Discussion:
[BENCHMARK] 2.5.51 with contest
(too old to reply)
Con Kolivas
2002-12-10 11:44:45 UTC
Permalink
Here are contest results (http://contest.kolivas.net) for 2.5.51 and related
kerneles using the dedicated osdl (http://www.osdl.org) hardware.

Uniprocessor:
noload:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 70.0 96 0 0 1.05
2.5.50 [5] 69.9 96 0 0 1.05
2.5.50-mm1 [5] 71.4 94 0 0 1.07
2.5.51 [2] 69.8 96 0 0 1.05

cacherun:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 67.4 99 0 0 1.01
2.5.50 [5] 67.3 99 0 0 1.01
2.5.50-mm1 [5] 67.8 99 0 0 1.02
2.5.51 [2] 67.2 99 0 0 1.01

process_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 85.2 79 17 20 1.28
2.5.50 [5] 84.8 79 17 19 1.27
2.5.50-mm1 [5] 86.6 78 18 20 1.30
2.5.51 [2] 85.2 79 17 20 1.28

dbench_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 210.5 37 2 50 3.15
2.5.50 [5] 189.2 40 2 49 2.83
2.5.50-mm1 [5] 243.3 34 2 51 3.64
2.5.51 [12] 195.8 39 2 50 2.93

ctar_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 106.1 82 2 9 1.59
2.5.50 [5] 107.5 81 3 9 1.61
2.5.50-mm1 [5] 88.0 83 1 4 1.32
2.5.51 [7] 107.0 81 3 9 1.60

xtar_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 184.8 70 3 8 2.77
2.5.50 [5] 189.5 61 4 9 2.84
2.5.50-mm1 [5] 104.9 70 1 6 1.57
2.5.51 [7] 163.7 67 3 8 2.45

io_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 127.4 57 14 13 1.91
2.5.50 [5] 142.6 54 19 14 2.14
2.5.50-mm1 [5] 174.2 46 24 15 2.61
2.5.51 [7] 125.6 58 14 12 1.88

io_other:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 97.4 75 7 11 1.46
2.5.50 [5] 106.9 69 10 11 1.60
2.5.50-mm1 [5] 101.8 70 9 11 1.52
2.5.51 [7] 105.1 69 9 11 1.57

read_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 88.2 80 15 6 1.32
2.5.50 [5] 88.5 80 15 7 1.33
2.5.50-mm1 [5] 86.6 80 3 2 1.30
2.5.51 [2] 88.4 80 15 7 1.32

list_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 81.4 85 0 8 1.22
2.5.50 [5] 81.2 85 0 8 1.22
2.5.50-mm1 [5] 82.4 84 0 7 1.23
2.5.51 [2] 80.8 85 0 8 1.21

mem_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 98.1 76 43 2 1.47
2.5.50 [5] 98.3 76 44 2 1.47
2.5.50-mm1 [5] 116.9 67 47 1 1.75
2.5.51 [7] 99.3 76 45 2 1.49

A little shorter compile times under io load and xtar load.


SMP:
noload:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [6] 39.3 181 0 0 1.09
2.5.50 [5] 39.3 180 0 0 1.09
2.5.50-mm1 [6] 39.4 181 0 0 1.09
2.5.51 [3] 39.6 180 0 0 1.09

cacherun:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [6] 36.6 194 0 0 1.01
2.5.50 [5] 36.5 194 0 0 1.01
2.5.50-mm1 [6] 36.6 194 0 0 1.01
2.5.51 [3] 36.5 195 0 0 1.01

process_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [6] 50.0 141 11 52 1.38
2.5.50 [5] 47.8 148 10 46 1.32
2.5.50-mm1 [5] 47.6 150 8 43 1.31
2.5.51 [3] 50.5 139 12 54 1.39

dbench_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 119.8 96 0 26 3.31
2.5.50 [5] 199.8 101 0 24 5.52
2.5.50-mm1 [5] 164.3 67 0 29 4.54
2.5.51 [7] 57.9 144 0 27 1.60

ctar_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 53.8 161 1 10 1.49
2.5.50 [5] 54.6 157 1 10 1.51
2.5.50-mm1 [5] 51.3 155 0 4 1.42
2.5.51 [7] 58.2 158 1 10 1.61

xtar_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 72.9 132 1 10 2.01
2.5.50 [5] 116.2 103 2 10 3.21
2.5.50-mm1 [5] 83.9 111 1 9 2.32
2.5.51 [7] 104.8 124 2 10 2.89

io_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 75.5 110 9 18 2.09
2.5.50 [5] 87.6 102 14 22 2.42
2.5.50-mm1 [5] 99.0 92 14 21 2.73
2.5.51 [7] 84.6 102 13 21 2.34

io_other:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 64.2 130 8 19 1.77
2.5.50 [5] 59.3 139 7 18 1.64
2.5.50-mm1 [5] 70.5 125 10 22 1.95
2.5.51 [7] 64.5 134 7 18 1.78

read_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 49.1 152 5 7 1.36
2.5.50 [5] 49.3 151 5 7 1.36
2.5.50-mm1 [5] 52.1 142 2 3 1.44
2.5.51 [3] 48.5 154 5 7 1.34

list_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 43.4 167 0 8 1.20
2.5.50 [5] 43.4 167 0 8 1.20
2.5.50-mm1 [5] 44.0 167 0 7 1.22
2.5.51 [3] 43.5 167 0 8 1.20

mem_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 62.5 145 35 3 1.73
2.5.50 [5] 63.3 141 36 3 1.75
2.5.50-mm1 [5] 67.1 126 39 3 1.85
2.5.51 [7] 62.6 148 38 3 1.73

The big difference in dbench_load in the SMP run is unusual but probably not
as significant as it appears. This load seems to occasionally create
pathological runs (probably the nature of running dbench as a load) and less
pathological runs occured in 2.5.51.

Overall the difference b/w 2.5.50 and 2.5.51 is small

Full hardware details and log runs can be found when the web server syncs up
at :
http://www.osdl.org/projects/ctdevel/results/

Con
Stan Bubrouski
2002-12-10 17:18:08 UTC
Permalink
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Here are contest results (http://contest.kolivas.net) for 2.5.51 and related
kerneles using the dedicated osdl (http://www.osdl.org) hardware.
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 70.0 96 0 0 1.05
2.5.50 [5] 69.9 96 0 0 1.05
2.5.50-mm1 [5] 71.4 94 0 0 1.07
2.5.51 [2] 69.8 96 0 0 1.05
I know this has been brought up before, but
these don't seem to mean much unless you
include 2.4.20 in the comaprison.

-Stan
Robert Love
2002-12-10 18:45:42 UTC
Permalink
Post by Stan Bubrouski
I know this has been brought up before, but
these don't seem to mean much unless you
include 2.4.20 in the comaprison.
Comparing this to 2.4 achieves nothing because so much changed.

The point of these benchmarks are not marketing, but to find
improvements or regressions from one version to the next and find out
what caused them.

Comparing the kernel to 2.4 has some uses (i.e. finding micro-ops) but
Con's mission is much different (and imo more useful).

Robert Love
Stan Bubrouski
2002-12-10 20:02:26 UTC
Permalink
Post by Robert Love
Post by Stan Bubrouski
I know this has been brought up before, but
these don't seem to mean much unless you
include 2.4.20 in the comaprison.
Comparing this to 2.4 achieves nothing because so much changed.
I disagree, 2.4.20 is the current stable kernel, it would
be nice to see how it compares to the current development,
what's faster, what's not... from Con's previous results
we can see that some things are indeed not as fast in 2.5.x
as in 2.4.x. It's just nice to be able to see the whole
picture. I often follow these threads for just this purpose.

-Stan
Post by Robert Love
The point of these benchmarks are not marketing, but to find
improvements or regressions from one version to the next and find out
what caused them.
Comparing the kernel to 2.4 has some uses (i.e. finding micro-ops) but
Con's mission is much different (and imo more useful).
Robert Love
Arador
2002-12-10 20:51:25 UTC
Permalink
On Tue, 10 Dec 2002 15:02:26 -0500
Post by Stan Bubrouski
I disagree, 2.4.20 is the current stable kernel, it would
be nice to see how it compares to the current development,
what's faster, what's not... from Con's previous results
we can see that some things are indeed not as fast in 2.5.x
as in 2.4.x. It's just nice to be able to see the whole
picture. I often follow these threads for just this purpose.
It's faster or slower. Comparing 2.5 vs 2.5 gives you a picture
of what's gone better, and why. IMHO.


Diego Calleja
Robert Love
2002-12-10 21:05:27 UTC
Permalink
Post by Stan Bubrouski
I disagree, 2.4.20 is the current stable kernel, it would
be nice to see how it compares to the current development,
what's faster, what's not... from Con's previous results
we can see that some things are indeed not as fast in 2.5.x
as in 2.4.x. It's just nice to be able to see the whole
picture. I often follow these threads for just this purpose.
Like I said, that may give users warm fuzzies or be helpful to marketing
folks but Con's benchmark is not really useful for _helping developers_
wrt comparing 2.4 vs 2.5.

A benchmark like AIM9, which is a bunch of micro-benchmarks, is useful
because we can say "look truncating a zero-length file is a lot slower
now".

But a contest result from 2.4 to 2.5 tells us what? Especially since
lower times in contest may not even be bad. Contest is invaluable for
testing one change vs. without. In fact, I would venture to say Con's
work is a big reason why 2.5 has the fairness and interactive
performance it does. But it is not so helpful to see changes since 2.4.

Robert Love

Con Kolivas
2002-12-10 20:24:40 UTC
Permalink
Post by Con Kolivas
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Here are contest results (http://contest.kolivas.net) for 2.5.51 and
related
kerneles using the dedicated osdl (http://www.osdl.org) hardware.
Kernel [runs] Time CPU% Loads LCPU% Ratio
2.5.49 [5] 70.0 96 0 0 1.05
2.5.50 [5] 69.9 96 0 0 1.05
2.5.50-mm1 [5] 71.4 94 0 0 1.07
2.5.51 [2] 69.8 96 0 0 1.05
I know this has been brought up before, but
these don't seem to mean much unless you
include 2.4.20 in the comaprison.
Repeated benchmarks of each successive release allow to detect subtle
differences of each change. If contest shows these changes the people who can
act on them will see them clearly if displayed only with relevant results. If
you want previous results, a full comparison is available at the full logs of
all previous benchmarks as I mention.

http://www.osdl.org/projects/ctdevel/results/

Con
Continue reading on narkive:
Loading...