fio.git
11 years agolfsr: Fix spin related bug
Alex Pyrgiotis [Sun, 10 Mar 2013 09:58:05 +0000 (10:58 +0100)]
lfsr: Fix spin related bug

On the previous patch on cycle detection, we would incremet spin value
when:

    num_vals % cycle_length == 0

which would evaluate always to true for the first iteration since we
start with num_vals = 0. Only one corner-case is affected by this bug
and this patch should fix it.

Signed-off-by: Alex Pyrgiotis <apyrgio@grnet.gr>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agolfsr: indent table
Jens Axboe [Fri, 8 Mar 2013 18:12:15 +0000 (19:12 +0100)]
lfsr: indent table

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd a simple test for LFSR generator
Alex Pyrgiotis [Fri, 8 Mar 2013 12:37:04 +0000 (14:37 +0200)]
Add a simple test for LFSR generator

Adds a simple test suite to check the speed of the LFSR
generator and verify its results. Just run:

    make t/lfsr-test
    ./t/lfsr-test

to compile the test suite and print its usage

Signed-off-by: Alex Pyrgiotis <apyrgio@grnet.gr>
 create mode 100644 t/lfsr-test.c
Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoImprove LFSR implementation
Alex Pyrgiotis [Fri, 8 Mar 2013 12:37:03 +0000 (14:37 +0200)]
Improve LFSR implementation

Changes:

1. Use Galois LFSR instead of Fibonacci LFSR.
2. Use XNOR gates instead of XOR gates.
3. Add tap sizes for LFSRs ranging from 3-bits to 15-bits.
4. Add spin parameter.

Rationale:

1. Fibonacci LFSRs have the following drawbacks:

   a. Their next state can not be computed in one cycle, since the input
      bit must be propagated serially through the XOR gates. Galois
  LFSRs however, can be computed instantly with XOR-masks.

   b. Their state changes cannot be considered "irregular", even by I/O
      standards. Actually, if the current state of an n-bit LFSR is x,
  then the next will either be (x >> 1) or (2^n + (x >> 1)). Galois
  LFSRs have instead their XOR gates interleaved with their bits,
  which means that the inner bits are changed as well, besides of
  the shifting. If the number of taps is z, this means that the
  different outcomes are 2^(z + 1).

2. An LFSR with XOR gates has the all-zeroes state as illegal. Since
   zero is valid for most I/O operations, it would be more intuitive to
   use XNOR gates, that have as the all-ones state as illegal.

3. Allow smaller I/O benchmarks to use LFSRs.

4. The spin parameter follows the same rationale as in 1b. To make the
   LFSR outcome "appear" less predictable, we can spin internally the
   LFSR state and produce the i-th number. To understand the spin
   parameter, consider the following state sequence of a 3-bit LFSR:

                             0, 2, 3, 5, 6, 1, 4
   Same LFSR, spin value 2:  0, 3, 6, 4, 2, 5, 1
   But what is the benefit from using spin? Well, the benefits are
   two-fold:

   a. For the same I/O size, we can create a different I/O sequences.

   b. Following the rationale of 1b, we can have more variable outcomes.
      If the spin value is "i" and the taps are "z", the number of
  different outcomes become i * 2^(z + 1).

Signed-off-by: Alex Pyrgiotis <apyrgio@grnet.gr>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix skewed latencies for rate IOPS
SEOKYOUNG KO [Thu, 7 Mar 2013 11:34:31 +0000 (12:34 +0100)]
Fix skewed latencies for rate IOPS

- when -rate_iops is specified, FIO periodically calls usleep() to limit IOPS

- Before usleep(), FIO always wait until the completion of all pending I/O

- For all I/O completions, FIO shows erroneous behavior of logging their
  latency, with that of the longest latency IO sample.

- w/ QD=8/ mixed R:W=33:66, up to 7 samples may get false latency log when
  waiting all I/O completions => False report of latency distribution

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix rate iops with ramp time
SEOKYOUNG KO [Thu, 7 Mar 2013 11:31:04 +0000 (12:31 +0100)]
Fix rate iops with ramp time

When the ramp time is over, fio resets the ->io_issues[]
counters, but not the rwmix counters. This causes a problem
for mixed read/write workloads that are rate limited and
include a ramp time.

Fix it by resetting rwmix_issues with the regular issues.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd --parse-only command line option
Jens Axboe [Thu, 7 Mar 2013 10:31:20 +0000 (11:31 +0100)]
Add --parse-only command line option

Sometimes it's useful just being able to verify whether a job
config is valid or not, without actually starting any jobs.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: fixup __sync_fetch_and_add()
Jens Axboe [Thu, 28 Feb 2013 21:13:29 +0000 (22:13 +0100)]
configure: fixup __sync_fetch_and_add()

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: add general libs for Solaris
Jens Axboe [Thu, 28 Feb 2013 21:10:22 +0000 (22:10 +0100)]
configure: add general libs for Solaris

Assume we just need them, this makes us find inet_aton()
on Solaris too.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agosolaris: ensure that TCP_NODELAY gets picked up
Jens Axboe [Thu, 28 Feb 2013 19:34:31 +0000 (20:34 +0100)]
solaris: ensure that TCP_NODELAY gets picked up

Use proper libs when linking, then we can also remove
those from the Makefile.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMerge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Thu, 28 Feb 2013 19:08:12 +0000 (20:08 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio

11 years agopowerpc: use mfspr for Time Base Register reads
Steven Noonan [Thu, 28 Feb 2013 19:08:06 +0000 (20:08 +0100)]
powerpc: use mfspr for Time Base Register reads

The 'mftb' and 'mftbu' instructions are deprecated in Power ISA v2.06 in favor
of mfspr. Since CPUs supporting mftb/mftbu also support the TBR SPRs, there's
no known reason to keep mftb/mftbu.

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix compile of RDMA engine for SunOS 5.x
Jens Axboe [Thu, 28 Feb 2013 07:28:05 +0000 (08:28 +0100)]
Fix compile of RDMA engine for SunOS 5.x

- byteswap.h include that doesn't exist there. Kill it, we
  have no use for it.

- No RLIMIT_MEMLOCK on SunOS. The use in the engine is a bit
  suspect, so not a problem if we don't have it. Add configure
  check for that.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agonet: fix warning for systems without TCP_NODELAY
Jens Axboe [Thu, 28 Feb 2013 07:24:23 +0000 (08:24 +0100)]
net: fix warning for systems without TCP_NODELAY

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoRevert "debug: only do getpid() if we have to"
Jens Axboe [Wed, 27 Feb 2013 11:32:43 +0000 (12:32 +0100)]
Revert "debug: only do getpid() if we have to"

This reverts commit 8037423f08649a7c3378dadf95327d6553105264.

11 years agoFIO-VERSION-GEN: use `foo` instead of $(foo)
Jens Axboe [Tue, 26 Feb 2013 20:59:33 +0000 (21:59 +0100)]
FIO-VERSION-GEN: use `foo` instead of $(foo)

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMerge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Tue, 26 Feb 2013 20:55:43 +0000 (21:55 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio

11 years agoconfigure: use `foo` instead of $(foo)
Jens Axboe [Tue, 26 Feb 2013 20:55:08 +0000 (21:55 +0100)]
configure: use `foo` instead of $(foo)

Seems to work everywhere, whereas $(foo) does not work on
some variants of FreeBSD.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agopowerpc: harden the clock ATB probe a bit
Jens Axboe [Tue, 26 Feb 2013 15:11:49 +0000 (16:11 +0100)]
powerpc: harden the clock ATB probe a bit

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agopowerpc: use ATB clock, if it's available
Jens Axboe [Tue, 26 Feb 2013 13:43:02 +0000 (14:43 +0100)]
powerpc: use ATB clock, if it's available

The ATB is more precise, but we don't know if it's available.
Check by forking off a process that attempts to use it, if
we die with SIGILL, then we know it's not there.

Based on code from Steven Noonan.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agogettime: print 64-bit variable with ULL
Jens Axboe [Tue, 26 Feb 2013 12:54:20 +0000 (13:54 +0100)]
gettime: print 64-bit variable with ULL

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agodebug: make __dprint() have printf() like argument type checking
Jens Axboe [Tue, 26 Feb 2013 07:16:43 +0000 (08:16 +0100)]
debug: make __dprint() have printf() like argument type checking

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agodebug: only do getpid() if we have to
Jens Axboe [Mon, 25 Feb 2013 13:02:42 +0000 (14:02 +0100)]
debug: only do getpid() if we have to

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agogettime: use 32-bit atomic sequences
Jens Axboe [Mon, 25 Feb 2013 09:23:58 +0000 (10:23 +0100)]
gettime: use 32-bit atomic sequences

Not all platforms have 64-bit wide atomic sync_and_fetch(). If
we just check for overflow, it should be OK to use a 32-bit
sequence number.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFixup wrong types for dprint()
Jens Axboe [Mon, 25 Feb 2013 09:18:33 +0000 (10:18 +0100)]
Fixup wrong types for dprint()

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agogettime: add some sanity checks to platform clock
Jens Axboe [Sun, 24 Feb 2013 20:29:35 +0000 (21:29 +0100)]
gettime: add some sanity checks to platform clock

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMerge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Sun, 24 Feb 2013 13:24:16 +0000 (14:24 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio

11 years agoAdd LICENSE to Windows installer, fix filename extensions and update URL.
Bruce Cran [Sun, 24 Feb 2013 11:15:26 +0000 (11:15 +0000)]
Add LICENSE to Windows installer, fix filename extensions and update URL.

Add the LICENSE file to the Windows installer, change installed filename
extensions from .rtf to .txt (assume a text editor is installed that
understands LF) and fix the "update url" to be http://bluestop.org/fio/
since that's where new Windows installers are.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd section 9 to HOWTO table of contents.
Bruce Cran [Sun, 24 Feb 2013 11:09:11 +0000 (11:09 +0000)]
Add section 9 to HOWTO table of contents.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd --eta-newline option
Jens Axboe [Fri, 22 Feb 2013 19:48:56 +0000 (20:48 +0100)]
Add --eta-newline option

For certain situations, it's handy to force a new line in
the ETA output. It allows you to easily track what
happened, without having to do detailed logging.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFio 2.0.14 fio-2.0.14
Jens Axboe [Thu, 21 Feb 2013 18:12:32 +0000 (19:12 +0100)]
Fio 2.0.14

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix problem with terminating on unaligned sizes
Jens Axboe [Thu, 21 Feb 2013 14:18:17 +0000 (15:18 +0100)]
Fix problem with terminating on unaligned sizes

Three separate little issues:

- Don't round up the number of blocks needed. That means we end
  up with potential partial blocks, which we can never do IO to.

- Fix an axmap bug where we only check against the specific bit
  in the mask, but we really want to check for "this bit or any
  higher bit". This makes axmap_next_free() behave more like it
  should, instead of failing way too often.

- For the almost-full case, we can return a next available bit
  that is just outside the allowed range. Punt to first-free for
  that case.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd FreeBSD cpu affinity support.
Bruce Cran [Thu, 21 Feb 2013 12:39:24 +0000 (12:39 +0000)]
Add FreeBSD cpu affinity support.

Since FreeBSD has different thread and process IDs, assume that fio_setaffinity
will always be called with a tid and fio_getaffinity with a pid.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoDeclare 'prev' and 'this' outside the loop to avoid clang warning about 'prev' being...
Bruce Cran [Thu, 21 Feb 2013 13:16:17 +0000 (14:16 +0100)]
Declare 'prev' and 'this' outside the loop to avoid clang warning about 'prev' being uninitialized

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFixup ia64 32-bit register clock cycle read
Jens Axboe [Thu, 21 Feb 2013 11:57:56 +0000 (12:57 +0100)]
Fixup ia64 32-bit register clock cycle read

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix x86 32-bit wrong variable type for rdtsc
Jens Axboe [Thu, 21 Feb 2013 11:39:42 +0000 (12:39 +0100)]
Fix x86 32-bit wrong variable type for rdtsc

Bruce points out, that we are putting the result into a 32-bit
variable. That is incorrect, make it a proper ULL instead.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoREADME: kill reference to non-functional SUSE package page
Jens Axboe [Thu, 21 Feb 2013 10:03:28 +0000 (11:03 +0100)]
README: kill reference to non-functional SUSE package page

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd recommendation that fio be run in a native console on Windows.
Bruce Cran [Wed, 20 Feb 2013 20:35:16 +0000 (20:35 +0000)]
Add recommendation that fio be run in a native console on Windows.

Windows console applications often don't run correctly under Cygwin shells such
as mintty because of the way they're implemented. Add a recommendation to the
README to run fio under a native console such as the Command Prompt or console2
to avoid these problems.

http://code.google.com/p/mintty/issues/detail?id=56 documents the problem.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoUpdate all Windows files except dobuild.cmd to use LF line endings
Bruce Cran [Thu, 21 Feb 2013 10:02:41 +0000 (11:02 +0100)]
Update all Windows files except dobuild.cmd to use LF line endings

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agocheck output_format before calling show_idle_prof_stats()
Huadong Liu [Thu, 21 Feb 2013 09:24:20 +0000 (10:24 +0100)]
check output_format before calling show_idle_prof_stats()

Do the proper checks for output format type and whether we are
a server backend or not.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agot/axmap: add pre/post bit set checks
Jens Axboe [Thu, 21 Feb 2013 09:23:51 +0000 (10:23 +0100)]
t/axmap: add pre/post bit set checks

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMerge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Mon, 18 Feb 2013 19:09:28 +0000 (20:09 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio

11 years agoFix replay of IO pattern with multiple open files
Nikolaus Jeremic [Mon, 18 Feb 2013 18:44:41 +0000 (19:44 +0100)]
Fix replay of IO pattern with multiple open files

Nikolaus writes:

I've tried to replay I/Os from a textual iolog file to a couple of files
using fio 2.0.13 on Linux. However, all I/Os were performed to the last
opened file, which is '/mnt/ssd5/rand-write.1.0_1_2' in this example .
My jobfile and iolog files contain the following:

[global]
name=replay_filetraces
description=Replay file traces
ioengine=libaio
iodepth=6
direct=1
continue_on_error=1

[replay]
numjobs=1
read_iolog=file_io_patterns
unlink=0

fio version 2 iolog
/mnt/ssd0/rand-write.1.0_0_0 add
/mnt/ssd1/rand-write.1.0_0_1 add
/mnt/ssd2/rand-write.1.0_0_2 add
/mnt/ssd3/rand-write.1.0_1_0 add
/mnt/ssd4/rand-write.1.0_1_1 add
/mnt/ssd5/rand-write.1.0_1_2 add
/mnt/ssd0/rand-write.1.0_0_0 open
/mnt/ssd1/rand-write.1.0_0_1 open
/mnt/ssd2/rand-write.1.0_0_2 open
/mnt/ssd3/rand-write.1.0_1_0 open
/mnt/ssd4/rand-write.1.0_1_1 open
/mnt/ssd5/rand-write.1.0_1_2 open
/mnt/ssd0/rand-write.1.0_0_0 write 0 65536
/mnt/ssd1/rand-write.1.0_0_1 write 0 65536
/mnt/ssd2/rand-write.1.0_0_2 write 0 65536
/mnt/ssd3/rand-write.1.0_1_0 write 0 65536
/mnt/ssd4/rand-write.1.0_1_1 write 0 65536
/mnt/ssd5/rand-write.1.0_1_2 write 0 65536
/mnt/ssd0/rand-write.1.0_0_0 write 65536 65536
/mnt/ssd1/rand-write.1.0_0_1 write 65536 65536
/mnt/ssd2/rand-write.1.0_0_2 write 65536 65536
/mnt/ssd3/rand-write.1.0_1_0 write 65536 65536
/mnt/ssd4/rand-write.1.0_1_1 write 65536 65536
/mnt/ssd5/rand-write.1.0_1_2 write 65536 65536
/mnt/ssd0/rand-write.1.0_0_0 close
/mnt/ssd1/rand-write.1.0_0_1 close
/mnt/ssd2/rand-write.1.0_0_2 close
/mnt/ssd3/rand-write.1.0_1_0 close
/mnt/ssd4/rand-write.1.0_1_1 close
/mnt/ssd5/rand-write.1.0_1_2 close

I've turned on debugging and added some extra printf in read_iolog2()
within iolog.c.

file     24594 add file /mnt/ssd0/rand-write.1.0_0_0
file     24594 resize file array to 2 files
file     24594 file 0x7fbe03f53090 "/mnt/ssd0/rand-write.1.0_0_0" added at 0
file     24594 add file /mnt/ssd1/rand-write.1.0_0_1
file     24594 file 0x7fbe03f53230 "/mnt/ssd1/rand-write.1.0_0_1" added at 1
file     24594 add file /mnt/ssd2/rand-write.1.0_0_2
file     24594 resize file array to 4 files
file     24594 file 0x7fbe03f533d0 "/mnt/ssd2/rand-write.1.0_0_2" added at 2
file     24594 add file /mnt/ssd3/rand-write.1.0_1_0
file     24594 file 0x7fbe03f53570 "/mnt/ssd3/rand-write.1.0_1_0" added at 3
file     24594 add file /mnt/ssd4/rand-write.1.0_1_1
file     24594 resize file array to 6 files
file     24594 file 0x7fbe03f53710 "/mnt/ssd4/rand-write.1.0_1_1" added at 4
file     24594 add file /mnt/ssd5/rand-write.1.0_1_2
file     24594 file 0x7fbe03f538b0 "/mnt/ssd5/rand-write.1.0_1_2" added at 5
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=0 file_action=1
fname=/mnt/ssd0/rand-write.1.0_0_0 act=open
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=1 file_action=1
fname=/mnt/ssd1/rand-write.1.0_0_1 act=open
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=2 file_action=1
fname=/mnt/ssd2/rand-write.1.0_0_2 act=open
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=3 file_action=1
fname=/mnt/ssd3/rand-write.1.0_1_0 act=open
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=4 file_action=1
fname=/mnt/ssd4/rand-write.1.0_1_1 act=open
io       24594 read_iolog2: Making note of file.
offset=7310584035475809390 bytes=811544671 fileno=5 file_action=1
fname=/mnt/ssd5/rand-write.1.0_1_2 act=open
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd0/rand-write.1.0_0_0 act=write
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd1/rand-write.1.0_0_1 act=write
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd2/rand-write.1.0_0_2 act=write
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd3/rand-write.1.0_1_0 act=write
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd4/rand-write.1.0_1_1 act=write
io       24594 read_iolog2: Making note of file. offset=0 bytes=65536
fileno=5 file_action=1 fname=/mnt/ssd5/rand-write.1.0_1_2 act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd0/rand-write.1.0_0_0
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd1/rand-write.1.0_0_1
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd2/rand-write.1.0_0_2
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd3/rand-write.1.0_1_0
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd4/rand-write.1.0_1_1
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=1 fname=/mnt/ssd5/rand-write.1.0_1_2
act=write
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=0 file_action=2 fname=/mnt/ssd0/rand-write.1.0_0_0
act=close
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=1 file_action=2 fname=/mnt/ssd1/rand-write.1.0_0_1
act=close
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=2 file_action=2 fname=/mnt/ssd2/rand-write.1.0_0_2
act=close
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=3 file_action=2 fname=/mnt/ssd3/rand-write.1.0_1_0
act=close
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=4 file_action=2 fname=/mnt/ssd4/rand-write.1.0_1_1
act=close
io       24594 read_iolog2: Making note of file. offset=65536
bytes=65536 fileno=5 file_action=2 fname=/mnt/ssd5/rand-write.1.0_1_2
act=close

Taking a look to the debugging output shows that on write action the
'fileno' variable is always set to 5 regardless of the file that should
be written. However, for open and close we have a different fileno value
for each file. Taking a look into the source code of read_iolog2()
reveals why: In case of actions 'open' or 'close', 'fileno' is looked up
through 'get_fileno(td, fname)', but for 'write', 'read' and the other 4
argument actions 'fileno' is not looked up. When I add a 'fileno =
get_fileno(td, fname);' at the end of the '(r == 4)' branch, I/Os are
replayed to the designated files.

                 if (r == 4) {
                         /*
                          * Check action first
                          */
                         if (!strcmp(act, "wait"))
                                 rw = DDIR_WAIT;
                         else if (!strcmp(act, "read"))
                                 rw = DDIR_READ;
                         else if (!strcmp(act, "write"))
                                 rw = DDIR_WRITE;
                         else if (!strcmp(act, "sync"))
                                 rw = DDIR_SYNC;
                         else if (!strcmp(act, "datasync"))
                                 rw = DDIR_DATASYNC;
                         else if (!strcmp(act, "trim"))
                                 rw = DDIR_TRIM;
                         else {
                                 log_err("fio: bad iolog file action: %s\n",
act);
                                 continue;
                         }
                 } else if (r == 2) {
                         rw = DDIR_INVAL;
                         if (!strcmp(act, "add")) {
                                 td->o.nr_files++;
                                 fileno = add_file(td, fname);
                                 file_action = FIO_LOG_ADD_FILE;
                                 continue;
                         } else if (!strcmp(act, "open")) {
                                 fileno = get_fileno(td, fname);
                                 file_action = FIO_LOG_OPEN_FILE;
                         } else if (!strcmp(act, "close")) {
                                 fileno = get_fileno(td, fname);
                                 file_action = FIO_LOG_CLOSE_FILE;
                         } else {
                                 log_err("fio: bad iolog file action: %s\n",
act);
                                 continue;
                         }
                 } else {
                         log_err("bad iolog2: %s", p);
                         continue;
                 }

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: missing Android config options
Aaron Carroll [Mon, 18 Feb 2013 07:52:18 +0000 (08:52 +0100)]
configure: missing Android config options

We need this for the run-time endianness check to pass.  Add supported
clock monotonic while we're at it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: attempt to get Android going again
Aaron Carroll [Tue, 12 Feb 2013 08:58:14 +0000 (09:58 +0100)]
configure: attempt to get Android going again

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoInitialize global and non-static variables
Jens Axboe [Mon, 11 Feb 2013 15:33:25 +0000 (16:33 +0100)]
Initialize global and non-static variables

Don't assume they get zeroed. The AIX linker throws a hissy fit on
these, refusing to link the program.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoTurn f->engine_data into a 64-bit value
Jens Axboe [Mon, 11 Feb 2013 13:35:43 +0000 (14:35 +0100)]
Turn f->engine_data into a 64-bit value

Even on 32-bit platforms, we store (uint64_t) -1 in there
for some cases to make it be 'invalid'.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: ensure that fatal errors kill config-host.h and mak files
Jens Axboe [Mon, 11 Feb 2013 12:21:54 +0000 (13:21 +0100)]
configure: ensure that fatal errors kill config-host.h and mak files

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoHandle end_fsync if a file is closed
Jens Axboe [Mon, 11 Feb 2013 10:26:55 +0000 (11:26 +0100)]
Handle end_fsync if a file is closed

Should not happen, but better be careful since it could have
an impact on data integrity.

Additionally, add a log_err() error if we do fail it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: allow --cpu to be specified
Jens Axboe [Sun, 10 Feb 2013 14:48:34 +0000 (15:48 +0100)]
configure: allow --cpu to be specified

And don't fail if we don't detect, it's not critical
for configure.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix rate limiting
Jens Axboe [Fri, 8 Feb 2013 18:05:25 +0000 (19:05 +0100)]
Fix rate limiting

Commit e0224c6b accidentally broke rate limiting, since
it bypassed the sleep logic. Rework it slightly to not
skip that.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoWindows: io cancellation often fails and causes crashes, so remove it.
Bruce Cran [Fri, 8 Feb 2013 01:45:17 +0000 (01:45 +0000)]
Windows: io cancellation often fails and causes crashes, so remove it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix bug with rate and read/write mixed workloads at 100% bias
Jens Axboe [Thu, 7 Feb 2013 18:55:24 +0000 (19:55 +0100)]
Fix bug with rate and read/write mixed workloads at 100% bias

The rate_ddir() might sometimes decide to send the other data
direction, even if we have a workload with rw=randrw and
rwmixread=0. Check for this condition.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agot/log: include minmax.h for min()
Jens Axboe [Thu, 7 Feb 2013 14:40:48 +0000 (15:40 +0100)]
t/log: include minmax.h for min()

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMakfile: use LINK for the link phase
Jens Axboe [Thu, 7 Feb 2013 14:40:37 +0000 (15:40 +0100)]
Makfile: use LINK for the link phase

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix zones for numjobs=x, where x > 1
Jens Axboe [Thu, 7 Feb 2013 12:06:18 +0000 (13:06 +0100)]
Fix zones for numjobs=x, where x > 1

In the options fixup, we auto-set open_files to 1 if it hasn't been set.
So any duplicated jobs will clear the zone. Fixup the check to properly
check for > 1 open files.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agostat: fix wrong type used for 32-bit compiles
Jens Axboe [Thu, 7 Feb 2013 11:56:09 +0000 (12:56 +0100)]
stat: fix wrong type used for 32-bit compiles

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix crash and precision of ETA with zones
Jens Axboe [Thu, 7 Feb 2013 11:54:10 +0000 (12:54 +0100)]
Fix crash and precision of ETA with zones

If zonesize was bigger than zoneskip, we could encounter a divide by
zero when calculating the number of bytes.

Additionally, the math was just wrong for most cases of zone settings.
Improve that.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoWindows: fix mlock, remove ftruncate and fix error handling.
Bruce Cran [Wed, 6 Feb 2013 23:55:46 +0000 (23:55 +0000)]
Windows: fix mlock, remove ftruncate and fix error handling.

Functions in posix.c need to use win_to_posix_err to convert errors from
Windows API calls to their equivalent POSIX error code so perror etc.
work: move it into posix.c and add the prototype to posix.h.

Fix error handling so -1 is always returned on error and errno is set.

Fix mlock to work with sizes greater than about 1MB: the working set size
needs to be increased for the new allocation.

Remove ftruncate from posix.c, since MinGW already provides it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoWindows: fix sysconf(_SC_PHYS_PAGES).
Bruce Cran [Wed, 6 Feb 2013 20:27:27 +0000 (20:27 +0000)]
Windows: fix sysconf(_SC_PHYS_PAGES).

sysconf(_SC_PHYS_PAGES) was returning the number of bytes of memory
instead of the number of pages. Also improve some error messages in
sysconf(). Update os_phys_mem() to use sysconf().

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoGet rid of fallocate on Windows
Jens Axboe [Wed, 6 Feb 2013 12:52:12 +0000 (13:52 +0100)]
Get rid of fallocate on Windows

Fallocate is only useful if it is a fast operation, helping the
file system allocate and setup meta data for the given size.
So don't punt to zero filling the entire thing.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoHandle normal output wrapping of the percentile list
Jens Axboe [Wed, 6 Feb 2013 07:51:57 +0000 (08:51 +0100)]
Handle normal output wrapping of the percentile list

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoCleanup the percentile output formatting
Jens Axboe [Wed, 6 Feb 2013 07:46:55 +0000 (08:46 +0100)]
Cleanup the percentile output formatting

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoHandle percentile lists with higher precision that 2 digits
Vincent Kang Fu [Wed, 6 Feb 2013 07:43:40 +0000 (08:43 +0100)]
Handle percentile lists with higher precision that 2 digits

We cap the output at %2.2f right now, that's not always enough.
Make the parser check and store the precision required to
output the list correctly.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd --build-32bit-win switch to configure --help output.
Bruce Cran [Tue, 5 Feb 2013 17:44:19 +0000 (17:44 +0000)]
Add --build-32bit-win switch to configure --help output.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agowindowsaio: fix file header format and improve error reporting.
Bruce Cran [Tue, 5 Feb 2013 11:33:04 +0000 (11:33 +0000)]
windowsaio: fix file header format and improve error reporting.

Use the same file header format as other engines.
Remove the copyright line.
Add calls to log_err to improve error reporting.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMakefile: fixup init.c dependency
Jens Axboe [Tue, 5 Feb 2013 10:03:12 +0000 (11:03 +0100)]
Makefile: fixup init.c dependency

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoEnable forced 32-bit build on Windows
Huadong Liu [Tue, 5 Feb 2013 07:43:14 +0000 (08:43 +0100)]
Enable forced 32-bit build on Windows

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix copy/paste error in windowsaio register/unregister function names.
Bruce Cran [Tue, 5 Feb 2013 00:56:59 +0000 (00:56 +0000)]
Fix copy/paste error in windowsaio register/unregister function names.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoUpdate the Windows installer
Bruce Cran [Mon, 4 Feb 2013 12:50:51 +0000 (13:50 +0100)]
Update the Windows installer

Update the EULA copyright year to 2013, auto-generate the product ID,
remove the comment field and allow per-user installs.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agogettime: fixup AMD constant TSC detection
Jens Axboe [Sun, 3 Feb 2013 19:54:26 +0000 (20:54 +0100)]
gettime: fixup AMD constant TSC detection

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix failure to exit IO loop on some IO sizes
Jens Axboe [Sun, 3 Feb 2013 13:20:44 +0000 (14:20 +0100)]
Fix failure to exit IO loop on some IO sizes

If the size of a file isn't a multiple of the block size being
used, it can cause fio to exit the IO loop, check bytes done,
and then decide to do one more loop since we didn't do quite
as much IO as we wanted to. This happens because the minimum
block size is larger than the remainder. So check for that,
and stop if we need to.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoUpdate the Windows section of the README file.
Bruce Cran [Sat, 2 Feb 2013 15:12:41 +0000 (15:12 +0000)]
Update the Windows section of the README file.

The 64-bit Windows build instructions can be simplified since Cygwin
includes all the required MinGW64 packages. Also, the pthread DLL is
available as an individual download.

Update the WiX Toolset version to 3.7 and fix the URL since it's no
longer hosted at Sourceforge.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoRemove duplicated Windows configure options.
Bruce Cran [Sat, 2 Feb 2013 20:59:31 +0000 (20:59 +0000)]
Remove duplicated Windows configure options.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoDefault to CS_GTOD if CONFIG_CLOCK_GETTIME isn't defined.
Bruce Cran [Sat, 2 Feb 2013 20:54:36 +0000 (20:54 +0000)]
Default to CS_GTOD if CONFIG_CLOCK_GETTIME isn't defined.

Defaulting the clock source to CS_CGETTIME broke on OS X because it
doesn't have clock_gettime. Use CS_GTOD if CONFIG_CLOCK_GETTIME isn't
defined.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agovalues.h is obsolete: use float.h and DBL_MIN/MAX instead.
Bruce Cran [Sat, 2 Feb 2013 14:31:24 +0000 (14:31 +0000)]
values.h is obsolete: use float.h and DBL_MIN/MAX instead.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoclock: hardwire tsc as unreliable on Solaris for now
Jens Axboe [Sat, 2 Feb 2013 09:05:07 +0000 (10:05 +0100)]
clock: hardwire tsc as unreliable on Solaris for now

Need to double check the cpuid test, it probably only is reliable
on Intel. Need to check the CPU vendor and flags appropriately.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoparse: use MIN/MAXDOUBLE instead of some representation of NAN
Jens Axboe [Sat, 2 Feb 2013 08:56:23 +0000 (09:56 +0100)]
parse: use MIN/MAXDOUBLE instead of some representation of NAN

Doesn't always work for C99.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix percentile_list option
Vincent Kang Fu [Sat, 2 Feb 2013 08:28:55 +0000 (09:28 +0100)]
Fix percentile_list option

Commit 802ad4a8 broke the parsing of the percentile list,
by neglecting to change the parser type fo a fio_fp64_t.

Also get rid of the def_percentile_list, just set the
default percentile list as the option default.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoMakefile: fixup dependency problem for files in sub dirs
Jens Axboe [Fri, 1 Feb 2013 14:51:23 +0000 (15:51 +0100)]
Makefile: fixup dependency problem for files in sub dirs

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoREADME: how to compile fio on Windows 64 bits
Neto, Antonio Jose Rodrigues [Fri, 1 Feb 2013 07:30:38 +0000 (08:30 +0100)]
README: how to compile fio on Windows 64 bits

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAllow override of CFLAGS
David M. Lee [Thu, 31 Jan 2013 17:37:41 +0000 (18:37 +0100)]
Allow override of CFLAGS

If you attempt to specify your own CFLAGS on the command line (such as
make CFLAGS="--whatever"), this would prevent -DFIO_VERSION from being
added to CFLAGS. This patch uses the override directive to allow the
flag to be appended to CFLAGS.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoOnly disable stdout ETA output if results are going to stdout
Jens Axboe [Thu, 31 Jan 2013 14:24:34 +0000 (15:24 +0100)]
Only disable stdout ETA output if results are going to stdout

If they are being redirected to a file with --output, then
we can still use the normal runtime ETA output on stdout.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agonet: 'nodelay' HOWTO/man page update
Jens Axboe [Thu, 31 Jan 2013 12:33:45 +0000 (13:33 +0100)]
net: 'nodelay' HOWTO/man page update

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: add TCP_NODELAY check
Jens Axboe [Thu, 31 Jan 2013 09:19:51 +0000 (10:19 +0100)]
configure: add TCP_NODELAY check

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agowindows: fix typo in <netinet/tcp.h> header
Jens Axboe [Thu, 31 Jan 2013 08:58:22 +0000 (09:58 +0100)]
windows: fix typo in <netinet/tcp.h> header

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix bugs in [v]snprintf usage
Ken Raeburn [Wed, 30 Jan 2013 21:31:09 +0000 (22:31 +0100)]
Fix bugs in [v]snprintf usage

When calling snprintf, supply the full buffer size instead of
one byte less.

When using the returned length from vsnprintf for logging, don't write
more than the actual buffer size (minus one for the trailing \0), in
case the formatted string was truncated.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd empty <netinet/tcp.h> for Windows
Jens Axboe [Wed, 30 Jan 2013 20:54:58 +0000 (21:54 +0100)]
Add empty <netinet/tcp.h> for Windows

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agonet: fix compile warning on Windows (and others)
Jens Axboe [Wed, 30 Jan 2013 20:52:37 +0000 (21:52 +0100)]
net: fix compile warning on Windows (and others)

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agonet engine: implement option "nodelay" for TCP sockets
Steven Noonan [Wed, 28 Nov 2012 22:52:36 +0000 (14:52 -0800)]
net engine: implement option "nodelay" for TCP sockets

With disks, O_DIRECT effectively bypasses all buffering/caching mechanisms and
ensures that the I/O is going directly to the disk. Since TCP is a streaming
protocol (like disk I/O), it also has a buffering mechanism. As with disks, it
is sometimes desirable to bypass buffering. To that end, we can use
TCP_NODELAY, which transmits the packet as soon as data is assembled,
regardless of whether it occupies a full frame.

Signed-off-by: Steven Noonan <snoonan@amazon.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoidletime: style cleanups
Jens Axboe [Wed, 30 Jan 2013 13:08:00 +0000 (14:08 +0100)]
idletime: style cleanups

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: add SCHED_IDLE check
Jens Axboe [Wed, 30 Jan 2013 12:58:58 +0000 (13:58 +0100)]
configure: add SCHED_IDLE check

We can't consider it always available on Linux. It fails on
Debian 6.x, for instance.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd support for idletime profiling
Huadong Liu [Wed, 30 Jan 2013 12:22:24 +0000 (13:22 +0100)]
Add support for idletime profiling

Idletime profiling allows a benchmark to run while filling the
idle cycles on the server, hence giving you some metric of how
much pressure the system is under. This is useful to be able
to profile and compare changes in a setup or driver, for instance.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoAdd unified_rw_reporting option
Jens Axboe [Wed, 30 Jan 2013 11:56:23 +0000 (12:56 +0100)]
Add unified_rw_reporting option

If this is set, then fio will sum and display just a single set
of statistics for any IO type. By default, fio accounts and reports
each data direction separately.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoHOWTO: indentation fix
Jens Axboe [Wed, 30 Jan 2013 11:53:55 +0000 (12:53 +0100)]
HOWTO: indentation fix

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agosmalloc: alloc failure cleanups
Jens Axboe [Tue, 29 Jan 2013 21:29:09 +0000 (22:29 +0100)]
smalloc: alloc failure cleanups

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoclient/server: fixup missing TRIM conversions
Jens Axboe [Tue, 29 Jan 2013 19:53:52 +0000 (20:53 +0100)]
client/server: fixup missing TRIM conversions

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoFix crash with absurdly but not impossibly deeply nested device stacks
Ken Raeburn [Tue, 29 Jan 2013 09:18:13 +0000 (10:18 +0100)]
Fix crash with absurdly but not impossibly deeply nested device stacks

Use the proper PATH_MAX for the max path for disk util, and use
snprintf() to avoid potentially overwriting it.

Modified by Jens to return NULL instead of exiting.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: enable --extra-cflags for Windows
Jens Axboe [Tue, 29 Jan 2013 09:06:30 +0000 (10:06 +0100)]
configure: enable --extra-cflags for Windows

Add configure help as well.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
11 years agoconfigure: add --extra-cflags
Jens Axboe [Tue, 29 Jan 2013 08:26:07 +0000 (09:26 +0100)]
configure: add --extra-cflags

And ensure that -D_GNU_SOURCE is being set for configure test
programs too.

Signed-off-by: Jens Axboe <axboe@kernel.dk>