Jens Axboe [Tue, 18 Dec 2012 08:41:59 +0000 (09:41 +0100)]
gettime: make last_cycles thread local too
Avoids a lot of cache line bouncing for threads.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 18 Dec 2012 08:12:27 +0000 (09:12 +0100)]
gettime: use pthread_{set,get}specific() for TLS
__thread doesn't work on OSX at least (thanks...), so resort to
using the pthread functions to maintaining a per-thread clock
check.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 18 Dec 2012 07:23:44 +0000 (08:23 +0100)]
gettime: fix race/bug with threads and time keeping
Sam writes:
First, the 1 second latency events come in batches and those batches
occur suspiciously close to a usec wrap (0.999999 us -> 1.000000 us).
Second, if you subtract exactly 1 second from these outlier latencies,
the remaining amount is very close to what our instrumented low level
driver records for the IO latency and consistent with the expected
latencies of our SSD. Similarly, the tv_usec portion of the timeval
structs shows increasing values. See snippet below.
Format is like start: <start_time.tv_sec>.<start_time.tv_usec>
latency:
1004657 us, lba:
1111289192, start:
1355776806.995294 issue:
1355776806.995312 complete:
1355776807.999969
latency:
1000494 us, lba:
203093568, start:
1355776806.999456 issue:
1355776806.999475 complete:
1355776807.999969
latency:
1000404 us, lba:
551350992, start:
1355776806.999546 issue:
1355776806.999565 complete:
1355776807.999969
latency:
1000477 us, lba:
449672928, start:
1355776806.999484 issue:
1355776806.999492 complete:
1355776807.999969
All this pointed to the time collection code being buggy. Reviewing the
code, I spotted this in fio_gettime():
/*
* If Linux is using the tsc clock on non-synced processors,
* sometimes time can appear to drift backwards. Fix that up.
*/
if (last_tv_valid) {
if (tp->tv_sec < last_tv.tv_sec)
tp->tv_sec = last_tv.tv_sec;
else if (last_tv.tv_sec == tp->tv_sec &&
tp->tv_usec < last_tv.tv_usec)
tp->tv_usec = last_tv.tv_usec;
}
last_tv_valid = 1;
memcpy(&last_tv, tp, sizeof(*tp));
This does not appear to be multi-thread safe. Pre-emption can occur
between either comparison and the subsequent update. Commenting it out
makes the problem go away (at the expense of being subject to drift).
How about making last_tv & last_tv_valid thread-local?
Reported-by: Sam Bradshaw <sbradshaw@micron.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 17 Dec 2012 13:44:08 +0000 (14:44 +0100)]
Add --cpuclock-test and clocksource= option help
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 17 Dec 2012 13:29:28 +0000 (14:29 +0100)]
gettime: include per-cpu clock calibration in cpu clock test
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 17 Dec 2012 13:23:47 +0000 (14:23 +0100)]
gettime: locking fix and debug check for identical sequence
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 17 Dec 2012 11:03:29 +0000 (12:03 +0100)]
cpu clock: add independent test for monotonic/sane TSC
Fire up all threads on the system, checking the clock 100,000
times on each. Then collect and compare results, to ensure we
never have a clock that goes backwards.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 14 Dec 2012 19:37:06 +0000 (20:37 +0100)]
clock: ensure that we re-init if the clocksource changes from the default
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 12 Dec 2012 08:07:00 +0000 (09:07 +0100)]
binject: fixup ->file_data breakage
Not sure how that was done, but I missed binject in doing
the ->file_data/->file_pos unification.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 12 Dec 2012 07:34:42 +0000 (08:34 +0100)]
file: unify ->file_data and ->file_pos
The only real use case of ->file_pos was by the sync engine to
avoid an lseek() if the offset was already correct. The e4defrag,
falloc, and fusion-aw engine also set ->file_pos, but that looks
like a case of copy-paste, since they don't actually use it.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 12 Dec 2012 07:16:27 +0000 (08:16 +0100)]
file hash: don't close file fd on lookup/add race
On Linux, udev often has a rule that triggers blkid to read data off
the device, if the device has been opened for writing. This triggers
when the device is closed.
Fio has an internal file hash for lookup of files. When threads start,
it can happen that one thread does a hash lookup without finding the
file, but when it has opened the file and attempts to insert it into
the hash, another thread has already completed that operation. When
that race happens, fio closes the file handle and does the lookup
again. That then triggers blkid to read pages off the device. As data
in the cache is invalidated on open of the device, we know have page
cache entries for the device again.
This is a problem for unbuffered workloads, where the existance of
page cache pages slows it down due to having to check for invalidation.
The user observed problem is that fio exhibits bi-modal performance
results, depending on whether the file hash race was hit during setup
or not.
Fix this by NOT closing the file if we hit this race, but instead wait
until the file is closed after the run.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 10 Dec 2012 09:16:45 +0000 (10:16 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Mon, 10 Dec 2012 09:15:59 +0000 (10:15 +0100)]
cpu clock: round up when dividing by samples
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 10 Dec 2012 07:36:35 +0000 (08:36 +0100)]
Define TSC arch_init() for PPC and IA64
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 10 Dec 2012 07:29:56 +0000 (08:29 +0100)]
gettime: fix CPU calibration reported mean
Bug introduced in
fa80feae. It only affected the reported mean
if --debug=time was used, cosmetic issue.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 10 Dec 2012 07:29:03 +0000 (08:29 +0100)]
gettime: calibration rounding error
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 10 Dec 2012 07:07:14 +0000 (08:07 +0100)]
Increase CPU clock calibration accuracy
Lets throw some more loops at it, it reduces the noise.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sun, 9 Dec 2012 19:29:00 +0000 (20:29 +0100)]
Add check for invariant TSC on x86 and use TSC is default clock if reliable
TSC is by far the fastest clock we can use. Check the CPUID bits for
whether it is both constant rate AND synced across cores. If it is,
we can use it as our default clock source.
Fio will default to this clock source on x86 if no other clock source
is specifically given with clocksource= in the job file.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Thu, 6 Dec 2012 19:59:42 +0000 (19:59 +0000)]
windowsaio: create a single completion port during init, associate files during open.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 6 Dec 2012 19:53:57 +0000 (20:53 +0100)]
Document the ioengine=net pingpong= option
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Thu, 6 Dec 2012 19:33:14 +0000 (20:33 +0100)]
Fix windows out-of-memory handling
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Thu, 6 Dec 2012 19:32:40 +0000 (20:32 +0100)]
Free io_u related structures before killing IO engine
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 6 Dec 2012 19:30:38 +0000 (20:30 +0100)]
windowsaio: initialize and map windowsaio IO structure to io_u
Instead of searching for a free entry everytime we enter
queue, do this when the io_u is setup. It's cleaner and
faster.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 6 Dec 2012 16:34:57 +0000 (17:34 +0100)]
parser: always match the correct option length for posval options
Right now we match rw=randr as rw=randrw, since we use the length
of the passed in option. Use a correct max of the ival and string
passed in, so that we fail if a full match isn't present.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 5 Dec 2012 20:15:01 +0000 (21:15 +0100)]
Fix man page indentation
Was broken after the io engines.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 5 Dec 2012 09:28:08 +0000 (10:28 +0100)]
lfsr: add HOWTO and man page documentation
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 4 Dec 2012 12:12:25 +0000 (13:12 +0100)]
lfsr: ensure that the cycle follows the randrepeat= setting
Use the regular block offset seed to "seed" the lfsr generator,
so that it obeys randrepeat=0/1 as well.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 4 Dec 2012 12:10:29 +0000 (13:10 +0100)]
Name the various random offsets we use
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sat, 1 Dec 2012 14:02:11 +0000 (15:02 +0100)]
Revert "Change preferred default clocksource to gettimeofday()"
This reverts commit
f7161017b75f5d3526147d42d3b583973922c444.
clock_gettime() is monotonic, so a superior clock source even
if it is slower.
Steven Noonan [Sat, 1 Dec 2012 08:16:55 +0000 (08:16 +0000)]
engines/net.c: fix compilation error due to missing signal.h include
Corrects this build failure:
engines/net.c: In function 'fio_netio_terminate':
engines/net.c:900:2: warning: implicit declaration of function 'kill' [-Wimplicit-function-declaration]
engines/net.c:900:16: error: 'SIGUSR2' undeclared (first use in this function)
engines/net.c:900:16: note: each undeclared identifier is reported only once for each function it appears in
Signed-off-by: Steven Noonan <snoonan@amazon.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 20:46:06 +0000 (21:46 +0100)]
Wire up SIGUSR2 to kill blocking threads
Get rid of io engine ops SIGTERM flag, it didn't really
work.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 18:43:31 +0000 (19:43 +0100)]
Change preferred default clocksource to gettimeofday()
It's faster than clock_gettime()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 18:37:46 +0000 (19:37 +0100)]
Cache layout improvements
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 15:22:31 +0000 (16:22 +0100)]
net: fix receiver start time
Reset epoch/start time to when we get a connection.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 08:59:20 +0000 (09:59 +0100)]
net: add basic ping/pong type workload support
If you use ioengine=net and then set pingpong=1, fio will operate
in a ping pong type fashion. When the reader receives a package
from the writer, it will return the package to the writer again.
The writer, in turn, will wait for this package before writing
a new package. This makes fio useful for roundrobin measurements
on networks.
The submission latency then measures how long it took to submit
a packet to the local system, while the completion latency is how
long it takes for the package to be received and returned by
the other end.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 08:39:02 +0000 (09:39 +0100)]
verify: treat as failure if given verify type is different from media
When fio verifies right now, it'll look at the on media stored type
and verify that type. This means that if verify=foo set and we
find older blocks that are written with verify=bar, we will
verify those against bar.
Change this so that if a specific verify type is given, that type
is ALWAYS used for verification. If none is given but asked to
verify, we retain the old behaviour of just verifying based on
the type given in the on-media block.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 30 Nov 2012 07:27:46 +0000 (08:27 +0100)]
net: sent udp open messages
The receiver goes directly into recvfrom() when it starts up,
which greatly skews the first submission latency (since it's
measure from when it enters recvfrom() and until we finally
get the first packet from the client). Add a file open
handshake to avoid this issue.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 20:56:06 +0000 (21:56 +0100)]
Merge branch 'rand-map'
Jens Axboe [Thu, 29 Nov 2012 20:48:55 +0000 (21:48 +0100)]
solaris: enable fadvise()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 20:37:11 +0000 (21:37 +0100)]
Fix broken 'norandommap'
commit
8055e41d broke the option, always falling through to
the failure case.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 20:33:03 +0000 (21:33 +0100)]
parse: fix wrong "might be used unitialized" warning on some compilers
GCC 3.4.3 on Solaris, I'm looking at you. Apparently:
if (a || b) {
if (a)
c = foo;
else if (b)
c = bar;
*c = foobar;
}
is too hard to figure out.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 20:27:20 +0000 (21:27 +0100)]
Wire up _weak posix_fadvise()
The commit adding Android support included a botched
variant of posix_fadvise().
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 19:02:50 +0000 (20:02 +0100)]
net: set runstate to SETTING_UP while waiting for a connection
Otherwise the status print thinks it's running, and displays
bad stats.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 18:57:35 +0000 (19:57 +0100)]
net: exit on WAITALL and !ret
Otherwise we can loop forever if the other end hung up.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 13:35:33 +0000 (14:35 +0100)]
net: fix segfault with receiver, tcp, and no hostname set
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 29 Nov 2012 13:24:34 +0000 (14:24 +0100)]
libfio: don't sent KILL signal to own process from signal handler
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 28 Nov 2012 20:29:14 +0000 (21:29 +0100)]
Add t/axmap tester
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 28 Nov 2012 20:24:46 +0000 (21:24 +0100)]
Rename the bitmap to axmap
It's not really a bitmap, it's a bitmap of bitmaps. Now named.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 28 Nov 2012 18:36:16 +0000 (19:36 +0100)]
bitmap: fix off-by-8 allocation error
Was multiplying by the bit-size, not the byte-size. No algorithmic
changes, we never used the 7/8ths of the memory at the end...
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Tue, 27 Nov 2012 12:16:07 +0000 (12:16 +0000)]
Fix windowsaio IO error handling and document device write access issue on Windows and FreeBSD.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 27 Nov 2012 07:27:50 +0000 (08:27 +0100)]
Fix compile for older Linux without mmap MAP_HUGETLB
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 27 Nov 2012 07:21:18 +0000 (08:21 +0100)]
lfsr: crank it 128 times before using the sequence
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 26 Nov 2012 18:59:14 +0000 (19:59 +0100)]
Ensure that lfsr_init() is always called
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Mon, 26 Nov 2012 07:43:47 +0000 (08:43 +0100)]
Add LFSR generator
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 22 Nov 2012 14:14:17 +0000 (15:14 +0100)]
Restore BUSY_OK bypassing of bitmap
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 22 Nov 2012 14:12:05 +0000 (15:12 +0100)]
Kill now unused bitmap defines from legacy code
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 22 Nov 2012 12:59:34 +0000 (13:59 +0100)]
bitmap: kill debug code
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 22 Nov 2012 12:52:56 +0000 (13:52 +0100)]
bitmap: fix bit_masks[] for 32-bit compiles
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 22 Nov 2012 12:50:29 +0000 (13:50 +0100)]
Rework file random map
Fio slows down at the end of a random IO run, when the random
map is used and it gets fuller. This causes slowdowns in
IOPS. This is largely due to the file random map being an
array of bits, and with random access to single bits of the
array at the time, locality is awful. The effect is observable
throughout a run, though, where it gradually gets slower and
slower. It just becomes more apparent at near the end of the
run, where the last 10% are fairly bad. This is even with
doing a bunch of tricks to reduce that cost.
Implement an N-level bitmap, where layer N uses a single bit
to represent 32/64-bits at layer N-1. The number of layers
depends on the number of blocks.
This has slightly higher overhead initially in theory, in
practice it performs about the same. As a bonus, the throughput
remains constant during the run, and even becomes faster
near the end.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Aaron Carroll [Wed, 21 Nov 2012 09:39:00 +0000 (10:39 +0100)]
Add support for Android
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 20 Nov 2012 12:39:59 +0000 (13:39 +0100)]
memory: wire up mmap huge page support
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sat, 17 Nov 2012 17:06:36 +0000 (10:06 -0700)]
Use unique seed for zipf/pareto init if rand_repeat is not set
By default, fio uses repeatable random sequences for anything.
This makes it easy to run the same job again. But if this is
disabled, ensure that the zipf/pareto sequences are also
randomized. Re-use the normal random io offset seed for
the random distribution too.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Sat, 17 Nov 2012 15:12:04 +0000 (08:12 -0700)]
Generate a new Windows installer product code for 2.0.11.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bruce Cran [Sat, 17 Nov 2012 15:11:22 +0000 (08:11 -0700)]
Add more examples to the Windows installer.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 22:40:41 +0000 (15:40 -0700)]
Fio 2.0.11
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 22:38:32 +0000 (15:38 -0700)]
Document random distribution
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 22:21:23 +0000 (15:21 -0700)]
Disable random map automatically if a non-uniform random distribution is given
Doesn't make sense to use a map, if the user specifically asked for
a non-uniform distribution.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 22:19:34 +0000 (15:19 -0700)]
genzipf: add help/usage screen (-h)
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 21:58:04 +0000 (14:58 -0700)]
Get rid of -fno-omit-frame-pointer
It costs a bit of performance. And it's trivial to re-enable
for debug/profiling purposes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Vincent Kang Fu [Thu, 15 Nov 2012 20:44:34 +0000 (13:44 -0700)]
t/genzfip: add -c option for csv output
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 15 Nov 2012 14:51:28 +0000 (07:51 -0700)]
server: properly reset stat_number in fio_reset_state()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 14 Nov 2012 20:09:45 +0000 (13:09 -0700)]
client/server: fixup "All clients" reporting
We didn't properly account for no group_reporting and multiple
jobs on multiple connections.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 15:31:24 +0000 (08:31 -0700)]
Get rid of uninitialized_var()
It causes complaints on other compilers (clang). One case
was actually not used at all, initialize the rest to 0.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 15:30:56 +0000 (08:30 -0700)]
Fix wrong return type on td_error_type()
It returns enum error_type_bit, not error_type.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 15:30:30 +0000 (08:30 -0700)]
e4defrag: fix always true comparison
len is unsigned so it is always >= 0.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 12:55:38 +0000 (05:55 -0700)]
json: fix off-by-one in memory alloc
Tighten it a bit too, checking for empty string.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 03:54:12 +0000 (20:54 -0700)]
genzipf: use regular array + qsort() instead of rbtree
Saves lots of allocations, and actually ends up being a bit faster.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 13 Nov 2012 03:52:10 +0000 (20:52 -0700)]
JSON: fix escape of '"' and '\' characters
Reported-by: Kepler Kramer <kkramer@fusionio.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sun, 11 Nov 2012 23:04:48 +0000 (16:04 -0700)]
genzipf: add size and percentage hit rates
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Sun, 11 Nov 2012 07:27:24 +0000 (08:27 +0100)]
zipf: use 64-bit safe hash for zipf/pareto
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Fri, 9 Nov 2012 08:05:24 +0000 (09:05 +0100)]
genzipf: more features
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 20:56:05 +0000 (21:56 +0100)]
Makefile: 2nd go at making CC assignment/check actually work
This time tested!
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 20:46:03 +0000 (21:46 +0100)]
Use more portable construct for setting/checking CC
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 20:38:35 +0000 (21:38 +0100)]
Use ETIMEDOUT instead of ETIME for exceeding max latency
ETIME isn't very portable...
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 12:43:21 +0000 (13:43 +0100)]
zipf: kill debug "generating series" messages
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 11:45:58 +0000 (12:45 +0100)]
zipf: cap range calculation at 10M
It's fast enough to get rid of the fio.zipf.* file caching,
yet precise enough that we need not care more about it.
Even on a slower machines, generating a sequence of 10M takes
1-2s tops.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Thu, 8 Nov 2012 07:36:00 +0000 (08:36 +0100)]
Zipf theta must be different than 1.0
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 18:47:47 +0000 (19:47 +0100)]
zipf: seed zipf/pareto rand with filename hash and job id
We don't want 4 jobs operating on the same file to use the
same hot spots.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 15:12:43 +0000 (16:12 +0100)]
zipf: cleanup
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 13:04:11 +0000 (14:04 +0100)]
zipf/pareto: ensure that 0 isn't always the hottest block
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 12:47:22 +0000 (13:47 +0100)]
Add safe checks for valid pareto input value
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 12:40:01 +0000 (13:40 +0100)]
Makefile: typo
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 12:39:18 +0000 (13:39 +0100)]
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio
Jens Axboe [Wed, 7 Nov 2012 12:35:41 +0000 (13:35 +0100)]
genzipf: fix off-by-one in output array calculation
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 10:43:50 +0000 (11:43 +0100)]
genzipf: improve accuracy
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 10:40:03 +0000 (11:40 +0100)]
rbtree: add rb_next()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 10:39:30 +0000 (11:39 +0100)]
zipf/pareto: mix blocks with hashing
We don't want to favor any end of the block spectrum.
Mix with a hash.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Wed, 7 Nov 2012 08:15:45 +0000 (09:15 +0100)]
Make the zipf/pareto state per file
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 6 Nov 2012 21:10:43 +0000 (22:10 +0100)]
Move code around to satisfy t/stest linkage
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe [Tue, 6 Nov 2012 20:54:30 +0000 (21:54 +0100)]
Fix t/ieee754 link
Signed-off-by: Jens Axboe <axboe@kernel.dk>