changes.txt

硬盘速度测试 linux下 源码 c/c

V2.31	(capps)
	Fixups for build under Linux.
	Added -j ### to support user setting the stride size for the 
	stride read benchmark.

V2.32 	(capps)
	Add support for IRIX and IRIX64.

V2.33 	(capps)
	Add support for POSIX async I/O benchmarking. Uses a library
	to interface to POSIX async I/O model. The library provides
	and extended async_read() interface. It takes the standard
	calling options of read() but also allows the application to
	perform read-ahead with a stride. (positive or negative)
	and allows the user to specify how much read ahead to 
	perform. 
	Tested on HP-UX 11.0, Linux, SGI Origin.

V2.34 	(capps)
	Added -k. This allows POSIX async I/O to utilize the buffer
	specified and not to perform any bcopys. Fixes to make
	multi-threadedness work on SGI Origin.

V2.34 	(capps)
	Added [-k #]. This allows POSIX async I/O to utilize the buffer
	specified and not to perform any bcopys. Fixes to make
	multi-threadedness work on SGI Origin.

V2.36 	(capps)
	Iozone is now a 64 bit application. It may be compiled for either
	64 bit or 32 bit machines. The makefile supports 64 and 32 bit
	targets for machines that support 32 & 64 bit targets.
	All version numbers are now automatically generated by
	RCS. This is the last time we have to bump the version
	number by hand.


-----------------------------------------------------------------------------------
Changed over to RCS source control here:
Version Numbers are reset at this point back to Version 1.1.
-----------------------------------------------------------------------------------


RCS file:        iozone.c,v;   Working file:    iozone.c
head:            1.94
locks:           ;  strict
access list:   
symbolic names:
comment leader:  " * "
total revisions: 94;    selected revisions: 94
description:
Initial rcs version of Iozone
----------------------------
Revision 1.94        
date: 99/01/18 13:02:57;  author: capps;  state: Exp;  lines added/del: 7/2
Call msync if writer wants sync in timing and terminates early in multi thread test case.
----------------------------
Revision 1.93        
date: 99/01/18 11:46:11;  author: capps;  state: Exp;  lines added/del: 309/126
Cleanup for include_flush and include_close for single and multi threaded operations.
----------------------------
Revision 1.92        
date: 99/01/15 10:53:58;  author: capps;  state: Exp;  lines added/del: 40/11
Add include_close support for throughput testing
----------------------------
Revision 1.91        
date: 98/12/07 09:26:22;  author: capps;  state: Exp;  lines added/del: 43/24
For Windows: Use the high resolution timers instead of timeofday();
Fix a few casting problems.
----------------------------
Revision 1.90        
date: 98/11/30 14:49:46;  author: capps;  state: Exp;  lines added/del: 24/17
Update the copyright and names and places
----------------------------
Revision 1.89        
date: 98/10/30 09:04:51;  author: capps;  state: Exp;  lines added/del: 1/2
An extra close(fd) causes HP-UX to fail future unmounts... 
----------------------------
Revision 1.88        
date: 98/10/29 09:47:25;  author: capps;  state: Exp;  lines added/del: 17/17
Cleanup the help screen
----------------------------
Revision 1.87        
date: 98/10/28 23:31:11;  author: capps;  state: Exp;  lines added/del: 7/6
Spelling error fix.
----------------------------
Revision 1.86        
date: 98/10/14 11:21:50;  author: capps;  state: Exp;  lines added/del: 23/68
Unified the time method to only have 2 ways to get time.
----------------------------
Revision 1.85        
date: 98/10/14 09:22:09;  author: capps;  state: Exp;  lines added/del: 91/91
Added code to remove the latency of gettimeofday() from the file performance measurements.
----------------------------
Revision 1.84        
date: 98/10/12 11:44:50;  author: capps;  state: Exp;  lines added/del: 107/8
Add time resolution output, and fix the divide by zero when the time in
a system call turns out to be Zero. This will introduce distortion for machines
that have very fast system calls and very poor time resolution. Windows 
has a 50 Milli second resolution on gettimeofday(). So... to fix it 
all calls that take less than 50 Milli seconds will be rounded up to 
cost 50 milliseconds.
----------------------------
Revision 1.83        
date: 98/10/06 09:58:16;  author: capps;  state: Exp;  lines added/del: 46/2
Add support for Windows build
----------------------------
Revision 1.82        
date: 98/09/23 09:48:02;  author: capps;  state: Exp;  lines added/del: 2/2
Fix bug where -i # was leaving tmp files after throughput test.
----------------------------
Revision 1.81        
date: 98/09/23 09:41:12;  author: capps;  state: Exp;  lines added/del: 1/3
Remove debug printf
----------------------------
Revision 1.80        
date: 98/09/23 09:29:01;  author: capps;  state: Exp;  lines added/del: 23/1
Add my_nap(). This allows the threads to switch processors to their
new bound processor before performing any work.
----------------------------
Revision 1.79        
date: 98/09/22 11:57:20;  author: capps;  state: Exp;  lines added/del: 8/8
Change xx back into an int so the modulo will work better.
----------------------------
Revision 1.78        
date: 98/09/18 16:27:05;  author: capps;  state: Exp;  lines added/del: 18/15
Remove create in rewrite path.
----------------------------
Revision 1.77        
date: 98/08/17 16:44:06;  author: capps;  state: Exp;  lines added/del: 23/1
Fixes for Solaris and the new processor bind feature.
----------------------------
Revision 1.76        
date: 98/08/17 16:17:45;  author: capps;  state: Exp;  lines added/del: 1/2
Remove debug code.
----------------------------
Revision 1.75        
date: 98/08/17 16:16:15;  author: capps;  state: Exp;  lines added/del: 92/5
Add support for binding procs/threads to cpus.
----------------------------
Revision 1.74        
date: 98/08/07 16:51:41;  author: capps;  state: Exp;  lines added/del: 4/3
Add fsync to the fwrite test case when the user specifies -e
----------------------------
Revision 1.73        
date: 98/08/07 16:47:38;  author: capps;  state: Exp;  lines added/del: 178/208
Add -c and -e to allow closes and fsyncs to be inside the timing calculations.
----------------------------
Revision 1.72        
date: 98/08/06 22:40:15;  author: capps;  state: Exp;  lines added/del: 9/1
Add setvbuf to fwrite and fread tests so that the internal fwrite and fread
buffer size is the same as the record size. This is what a well tuned application 
would do.
----------------------------
Revision 1.71        
date: 98/08/06 09:03:06;  author: capps;  state: Exp;  lines added/del: 2/3
Fix fsync filename problem in fwrite_perf_test
----------------------------
Revision 1.70        
date: 98/08/05 18:06:41;  author: capps;  state: Exp;  lines added/del: 6/2
Add fsync after fwrite test case so the fread will start with a 
clean buffer cache and no writes in progress.
----------------------------
Revision 1.69        
date: 98/08/03 10:45:49;  author: capps;  state: Exp;  lines added/del: 3/3
Bug fix for -V option not filling the entire buffer.
----------------------------
Revision 1.68        
date: 98/07/30 22:11:11;  author: capps;  state: Exp;  lines added/del: 2/3
Fix for solaris
----------------------------
Revision 1.67        
date: 98/07/30 22:08:19;  author: capps;  state: Exp;  lines added/del: 2/2
Fix for solaris
----------------------------
Revision 1.66        
date: 98/07/30 22:05:02;  author: capps;  state: Exp;  lines added/del: 43/15
Add support for Solaris
----------------------------
Revision 1.65        
date: 98/07/01 14:19:19;  author: capps;  state: Exp;  lines added/del: 80/82
Move end_async inside the timing loops as in async I/O it counts.
----------------------------
Revision 1.64        
date: 98/06/16 17:04:36;  author: capps;  state: Exp;  lines added/del: 13/2
Correct problem where user specifies pread tests on hpux... which does not 
support these operations. The test now prints an error message and exits.
----------------------------
Revision 1.63        
date: 98/06/16 16:54:22;  author: capps;  state: Exp;  lines added/del: 1/2
Remove exit from auto_test. This allows the message "iozone test complete" to 
be printed when in auto test mode.
----------------------------
Revision 1.62        
date: 98/06/10 10:54:28;  author: capps;  state: Exp;  lines added/del: 175/173
All exit()s now have a unique exit value.
----------------------------
Revision 1.61        
date: 98/05/18 13:34:03;  author: capps;  state: Exp;  lines added/del: 17/18
Move .dat file descriptors to global data. Needed to prevent re-opens.
----------------------------
Revision 1.60        
date: 98/05/18 13:24:22;  author: capps;  state: Exp;  lines added/del: 6/3
Bug fix. Prevents re-opending .dat files when in auto mode.
----------------------------
Revision 1.59        
date: 98/05/08 13:03:02;  author: capps;  state: Exp;  lines added/del: 21/3
Enhance throughput tests to follow the -i test number to run options.
----------------------------
Revision 1.58        
date: 98/05/07 14:15:49;  author: capps;  state: Exp;  lines added/del: 109/39
Make VXFS a define in the make command. This makes moving to other targets
easier. It removes the binding of HPUX and VXFS.
Also, Added -Q to support offset/latency file generation for later use
as inputs to plot program.
----------------------------
Revision 1.57        
date: 98/05/06 15:09:43;  author: capps;  state: Exp;  lines added/del: 100/27
Add -N to provide results in microseconds per operation.
----------------------------
Revision 1.56        
date: 98/05/05 13:23:29;  author: capps;  state: Exp;  lines added/del: 3/10
If the user specifies -i 0 then run both write and rewrite tests.
----------------------------
Revision 1.55        
date: 98/04/30 15:19:02;  author: capps;  state: Exp;  lines added/del: 1/1
No change
----------------------------
Revision 1.54        
date: 98/04/30 15:09:58;  author: capps;  state: Exp;  lines added/del: 2/2
Unlink the vxfstest when the test fails.
----------------------------
Revision 1.53        
date: 98/04/30 13:07:21;  author: capps;  state: Exp;  lines added/del: 7/5
Cleanup help output.
----------------------------
Revision 1.52        
date: 98/04/30 12:58:29;  author: capps;  state: Exp;  lines added/del: 21/4
Add async I/O with no bcopy to throughput tests.
----------------------------
Revision 1.51        
date: 98/04/29 15:29:29;  author: capps;  state: Exp;  lines added/del: 5/1
Fixes so it will compile on the SGI Origin.
----------------------------
Revision 1.50        
date: 98/04/29 11:57:58;  author: capps;  state: Exp;  lines added/del: 5/1
Do not need to limit async ops. Fix is in libasync.c
----------------------------
Revision 1.49        
date: 98/04/29 10:45:19;  author: capps;  state: Exp;  lines added/del: 61/3
Add async I/O to throughput testing for writes
----------------------------
Revision 1.48        
date: 98/04/28 11:57:13;  author: capps;  state: Exp;  lines added/del: 5/1
Limit max async operations to 60. Beyond this there be dragons.
----------------------------
Revision 1.47        
date: 98/04/28 10:16:09;  author: capps;  state: Exp;  lines added/del: 108/21
Completed support for no_bcopy POSIX async I/O in the async_write_no_copy path.
This allows write tests to perform async I/O with buffers released when 
the write is completed.
----------------------------
Revision 1.46        
date: 98/04/27 16:58:38;  author: capps;  state: Exp;  lines added/del: 43/10
Add aio_write() to the write and re-write tests. This provides
POSIX async I/O for the those tests.
----------------------------
Revision 1.45        
date: 98/04/25 09:53:39;  author: capps;  state: Exp;  lines added/del: 3/2
direct_flag is an int. Was a char in one place and
an int in another.
----------------------------
Revision 1.44        
date: 98/04/25 09:17:42;  author: capps;  state: Exp;  lines added/del: 27/15
More support for vx_direct support in the write path
----------------------------
Revision 1.43        
date: 98/04/24 16:33:44;  author: capps;  state: Exp;  lines added/del: 115/77
Move VX_DIRECT to libasync. But keep the VX_DIRECT support also
in iozone. So one can use VX_DIRECT with and without async I/O
----------------------------
Revision 1.42        
date: 98/04/24 16:20:34;  author: capps;  state: Exp;  lines added/del: 127/60
Move VX_DIRECT to the libasync module.
----------------------------
Revision 1.41        
date: 98/04/24 15:50:54;  author: capps;  state: Exp;  lines added/del: 190/7
Add support for VxFS VX_DIRECT 
Idea is to use VX_DIRECT and POSIX async I/O together
----------------------------
Revision 1.40        
date: 98/04/22 16:38:25;  author: capps;  state: Exp;  lines added/del: 5/5