svn commit: r211152 - user/des/phybs

[Dag-Erling Smorgrav]

Author: des
Date: Tue Aug 10 18:58:06 2010
New Revision: 211152
URL: http://svn.freebsd.org/changeset/base/211152

Log:
  Increase MINSIZE; there isn't much point in doing 512-byte
  transactions, as they will always be either perfectly aligned or
  short.
  
  Increase STEP to MAXSIZE * 4; having STEP equal to MAXSIZE means that
  the last round actually tests sequential reading / writing, which is
  not what we're after.
  
  Instead of testing all offsets from 0 to size, test 0 and power-of-two
  multiples of BSIZE, since we assume that the physical block size is a
  power-of-two multiple of BSIZE.
  
  This should both shorten the test's run time and improve the
  significance of its output.

Modified:
  user/des/phybs/phybs.c

Modified: user/des/phybs/phybs.c
==============================================================================
--- user/des/phybs/phybs.c	Tue Aug 10 18:29:39 2010	(r211151)
+++ user/des/phybs/phybs.c	Tue Aug 10 18:58:06 2010	(r211152)
@@ -36,9 +36,10 @@
 #include <string.h>
 #include <unistd.h>
 
-#define MINSIZE		512
+#define BSIZE		512
+#define MINSIZE		1024
 #define MAXSIZE		8192
-#define STEP		MAXSIZE
+#define STEP		(MAXSIZE * 4)
 #define COUNT		65536
 
 static int opt_r = 0;
@@ -137,9 +138,12 @@ main(int argc, char *argv[])
 	printf("%8s%8s%8s%8s%12s%8s%8s\n",
 	    "count", "size", "offset", "step",
 	    "msec", "tps", "kBps");
-	for (size_t size = MINSIZE; size <= MAXSIZE; size *= 2)
-		for (off_t offset = 0; offset < (off_t)size; offset += 512)
+	for (size_t size = MINSIZE; size <= MAXSIZE; size *= 2) {
+		printf("\n");
+		scan(fd, size, 0, STEP, COUNT);
+		for (off_t offset = BSIZE; offset <= (off_t)size; offset *= 2)
 			scan(fd, size, offset, STEP, COUNT);
+	}
 	close(fd);
 	exit(0);
 }