dbench scalability testing

[Anton Blanchard]

> Actually, I'll round up my patch to use MSG_TRUNC in smbtorture. This
> improves client performance somewhat as we dont waste cpu cycles copying
> data between kernel and user space. (When things dont run fast enough, fix
> the benchmark :)

Here it is.

Anton

Index: source/utils/nbio.c
===================================================================
RCS file: /data/cvs/samba/source/utils/nbio.c,v
retrieving revision 1.4
diff -u -u -r1.4 nbio.c
--- source/utils/nbio.c	2000/06/13 09:33:43	1.4
+++ source/utils/nbio.c	2001/03/10 02:22:45
@@ -145,12 +145,16 @@
 		       line_count, handle, size, offset);
 		return;
 	}
+#if 0
 	if ((ret=cli_read(c, ftable[i].fd, buf, offset, size)) != size) {
 #if NBDEBUG
 		printf("(%d) read failed on handle %d ofs=%d size=%d res=%d\n", 
 		       line_count, handle, offset, size, ret);
 #endif
 	}
+#else
+	recv(ftable[i].fd, buf, size, MSG_TRUNC);
+#endif
 }
 
 void nb_close(int handle)