Re: stock 2.4.18-pre6 vs low latency

From: J Sloan (jjs@pobox.com)
Date: Sat Jan 26 2002 - 02:32:48 EST


This is a followup to my earlier post about the
effect of the low latency patch on 2.4.18-pre6.

As it was pointed out, the stock kernel didn't
do badly. This is true, so I have conducted a
more stressful test to help highlight the effects
of the low latency patch.

In this test I ran a dbench sequence of 3 reps
each of dbench (1, 2, 4 ... 128), while doing
normal desktop activities e.g. pysol, reading
mail, running aim. The tested system also acts
as an iptables firewall for 3 internal hosts, but
the load is light.

As the results show, the histogram for the
low latency kernel is short and sweet. The
histogram for the stock kernel is rather more
involved. The low latency kernel also shows
no degredation in dbench results compared
to the stock kernel.

The histograms were generated as before:

1. start realfeel
2. run the dbench test
3. send sigint to realfeel to dump the histograms
4. generate the graphs with the included script
5. compare and contrast the results

The dbench results:

2.4.18-pre6 + lowlat
-----------------------
Throughput 106.33 MB/sec (NB=132.913 MB/sec 1063.3 MBit/sec) 1 procs
Throughput 108.236 MB/sec (NB=135.295 MB/sec 1082.36 MBit/sec) 1 procs
Throughput 108.818 MB/sec (NB=136.022 MB/sec 1088.18 MBit/sec) 1 procs

Throughput 107.13 MB/sec (NB=133.913 MB/sec 1071.3 MBit/sec) 2 procs
Throughput 107.611 MB/sec (NB=134.514 MB/sec 1076.11 MBit/sec) 2 procs
Throughput 106.817 MB/sec (NB=133.521 MB/sec 1068.17 MBit/sec) 2 procs

Throughput 107.051 MB/sec (NB=133.814 MB/sec 1070.51 MBit/sec) 4 procs
Throughput 106.648 MB/sec (NB=133.31 MB/sec 1066.48 MBit/sec) 4 procs
Throughput 103.107 MB/sec (NB=128.883 MB/sec 1031.07 MBit/sec) 4 procs

Throughput 105.72 MB/sec (NB=132.15 MB/sec 1057.2 MBit/sec) 8 procs
Throughput 106.87 MB/sec (NB=133.588 MB/sec 1068.7 MBit/sec) 8 procs
Throughput 106.091 MB/sec (NB=132.613 MB/sec 1060.91 MBit/sec) 8 procs

Throughput 105.146 MB/sec (NB=131.433 MB/sec 1051.46 MBit/sec) 16 procs
Throughput 105.636 MB/sec (NB=132.045 MB/sec 1056.36 MBit/sec) 16 procs
Throughput 106.289 MB/sec (NB=132.861 MB/sec 1062.89 MBit/sec) 16 procs

Throughput 47.5776 MB/sec (NB=59.472 MB/sec 475.776 MBit/sec) 32 procs
Throughput 47.1291 MB/sec (NB=58.9113 MB/sec 471.291 MBit/sec) 32 procs
Throughput 48.5488 MB/sec (NB=60.686 MB/sec 485.488 MBit/sec) 32 procs

Throughput 30.9873 MB/sec (NB=38.7341 MB/sec 309.873 MBit/sec) 64 procs
Throughput 30.2279 MB/sec (NB=37.7849 MB/sec 302.279 MBit/sec) 64 procs
Throughput 31.0719 MB/sec (NB=38.8399 MB/sec 310.719 MBit/sec) 64 procs

Throughput 18.1556 MB/sec (NB=22.6945 MB/sec 181.556 MBit/sec) 128 procs
Throughput 16.0057 MB/sec (NB=20.0071 MB/sec 160.057 MBit/sec) 128 procs
Throughput 17.7952 MB/sec (NB=22.244 MB/sec 177.952 MBit/sec) 128 procs

2.4.18-pre6 (no lowlat)
-----------------------
Throughput 108.359 MB/sec (NB=135.449 MB/sec 1083.59 MBit/sec) 1 procs
Throughput 107.866 MB/sec (NB=134.833 MB/sec 1078.66 MBit/sec) 1 procs
Throughput 108.862 MB/sec (NB=136.077 MB/sec 1088.62 MBit/sec) 1 procs

Throughput 107.715 MB/sec (NB=134.643 MB/sec 1077.15 MBit/sec) 2 procs
Throughput 98.9497 MB/sec (NB=123.687 MB/sec 989.497 MBit/sec) 2 procs
Throughput 102.91 MB/sec (NB=128.637 MB/sec 1029.1 MBit/sec) 2 procs

Throughput 72.1596 MB/sec (NB=90.1995 MB/sec 721.596 MBit/sec) 4 procs
Throughput 85.7308 MB/sec (NB=107.164 MB/sec 857.308 MBit/sec) 4 procs
Throughput 103.053 MB/sec (NB=128.816 MB/sec 1030.53 MBit/sec) 4 procs

Throughput 84.795 MB/sec (NB=105.994 MB/sec 847.95 MBit/sec) 8 procs
Throughput 88.9176 MB/sec (NB=111.147 MB/sec 889.176 MBit/sec) 8 procs
Throughput 104.329 MB/sec (NB=130.411 MB/sec 1043.29 MBit/sec) 8 procs

Throughput 68.186 MB/sec (NB=85.2324 MB/sec 681.86 MBit/sec) 16 procs
Throughput 47.0743 MB/sec (NB=58.8429 MB/sec 470.743 MBit/sec) 16 procs
Throughput 46.0652 MB/sec (NB=57.5814 MB/sec 460.652 MBit/sec) 16 procs

Throughput 50.2502 MB/sec (NB=62.8127 MB/sec 502.502 MBit/sec) 32 procs
Throughput 51.1178 MB/sec (NB=63.8973 MB/sec 511.178 MBit/sec) 32 procs
Throughput 49.8005 MB/sec (NB=62.2506 MB/sec 498.005 MBit/sec) 32 procs

Throughput 27.9207 MB/sec (NB=34.9009 MB/sec 279.207 MBit/sec) 64 procs
Throughput 30.3247 MB/sec (NB=37.9059 MB/sec 303.247 MBit/sec) 64 procs
Throughput 31.0434 MB/sec (NB=38.8042 MB/sec 310.434 MBit/sec) 64 procs

Throughput 19.0205 MB/sec (NB=23.7757 MB/sec 190.205 MBit/sec) 128 procs
Throughput 18.0334 MB/sec (NB=22.5418 MB/sec 180.334 MBit/sec) 128 procs
Throughput 18.2095 MB/sec (NB=22.7619 MB/sec 182.095 MBit/sec) 128 procs

The latency histograms:

2.4.18-pre6 + lowlat
-----------------------
0.0 8580075
0.1 13384
0.2 350
0.3 167
0.4 54
0.5 7
0.6 5
0.7 3
0.8 1
2.7 1
33.6 1

2.4.18-pre6 (no lowlat)
-----------------------
0.0 8224742
0.1 120095
0.2 54023
0.3 56931
0.4 10825
0.5 2017
0.6 823
0.7 560
0.8 480
0.9 386
1.0 319
1.1 313
1.2 214
1.3 190
1.4 142
1.5 107
1.6 74
1.7 46
1.8 41
1.9 31
2.0 25
2.1 15
2.2 17
2.3 12
2.4 12
2.5 12
2.6 12
2.7 12
2.8 10
2.9 9
3.0 3
3.1 5
3.2 1
3.3 2
3.4 2
3.5 1
3.6 2
3.7 2
3.8 2
3.9 1
4.0 2
4.1 1
4.2 1
4.3 2
4.5 1
4.6 2
4.7 2
4.8 2
4.9 3
5.0 1
5.1 1
5.2 1
5.4 1
5.5 1
5.6 1
5.8 1
5.9 2
6.0 2
6.2 1
6.5 1
6.7 1
7.0 1
7.7 1
7.8 1
7.9 1
8.7 1
8.9 1
9.5 1
9.7 1
9.9 1
10.1 1
11.2 1
11.4 1
12.0 1
12.6 1
12.7 2
13.4 1
13.9 1
14.2 1
14.4 1
15.8 1
16.9 1
17.0 1
17.3 1
17.5 1
19.1 1
19.3 1
20.8 1
21.9 1
23.6 1
24.3 1
24.8 1
25.1 1
26.1 1
26.9 1
27.0 1
28.8 1
30.2 1
31.7 1
32.7 1
33.5 1
33.7 1
34.7 1
36.8 1
36.9 1
41.8 1
42.2 1
43.7 1
44.8 1
45.1 1
45.2 1
45.6 1
45.9 1

The 2 small pngs are attached.

Best Regards,

Joe





-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Thu Jan 31 2002 - 21:00:39 EST