Patch for --noswap on Mac. Untested.

doc/release_new_version

@@ -139,8 +139,7 @@ http://freshmeat.net/projects/parallel/releases/new
 
 == Update Diaspora Twitter ==
 
-New version of #GNU Parallel http://pi.dk/0 released. See what is new
+New release of #GNU Parallel pi.dk/0 New in this release pi.dk/2 See the intro videos pi.dk/1
-in this release http://pi.dk/2 Watch the intro videos http://pi.dk/1
 
 [x] Twitter
 Aspect: Public

src/parallel

@@ -543,7 +543,7 @@ sub get_options_from_array {
 sub parse_options {
     # Returns: N/A
     # Defaults:
-    $Global::version = 20120222;
+    $Global::version = 20120224;
     $Global::progname = 'parallel';
     $Global::infinity = 2**31;
     $Global::debug = 0;
@@ -2182,7 +2182,9 @@ sub swap_activity {
         -e $ENV{'HOME'}."/.parallel" or mkdir $ENV{'HOME'}."/.parallel";
         -e $ENV{'HOME'}."/.parallel/tmp" or mkdir $ENV{'HOME'}."/.parallel/tmp";
         my $swap_activity;
-	$swap_activity = "vmstat 1 2 | tail -n1 | awk '{print \$7*\$8}'";
+	$swap_activity = ("vmstat 1 2 2>/dev/null | tail -n1 | awk '{print \$7*\$8}' || ".
+			  # If the (remote) machine is Mac and the above fails, try this:
+			  "vm_stat 1 | head -n 3 | tail -n1 | awk '{print \$9*\$10}'");
         if($self->{'string'} ne ":") {
             $swap_activity = $self->sshcommand() . " " . $self->serverlogin() . " " .
 		::shell_quote_scalar($swap_activity);
@@ -2618,13 +2620,27 @@ sub no_of_cpus {
     # Returns:
     #   Number of physical CPUs
     local $/="\n"; # If delimiter is set, then $/ will be wrong
-    my $no_of_cpus = (no_of_cpus_freebsd()
+    my $no_of_cpus;
-                      || no_of_cpus_darwin()
+    if ($^O eq 'linux') {
-                      || no_of_cpus_solaris()
+        $no_of_cpus = no_of_cpus_gnu_linux();
-                      || no_of_cpus_aix()
+    } elsif ($^O eq 'freebsd') {
-                      || no_of_cpus_gnu_linux()
+        $no_of_cpus = no_of_cpus_freebsd();
-        );
+    } elsif ($^O eq 'solaris') {
+        $no_of_cpus = no_of_cpus_solaris();
+    } elsif ($^O eq 'aix') {
+        $no_of_cpus = no_of_cpus_aix();
+    } elsif ($^O eq 'darwin') {
+	$no_of_cpus = no_of_cpus_darwin();
+    } else {
+	$no_of_cpus = (no_of_cpus_freebsd()
+		       || no_of_cpus_darwin()
+		       || no_of_cpus_solaris()
+		       || no_of_cpus_aix()
+		       || no_of_cpus_gnu_linux()
+	    );
+    }
     if($no_of_cpus) {
+	chomp $no_of_cpus;
         return $no_of_cpus;
     } else {
         warn("parallel: Cannot figure out number of cpus. Using 1");
@@ -2636,13 +2652,27 @@ sub no_of_cores {
     # Returns:
     #   Number of CPU cores
     local $/="\n"; # If delimiter is set, then $/ will be wrong
-    my $no_of_cores = (no_of_cores_freebsd()
+    my $no_of_cores;
-                       || no_of_cores_darwin()
+    if ($^O eq 'linux') {
-                       || no_of_cores_solaris()
+	$no_of_cores = no_of_cores_gnu_linux();
-                       || no_of_cores_aix()
+    } elsif ($^O eq 'freebsd') {
-                       || no_of_cores_gnu_linux()
+        $no_of_cores = no_of_cores_freebsd();
-        );
+    } elsif ($^O eq 'solaris') {
+	$no_of_cores = no_of_cores_solaris();
+    } elsif ($^O eq 'aix') {
+        $no_of_cores = no_of_cores_aix();
+    } elsif ($^O eq 'darwin') {
+	$no_of_cores = no_of_cores_darwin();
+    } else {
+	$no_of_cores = (no_of_cores_freebsd()
+			|| no_of_cores_darwin()
+			|| no_of_cores_solaris()
+			|| no_of_cores_aix()
+			|| no_of_cores_gnu_linux()
+	    );
+    }
     if($no_of_cores) {
+	chomp $no_of_cores;
         return $no_of_cores;
     } else {
         warn("parallel: Cannot figure out number of CPU cores. Using 1");

src/parallel.pod

@@ -3133,23 +3133,59 @@ echo 1,2,3 | parallel -vkd, "echo 'a'{}'b'"
 echo 1,2,3 | parallel -vkd, "echo 'a'"{}"'b'"
 
 
-=head2 Startup speed
+=head2 Speed
 
-GNU B<parallel> is slow at starting up. Half of the startup time on
+=head3 Startup
-the local computer is spent finding the maximal length of a command
+
-line. Setting B<-s> will remove this part of the startup time.
+GNU B<parallel> is slow at starting up - around 250 ms. Half of the
+startup time on the local computer is spent finding the maximal length
+of a command line. Setting B<-s> will remove this part of the startup
+time.
+
+=head3 Job startup
+
+Starting a job takes around 3 ms. This can be a big overhead if the
+job takes very few ms to run. Often you can group small jobs together
+using B<-X> which will make the overhead less significant.
+
+Using B<--ungroup> the 3 ms can be lowered to around 2 ms.
+
+=head3 SSH
 
 When using multiple computers GNU B<parallel> opens B<ssh> connections
 to them to figure out how many connections can be used reliably
 simultaneously (Namely SSHD's MaxStartup). This test is done for each
-host in serial, so if your --sshloginfile contains many hosts it may
+host in serial, so if your B<--sshloginfile> contains many hosts it may
 be slow.
 
+=head3 Disk access
+
+A single disk can normally read data faster if it reads one file at a
+time instead of reading a lot of files in parallel, as this will avoid
+disk seeks. However, newer disk systems with multiple drives can read
+faster if reading from multiple files in parallel.
+
+If the jobs are of the form read-all-compute-all-write-all, so
+everything is read before anything is written, it may be faster to
+force only one disk access at the time:
+
+  sem --id diskio cat file | compute | sem --id diskio cat > file
+
+If the jobs are of the form read-compute-write, so writing starts
+before all reading is done, it may be faster to force only one reader
+and writer at the time:
+
+  sem --id read cat file | compute | sem --id write cat > file
+
+If the jobs are of the form read-compute-read-compute, it may be
+faster to run more jobs in parallel than the system has CPUs, as some
+of the jobs will be stuck waiting for disk access.
+
 =head2 --nice limits command length
 
 The current implementation of B<--nice> is too pessimistic in the max
 allowed command length. It only uses a little more than half of what
-it could. This affects -X and -m. If this becomes a real problem for
+it could. This affects B<-X> and B<-m>. If this becomes a real problem for
 you file a bug-report.
 
 =head2 Aliases and functions do not work

src/parallel.texi

@@ -3337,25 +3337,69 @@ echo 1,2,3 | parallel -vkd, "echo 'a'@{@}'b'"
 
 echo 1,2,3 | parallel -vkd, "echo 'a'"@{@}"'b'"
 
-@section Startup speed
+@section Speed
-@anchor{Startup speed}
+@anchor{Speed}
 
-GNU @strong{parallel} is slow at starting up. Half of the startup time on
+@subsection Startup
-the local computer is spent finding the maximal length of a command
+@anchor{Startup}
-line. Setting @strong{-s} will remove this part of the startup time.
+
+GNU @strong{parallel} is slow at starting up - around 250 ms. Half of the
+startup time on the local computer is spent finding the maximal length
+of a command line. Setting @strong{-s} will remove this part of the startup
+time.
+
+@subsection Job startup
+@anchor{Job startup}
+
+Starting a job takes around 3 ms. This can be a big overhead if the
+job takes very few ms to run. Often you can group small jobs together
+using @strong{-X} which will make the overhead less significant.
+
+Using @strong{--ungroup} the 3 ms can be lowered to around 2 ms.
+
+@subsection SSH
+@anchor{SSH}
 
 When using multiple computers GNU @strong{parallel} opens @strong{ssh} connections
 to them to figure out how many connections can be used reliably
 simultaneously (Namely SSHD's MaxStartup). This test is done for each
-host in serial, so if your --sshloginfile contains many hosts it may
+host in serial, so if your @strong{--sshloginfile} contains many hosts it may
 be slow.
 
+@subsection Disk access
+@anchor{Disk access}
+
+A single disk can normally read data faster if it reads one file at a
+time instead of reading a lot of files in parallel, as this will avoid
+disk seeks. However, newer disk systems with multiple drives can read
+faster if reading from multiple files in parallel.
+
+If the jobs are of the form read-all-compute-all-write-all, so
+everything is read before anything is written, it may be faster to
+force only one disk access at the time:
+
+@verbatim
+  sem --id diskio cat file | compute | sem --id diskio cat > file
+@end verbatim
+
+If the jobs are of the form read-compute-write, so writing starts
+before all reading is done, it may be faster to force only one reader
+and writer at the time:
+
+@verbatim
+  sem --id read cat file | compute | sem --id write cat > file
+@end verbatim
+
+If the jobs are of the form read-compute-read-compute, it may be
+faster to run more jobs in parallel than the system has CPUs, as some
+of the jobs will be stuck waiting for disk access.
+
 @section --nice limits command length
 @anchor{--nice limits command length}
 
 The current implementation of @strong{--nice} is too pessimistic in the max
 allowed command length. It only uses a little more than half of what
-it could. This affects -X and -m. If this becomes a real problem for
+it could. This affects @strong{-X} and @strong{-m}. If this becomes a real problem for
 you file a bug-report.
 
 @section Aliases and functions do not work