FAQ

Dennis Clarke wrote:
A little data never hurts. Even if the numbers mean little.

test 1 - Debian Linux 6.0.5 on x86_64
Given the fact, that you did not run star -no-fifo, you compare an insecure
implementation (gtar never calls fsync(2)) with a secure by default
implementation (star).

Also note: ZFS has one real problem: it is really slow if you force it to grant
a stable sate on the medium. This is why ZFS is aprox. 4x slower in your test
star (without -no-fsync) compared to gtar.

ext3 is slow with star (without -no-fsync) because ext3 it is not optimized.
ext3 is fast with gtar because it cheats.

Solaris with UFS:

bzip2 -d < /tmp/linux-3.5.1.tar.bz2 > /dev/null
19.909r 19.770u 0.110s 99% 0M 0+0k 0st 0+0io 0pf+0w


OK, 19.77 seconds user CPU time.....

star -xp -xdot -time < /tmp/linux-3.5.1.tar.bz2
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 70.477sec (6647 kBytes/sec)
1:10.489r 23.020u 9.010s 45% 0M 0+0k 0st 0+0io 0pf+0w

star -xp -xdot -time -no-fsync < /tmp/linux-3.5.1.tar.bz2
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 74.161sec (6317 kBytes/sec)
1:14.174r 21.840u 4.640s 35% 0M 0+0k 0st 0+0io 0pf+0w

Only half of the System CPU time here because fsync(2) calls are missing...

gtar --totals -xf /tmp/linux-3.5.1.tar.bz2
Gesamtzahl gelesener Bytes: 479733760 (458MiB, 4,5MiB/s)
1:42.658r 23.150u 5.530s 27% 0M 0+0k 0st 0+0io 0pf+0w

gtar does not have a FIFO and thus is slower than star.


Now uncompressed (you see that the user CPU time for bzip2 is missing):

star -xp -xdot -time < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 70.438sec (6651 kBytes/sec)
1:10.449r 0.520u 8.190s 12% 0M 0+0k 0st 0+0io 0pf+0w

star -xp -xdot -time -no-fsync < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 86.624sec (5408 kBytes/sec)
1:26.636r 0.300u 3.960s 4% 0M 0+0k 0st 0+0io 0pf+0w

gtar --totals -xf /tmp/linux-3.5.1.tar
Gesamtzahl gelesener Bytes: 479733760 (458MiB, 4,7MiB/s)
1:38.829r 0.440u 4.570s 5% 0M 0+0k 0st 0+0io 0pf+0w


Now ZFS being on a single disk like the UFS test before (let us omit
the test with the compressed archive):

star -xp -xdot -time < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 394.783sec (1186 kBytes/sec)
6:34.795r 0.580u 8.250s 2% 0M 0+0k 0st 0+0io 0pf+0w

As expected: ZFS is slow if you force it to grant a stable state.

star -xp -xdot -time -no-fsync < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 11.085sec (42259 kBytes/sec)
11.096r 0.290u 4.380s 42% 0M 0+0k 0st 0+0io 0pf+0w

gtar --totals -xf /tmp/linux-3.5.1.tar
Gesamtzahl gelesener Bytes: 479733760 (458MiB, 39MiB/s)
11.929r 0.360u 4.260s 38% 0M 0+0k 0st 0+0io 0pf+0w

As you see, if you permit star to be as insecure as gtar always is, star is
always faster than gtar.

Now the same on a thumper (ZFS on RAIDZ2):

star -xp -xdot -time < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 349.575sec (1340 kBytes/sec)
5:49.595r 0.690u 11.270s 3% 0M 0+0k 0st 0+0io 0pf+0w

star -xp -xdot -time -no-fsync < /tmp/linux-3.5.1.tar
star: 46849 blocks + 0 bytes (total of 479733760 bytes = 468490.00k).
star: Total time 4.626sec (101251 kBytes/sec)
4.654r 0.330u 4.990s 115% 0M 0+0k 0st 0+0io 0pf+0w

gtar --totals -xf /tmp/linux-3.5.1.tar
Total bytes read: 479733760 (458MiB, 85MiB/s)
5.510r 0.430u 4.130s 82% 0M 0+0k 0st 0+0io 0pf+0w

J?rg

--
EMail:joerg at schily.isdn.cs.tu-berlin.de (home) J?rg Schilling D-13353 Berlin
js at cs.tu-berlin.de (uni)
joerg.schilling at fokus.fraunhofer.de (work) Blog: http://schily.blogspot.com/
URL: http://cdrecord.berlios.de/private/ ftp://ftp.berlios.de/pub/schily

Search Discussions

Discussion Posts

Previous

Follow ups

Related Discussions

People

Translate

site design / logo © 2017 Grokbase