whatsnew-mar30,1996

Linux磁盘测试的源代码,测试磁盘的读写性能

March 30, 1996 by Robin Miller

	The following changes have been made to 'dt' Version 10.1:

     o	An I/O mode option "iomode={copy, test, or verify}" has been added.
	The default mode is "test", so existing functionality does not change.
	The "copy" mode allows you to specify both input and output devices,
	the same as 'dd', for copying data from one device to another.  After
	the copy operation completes the data is automatically verified, unless
	disabled via specify "disable=verify".  The "verify" mode simply does
	a data verification of the two devices (or files) specified.

	Why was this option added? (my little story)

	In the past, I've purposely omitted 'dd' functionality, since they serve
	different needs (copy versus testing).  Well, one of my disks developed
	a bad block which could not be reassigned.  Since the bad block was in
	/etc, I decided to copy the root partition to salvage what I could.  It
	was then, that I discovered the 'dd's "conv=noerror,sync" options did
	not work as advertised (yes, I took the time to QAR 'dd').  Well, to
	make a long story shorter, I added the necessary logic to 'dt' to copy
	and properly skip bad blocks, and since 'dt' already did verification,
	this was an added bonus.

	Caveats:

	-  writes get done synchronously when AIO is chosen (read are async).
	-  the seek, skip, and block size options apply to both devices.

	These new options open a new realm of possibilities over 'dd' (IMHO):

	-  performance statistics can be obtained for device copy operations.
	-  the copy w/verify gives you that "warm and cozy" feeling when done.
	-  if errors occur during a copy, better error information is available.

     o	In conjuction with the above, proper handling of bad blocks is now done.
	Hmmm, what does that mean?  Well, of course EIO doesn't tell you a whole
	lot, but if you're testing a random I/O device, and you specify an error
	limit greater than 1, 'dt' will attempt to lseek() past the erroring lba
	to continue the copy/test/verify operation (this is 'dd' bug w/noerror).

     o	When random I/O is selected, it's no longer necessary to specify a data
	limit or record count with random access devices.  'dt' automatically
	uses an algorithm using lseek()/read()'s to find the maximum capacity.

     o	When selecting random I/O and executing multiple processes, 'dt' now
	uses both time(0) and the pid to ensure each process gets a different
	seed for random number generator.

	The latest version of 'dt' is available on our production machines
in the following account:

		Location: wasted:~rmiller/dt.d/{dt,dt-ptos}
				or via
		URL: http://www.zk3.dec.com/~rmiller/dt.html

Cheers,
Robin
================================================================================
   [ For those who don't know, here's how SCSI disk bad blocks can be made. ]

imfaster% scu -f /dev/rrz8c
scu> write media lba 1897 count 1
Writing 1 block (1897 - 1897) on /dev/rrz8c (RZ25M) with pattern 0x39c39c39...
scu> show nexus
Device: RZ25M, Bus: 1, Target: 0, Lun: 0, Type: Direct Access
scu> mbad lba 1897 hard
Making HARD Error @ Logical Block 1897 on device /dev/rrz8c (RZ25M)...
scu> verify media
Verifying 1046206 blocks (0 - 1046205) on /dev/rrz8c (RZ25M), please be patient...
scu: Verify error at logical block number 1897 (0x769).
scu: Sense Key = 0x3 = MEDIUM ERROR - Nonrecoverable medium error,
     Sense Code/Qualifier = (0x11, 0) = Unrecovered read error
scu: Error number 1 occurred on Sat Mar 30 07:45:42 1996
scu> read media lba 1897
Reading 1 block (1897 - 1897) on /dev/rrz8c (RZ25M) using pattern 0x39c39c39...
scu: Read error at logical block number 1897 (0x769).
scu: Sense Key = 0x3 = MEDIUM ERROR - Nonrecoverable medium error,
     Sense Code/Qualifier = (0x11, 0) = Unrecovered read error
scu: Error number 1 occurred on Sat Mar 30 07:45:54 1996
scu> quit
imfaster%

================================================================================
   [ This example shows a copy w/verify using the bad block generated above. ]

imfaster% dt if=/dev/rrz8a of=/dev/rrz8b iomode=copy limit=5m errors=5
dt: 'read' - I/O error
dt: Relative block number where the error occcured is 1897
dt: Error number 1 occurred on Sat Mar 30 08:10:41 1996
Copy Statistics:
    Data operation performed: Copied '/dev/rrz8a' to '/dev/rrz8b'.
     Total records processed: 10240 @ 512 bytes/record (0.500 Kbytes)
     Total bytes transferred: 5242880 (5120.000 Kbytes, 5.000 Mbytes)
      Average transfer rates: 30039 bytes/sec, 29.335 Kbytes/sec
      Total passes completed: 0/1
       Total errors detected: 1/5
          Total elapsed time: 02m54.53s
           Total system time: 00m01.66s
             Total user time: 00m00.08s

dt: 'read' - I/O error
dt: Relative block number where the error occcured is 1897
dt: Error number 1 occurred on Sat Mar 30 08:13:35 1996
Verify Statistics:
    Data operation performed: Verified '/dev/rrz8a' with '/dev/rrz8b'.
     Total records processed: 10240 @ 512 bytes/record (0.500 Kbytes)
     Total bytes transferred: 5242880 (5120.000 Kbytes, 5.000 Mbytes)
      Average transfer rates: 29809 bytes/sec, 29.110 Kbytes/sec
      Total passes completed: 1/1
       Total errors detected: 1/5
          Total elapsed time: 02m55.88s
           Total system time: 00m01.66s
             Total user time: 00m00.70s

Total Statistics:
      Input device/file name: /dev/rrz8a (Device: RZ25M, type=disk)
     Total records processed: 20480 @ 512 bytes/record (0.500 Kbytes)
     Total bytes transferred: 10485760 (10240.000 Kbytes, 10.000 Mbytes)
      Average transfer rates: 29924 bytes/sec, 29.222 Kbytes/sec
      Total passes completed: 1/1
       Total errors detected: 2/5
          Total elapsed time: 05m50.41s
           Total system time: 00m03.33s
             Total user time: 00m00.78s
               Starting time: Sat Mar 30 08:09:25 1996
                 Ending time: Sat Mar 30 08:15:16 1996

imfaster%