size and request latency
statistics from blktrace data. It provides histograms as well as data that
can be used to calculate min, max, average and variance. For this purpose,
it consumes D and C traces read from stdin.

There are options for binary output and human-readable output to files and
stdout. Output to a message queue is supported as well.

#blktrace /dev/sdw -a issue -a complete -w 200 -o - | blkiomon -I 8 -h -

time: Tue Sep 30 17:39:25 2008
device: 65,96
requests: read 62, write 40, bidir: 0
sizes: num 102, min 4096, max 430080, sum 13312000, squ 3102442782720,
 avg 130509.8, var 13383296793.3
d2c: num 102, min 393, max 14261, sum 359441, squ 2830211755, avg 3523.9,
 var 15329081.8
sizes histogram (in kB):
            0:     0         1024:     0         2048:     0         4096:     6
         8192:     0        16384:    15        32768:     4        65536:    24
       131072:    11       262144:    30       524288:    12      1048576:     0
      2097152:     0      4194304:     0      8388608:     0    > 8388608:     0
d2c histogram (in usec):
            0:     0            8:     0           16:     0           32:     0
           64:     0          128:     0          256:     0          512:    13
         1024:    21         2048:    27         4096:    14         8192:     8
        16384:    19        32768:     0        65536:     0       131072:     0
       262144:     0       524288:     0      1048576:     0      2097152:     0
      4194304:     0      8388608:     0     16777216:     0     33554432:     0
    >33554432:     0

time: Tue Sep 30 17:39:33 2008
device: 65,96
requests: read 312, write 47, bidir: 0
sizes: num 359, min 4096, max 430080, sum 13197312, squ 1575816790016,
 avg 36761.3, var 3038067547.5
d2c: num 359, min 294, max 9211, sum 387134, squ 1262489694, avg 1078.4,
 var 2353807.5
sizes histogram (in kB):
            0:     0         1024:     0         2048:     0         4096:    32
         8192:    17        16384:   133        32768:    87        65536:    59
       131072:     9       262144:    18       524288:     4      1048576:     0
      2097152:     0      4194304:     0      8388608:     0    > 8388608:     0
d2c histogram (in usec):
            0:     0            8:     0           16:     0           32:     0
           64:     0          128:     0          256:     0          512:   129
         1024:   164         2048:    33         4096:    15         8192:    13
        16384:     5        32768:     0        65536:     0       131072:     0
       262144:     0       524288:     0      1048576:     0      2097152:     0
      4194304:     0      8388608:     0     16777216:     0     33554432:     0
    >33554432:     0

Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoRemoved excessive amounts of seek modes (for random sets of I/Os)
Alan D. Brunelle [Fri, 10 Oct 2008 12:40:57 +0000 (08:40 -0400)]
Removed excessive amounts of seek modes (for random sets of I/Os)

When doing a random load, we'd get a LARGE amount of single-seek buckets,
this patch just notes that fact, without dumping all the data...

15 years agoMerge branch 'master' of ssh://alanbrunelle@git.kernel.dk/data/git/blktrace
Alan D. Brunelle [Fri, 10 Oct 2008 12:40:21 +0000 (08:40 -0400)]
Merge branch 'master' of ssh://alanbrunelle@git.kernel.dk/data/git/blktrace

15 years agospec file tweak
Nathan Scott [Fri, 26 Sep 2008 09:05:51 +0000 (11:05 +0200)]
spec file tweak

I found I needed this tweak to make the spec file usable on a RHEL5
system (I hit rpmbuild errors about unpackaged files without this).
Not sure if its correct though, maybe someone with stronger rpm-fu
than I have could take a closer look?

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoMerge branch 'master' of ssh://axboe@router.home.kernel.dk/data/git/blktrace
Jens Axboe [Fri, 26 Sep 2008 09:05:16 +0000 (11:05 +0200)]
Merge branch 'master' of ssh://axboe@router.home.kernel.dk/data/git/blktrace

15 years agoman page typo
Nathan Scott [Fri, 26 Sep 2008 09:05:13 +0000 (11:05 +0200)]
man page typo

This fixes up a formatting problem when displaying the blktrace
man pages - just a typo (bold mode not correctly terminated, so
it flows on until the next bold-mode-termination a few lines on).

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoAdded in %done for btt
Alan D. Brunelle [Tue, 23 Sep 2008 12:06:41 +0000 (08:06 -0400)]
Added in %done for btt

15 years agoAdd documentation of 'D' discard operation
David Woodhouse [Fri, 15 Aug 2008 09:12:20 +0000 (11:12 +0200)]
Add documentation of 'D' discard operation

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoblktrace: support discard requests
David Woodhouse [Fri, 15 Aug 2008 08:44:39 +0000 (10:44 +0200)]
blktrace: support discard requests

Add support for discard requests to blktrace userspace tools.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agospelling and grammar fixes for btreplay.tex
Jeff Moyer [Tue, 1 Jul 2008 11:35:29 +0000 (13:35 +0200)]
spelling and grammar fixes for btreplay.tex

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
15 years agoPut message notes from kernel into a separate file for easy tracking
Alan D. Brunelle [Tue, 27 May 2008 14:39:14 +0000 (10:39 -0400)]
Put message notes from kernel into a separate file for easy tracking

Also made made_dev_hdr standard for all usages.

15 years agoAdded in new message updates to the documentation.
Alan D. Brunelle [Tue, 27 May 2008 12:34:13 +0000 (08:34 -0400)]
Added in new message updates to the documentation.

15 years agoAdded in handling of MESSAGE notes
Alan D. Brunelle [Tue, 27 May 2008 12:19:57 +0000 (08:19 -0400)]
Added in handling of MESSAGE notes

Sample output:

  8,16   1   691118    17.417000000     0  C   R 2660776 + 8 [0]
  8,16   1   691119    17.417000000     0  D   R 2660792 + 8 [swapper]
  8,16   1   691120    17.417000000  4688  U   N [dd] 42
  8,16   1        0    17.418000000     0  m   N elv switch: noop
  8,16   1   691121    17.418000000  4638  C   R 2660784 + 8 [0]
  8,16   1   691122    17.418000000  4638  D   R 2660800 + 8 [bash]
  8,16   1   691123    17.418000000  4638  C   R 2660792 + 8 [0]

Thanks to Carl Henrik Lunde <chlunde@ping.uio.no> for adding in sequence
printing & time-stamp correction.

15 years agoHandled no difference in seek times
Alan D. Brunelle [Wed, 21 May 2008 19:55:57 +0000 (15:55 -0400)]
Handled no difference in seek times

For some reason recent kernels (2.6.25.4, for example) we've lost a lot
of resolution in our blktrace times. This can result in lots of things
happening "simultaneously." This change at least tries to handle the
case where all the seeks happen at once.

Probably have other issues that need to be looked into...

15 years agoAdded in -m option, seeks-per-second
Alan D. Brunelle [Wed, 21 May 2008 14:55:20 +0000 (10:55 -0400)]
Added in -m option, seeks-per-second

btt can now output data files containing seeks-per-second information.

16 years agoblkparse: cope with missing process notify event
Jens Axboe [Sun, 18 May 2008 18:55:25 +0000 (20:55 +0200)]
blkparse: cope with missing process notify event

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoFixed percentage calculations for phases of an I/O
Alan D. Brunelle [Mon, 12 May 2008 13:56:33 +0000 (09:56 -0400)]
Fixed percentage calculations for phases of an I/O

16 years agoAdded S2G times + fixed up -X output to include X2X
Alan D. Brunelle [Fri, 9 May 2008 17:46:47 +0000 (13:46 -0400)]
Added S2G times + fixed up -X output to include X2X

Including Q2Q, Q2G, S2G, G2I, Q2M, I2D, M2D, D2C, Q2C.

S2G is part of Q2G, and shows the number of times we had to sleep to
get a request.

Ignored 0-byte I/Os - coming from barrier I/Os...

16 years agoAdded -X option - generate easily parseable file
Alan D. Brunelle [Thu, 8 May 2008 19:28:32 +0000 (15:28 -0400)]
Added -X option - generate easily parseable file

Writes a portion of the default output into a separate file that more
easily parsed.

16 years agoAdd -x accellerator option
Luis Useche [Mon, 5 May 2008 18:53:13 +0000 (20:53 +0200)]
Add -x accellerator option

This patch adds a new functionality to the btreplay tool, the -x option.
This parameter accelerate the replication by the factor specified. This
means that the stall time is divided by the number introduced.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoFix problem with -w option
Luis Useche [Mon, 5 May 2008 18:53:10 +0000 (20:53 +0200)]
Fix problem with -w option

This patch fixes the problem when the -w option is used in the file mode
(i.e., no fifo mode). It just consists of moving the checking of the
stopwatch_end after the time is updated with genesis. This also includes
the checking of the stopwatch_start.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoeliminate check of empty -F format
Luis Useche [Mon, 5 May 2008 18:53:07 +0000 (20:53 +0200)]
eliminate check of empty -F format

This patch eliminates the checking of -F format when it is empty. I am
using this in order to blank out the events that I do not want for
certain act mask. Note that there is no real motivation to have this
checking.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblkparse: Introduce optional accounting of PC requests
Christof Schmitt [Wed, 2 Apr 2008 07:21:08 +0000 (09:21 +0200)]
blkparse: Introduce optional accounting of PC requests

The current blkparse lists only PC requests, but does not account for
number of PC requests and the size of the attached data.  This patch
introduces optional accounting for PC requests.  If any PC event was
encountered, the summary output will also include data from PC
requests.

For PC completion events the amount of transferred data is not
available, since the kernel uses the data_len field as residual bytes
count on event completion.

Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblkparse: Add PC requests to depth trace
Christof Schmitt [Wed, 2 Apr 2008 07:21:08 +0000 (09:21 +0200)]
blkparse: Add PC requests to depth trace

The PC requests use the same queue and also contribute to the
queue length, so they should be added to the queue depth trace.

Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoFix section of btrecord and btreplay man pages
Bas Zoetekouw [Fri, 22 Feb 2008 11:58:12 +0000 (12:58 +0100)]
Fix section of btrecord and btreplay man pages

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoDon't like btrecord against libaio and librt, as it doesn't use any of their symbols
Bas Zoetekouw [Fri, 22 Feb 2008 11:51:42 +0000 (12:51 +0100)]
Don't like btrecord against libaio and librt, as it doesn't use any of their symbols

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoReally commit the changes to btt.1
Bas Zoetekouw [Fri, 22 Feb 2008 10:27:04 +0000 (11:27 +0100)]
Really commit the changes to btt.1

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdded man page for bno_plot and updated btt man page to refer to bno_plot
Bas Zoetekouw [Fri, 22 Feb 2008 10:25:51 +0000 (11:25 +0100)]
Added man page for bno_plot and updated btt man page to refer to bno_plot

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoCleanups: Fixed IOPs in btt left over at end of run
Alan D. Brunelle [Wed, 13 Feb 2008 16:46:22 +0000 (11:46 -0500)]
Cleanups: Fixed IOPs in btt left over at end of run

o  Using valgrind, determined we had Q IOPs left over that weren't
   used. Added "all" list, and then deleted these at end.

o  Removed old debug stuff (COUNT_IOS, DEBUG, ...)

o  Fixed a bunch of white space at end of lines.

Signed-off-by: Alan D. Brunelle <alan.brunelle@hp.com>
16 years agobtt: fix missing cleanup call
Aaron Carroll [Wed, 13 Feb 2008 07:48:14 +0000 (18:48 +1100)]
btt: fix missing cleanup call

clean_args() exists but is never called, thus any files opened in args remain
open while the buffers backing those files are closed.

Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
16 years agobtt: fix iostat interval default
Aaron Carroll [Wed, 13 Feb 2008 07:46:40 +0000 (18:46 +1100)]
btt: fix iostat interval default

Change the default iostat interval from 1ns to 1s, which is the
default implied by btt.1 and common sense.

Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
16 years agoAdded info about bno_plot.py and clean ups
Alan D. Brunelle [Tue, 12 Feb 2008 13:58:25 +0000 (08:58 -0500)]
Added info about bno_plot.py and clean ups

o  Added description of bno_plot.py, including sample screenshot
o  Various spelling clean ups
o  Updated sample btt output
o  Cleaned up table of contents

16 years agoMerge branch 'bno_plot.submit'
Alan D. Brunelle [Fri, 8 Feb 2008 15:03:13 +0000 (10:03 -0500)]
Merge branch 'bno_plot.submit'

16 years agoAdded new IOs per unplug table
Alan D. Brunelle [Tue, 5 Feb 2008 22:31:21 +0000 (17:31 -0500)]
Added new IOs per unplug table

==================== Plug Information ====================

       DEV |    # Plugs # Timer Us  | % Time Q Plugged
---------- | ---------- ----------  | ----------------
 (  8,  0) |       1171(       123) |   0.280946640%
 (  8, 32) |          4(         0) |   0.000325469%
---------- | ---------- ----------  | ----------------
   Overall |    # Plugs # Timer Us  | % Time Q Plugged
   Average |        587(        61) |   0.140636055%

       DEV |    IOs/Unp   IOs/Unp(to)
---------- | ----------   ----------
 (  8,  0) |        9.2          8.8
 (  8, 32) |        2.5          0.0
---------- | ----------   ----------
       DEV |    IOs/Unp   IOs/Unp(to)
   Overall |        9.2          8.8

16 years agoAdded bno_plot.py - generate interactive 3D plot of IO blocks and sizes
Alan D. Brunelle [Tue, 5 Feb 2008 18:46:51 +0000 (13:46 -0500)]
Added bno_plot.py - generate interactive 3D plot of IO blocks and sizes

bno_plot.py
[ -h | --help       ]
[ -K | --keys-below ]
[ -v | --verbose    ]
[ <file...>         ]

Utilizes gnuplot to generate a 3D plot of the block number output
from btt.  If no <files> are specified, it will utilize all files
generated after btt was run with -B blknos (meaning: all files of the
form blknos*[rw].dat).

The -K option forces bno_plot.py to put the keys below the graph,
typicall all keys for input files are put in the upper right corner of
the graph. If the number of devices exceed 10, then bno_plot.py will
automatically push the keys under the graph.

To exit the plotter, enter 'quit' or ^D at the 'gnuplot> ' prompt.

16 years agoUNPLUG does the timing stuff, UNPLUG TIMEOUT only does timeout
Alan D. Brunelle [Thu, 31 Jan 2008 22:10:52 +0000 (17:10 -0500)]
UNPLUG does the timing stuff, UNPLUG TIMEOUT only does timeout

Each UNPLUG TIMEOUT should be followed by an UNPLUG, so we were getting
double information on time outs.

16 years agoFixed excess bucket information for unplug histograms.
Alan D. Brunelle [Thu, 31 Jan 2008 18:02:24 +0000 (13:02 -0500)]
Fixed excess bucket information for unplug histograms.