Re: cloning the kernel - why long time in "Resolving 313037 deltas"

Johannes Schindelin [off-list ref] writes:

Hi,

in a very unscientific test, without your patch local cloning of the 
LilyPond repo takes 1m33s (user), and with your patch (pread() instead of 
mmap()) it takes 1m13s (user). The real times are somewhat bogus, but 
still in favour of pread(), but only by 8 seconds instead of 20.

This is on Linux 2.4.32.

Interesting.  Anybody have numbers from 2.6?
 

On Tue, Dec 19, 2006 at 01:10:59AM -0800, Junio C Hamano wrote:

quoted

in a very unscientific test, without your patch local cloning of the 
LilyPond repo takes 1m33s (user), and with your patch (pread() instead of 
mmap()) it takes 1m13s (user). The real times are somewhat bogus, but 
still in favour of pread(), but only by 8 seconds instead of 20.

This is on Linux 2.4.32.

Interesting.  Anybody have numbers from 2.6?

Similar results to those posted by Marco Roeland (numbers below are
index-pack of linux-2.6 on 2.6.18 i386 smp):

mmap
46.78user 4.18system 0:51.08elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+725627minor)pagefaults 0swaps

pread
45.70user 3.12system 0:48.96elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+363794minor)pagefaults 0swaps

Junio C Hamano wrote:

Johannes Schindelin [off-list ref] writes:

quoted

Hi,

in a very unscientific test, without your patch local cloning of the 
LilyPond repo takes 1m33s (user), and with your patch (pread() instead of 
mmap()) it takes 1m13s (user). The real times are somewhat bogus, but 
still in favour of pread(), but only by 8 seconds instead of 20.

This is on Linux 2.4.32.

Interesting.  Anybody have numbers from 2.6?

On my debian etch system:

Linux version 2.6.17-2-686 (Debian 2.6.17-9) (waldi@debian.org) (gcc 
version 4.1.2 20060901 (prerelease) (Debian 4.1.1-13)) #1 SMP Wed Sep 13 
16:34:10 UTC 2006

I did one run first (not included) to get things nice and warm, then 
three runs of each.  Overall the same as reported elsewhere marginally 
better with pread().  I guess you could say a 20-30% improvement in 
system time which isn't to be sniffed at.

-apw

mmap():

real    1m5.187s
user    1m0.844s
sys     0m2.900s

real    1m6.748s
user    1m0.868s
sys     0m3.064s

real    1m5.604s
user    1m0.760s
sys     0m3.124s


pread():

real    1m4.676s
user    1m0.168s
sys     0m2.340s

real    1m3.563s
user    0m59.796s
sys     0m2.248s

real    1m4.066s
user    1m0.156s
sys     0m2.304s

It was reported by Randal L. Schwartz [off-list ref] that 
indexing the Linux repository ~150MB pack takes about an hour on OS x 
while it's a minute on Linux.  It seems that the OS X mmap() 
implementation is more than 2 orders of magnitude slower than the Linux 
one.

Linus proposed a patch replacing mmap() with pread() bringing index-pack 
performance on OS X in line with the Linux one.  The performances on 
Linux also improved by a small margin.

Signed-off-by: Nicolas Pitre <redacted>

---

OK looks like this has been sorted out while I was away.  Good!

On Tue, 19 Dec 2006, Junio C Hamano wrote:

Johannes Schindelin [off-list ref] writes:

quoted

Hi,

in a very unscientific test, without your patch local cloning of the 
LilyPond repo takes 1m33s (user), and with your patch (pread() instead of 
mmap()) it takes 1m13s (user). The real times are somewhat bogus, but 
still in favour of pread(), but only by 8 seconds instead of 20.

This is on Linux 2.4.32.

Interesting.  Anybody have numbers from 2.6?

My 37 seconds of yesterday dropped to 32.

This is Linus's patch plus a few cosmetic changes.

diff --git a/index-pack.c b/index-pack.c
index 6d6c92b..e08a687 100644
--- a/index-pack.c
+++ b/index-pack.c

@@ -1,3 +1,8 @@
+#define _XOPEN_SOURCE 500
+#include <unistd.h>
+#include <sys/time.h>
+#include <signal.h>
+
 #include "cache.h"
 #include "delta.h"
 #include "pack.h"

@@ -6,8 +11,6 @@
 #include "commit.h"
 #include "tag.h"
 #include "tree.h"
-#include <sys/time.h>
-#include <signal.h>
 
 static const char index_pack_usage[] =
 "git-index-pack [-v] [-o <index-file>] [{ ---keep | --keep=<msg> }] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";

@@ -87,7 +90,7 @@ static unsigned display_progress(unsigned n, unsigned total, unsigned last_pc)
 static unsigned char input_buffer[4096];
 static unsigned long input_offset, input_len, consumed_bytes;
 static SHA_CTX input_ctx;
-static int input_fd, output_fd, mmap_fd;
+static int input_fd, output_fd, pack_fd;
 
 /* Discard current buffer used content. */
 static void flush(void)

@@ -148,14 +151,14 @@ static const char *open_pack_file(const char *pack_name)
 			output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
 		if (output_fd < 0)
 			die("unable to create %s: %s\n", pack_name, strerror(errno));
-		mmap_fd = output_fd;
+		pack_fd = output_fd;
 	} else {
 		input_fd = open(pack_name, O_RDONLY);
 		if (input_fd < 0)
 			die("cannot open packfile '%s': %s",
 			    pack_name, strerror(errno));
 		output_fd = -1;
-		mmap_fd = input_fd;
+		pack_fd = input_fd;
 	}
 	SHA1_Init(&input_ctx);
 	return pack_name;

@@ -279,27 +282,25 @@ static void *get_data_from_pack(struct object_entry *obj)
 {
 	unsigned long from = obj[0].offset + obj[0].hdr_size;
 	unsigned long len = obj[1].offset - from;
-	unsigned pg_offset = from % getpagesize();
-	unsigned char *map, *data;
+	unsigned char *src, *data;
 	z_stream stream;
 	int st;
 
-	map = mmap(NULL, len + pg_offset, PROT_READ, MAP_PRIVATE,
-		   mmap_fd, from - pg_offset);
-	if (map == MAP_FAILED)
-		die("cannot mmap pack file: %s", strerror(errno));
+	src = xmalloc(len);
+	if (pread(pack_fd, src, len, from) != len)
+		die("cannot pread pack file: %s", strerror(errno));
 	data = xmalloc(obj->size);
 	memset(&stream, 0, sizeof(stream));
 	stream.next_out = data;
 	stream.avail_out = obj->size;
-	stream.next_in = map + pg_offset;
+	stream.next_in = src;
 	stream.avail_in = len;
 	inflateInit(&stream);
 	while ((st = inflate(&stream, Z_FINISH)) == Z_OK);
 	inflateEnd(&stream);
 	if (st != Z_STREAM_END || stream.total_out != obj->size)
 		die("serious inflate inconsistency");
-	munmap(map, len + pg_offset);
+	free(src);
 	return data;
 }