computerの日記

Cisco,SHELL,C,Qt,C++,Linux,ネットワーク,Windows Scriptなどの発言です

Spectre & Meltdown Patch 適用前、適用後、nopti による、UnixBench の結果。

Spectre & Meltdown Patch 適用前、適用後、nopti による、UnixBench の結果。

OS: CentOS 6.9

古いカーネル
2.6.32-696.el6

BYTE UNIX Benchmarks (Version 5.1.3)

System: localhost.localdomain: GNU/Linux
OS: GNU/Linux -- 2.6.32-696.el6.x86_64 -- #1 SMP Tue Mar 21 19:29:05 UTC
2017
Machine: x86_64 (x86_64)
Language: en_US.utf8 (charmap="UTF-8", collate="UTF-8")
CPU 0: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz (5184.1 bogomips)
x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
21:58:20 up 29 min, 2 users, load average: 0.21, 0.08, 0.02; runlevel
2018-02-20

------------------------------------------------------------------------
Benchmark Run: 火 2月 20 2018 21:58:20 - 22:26:29
1 CPU in system; running 1 parallel copy of tests

Dhrystone 2 using register variables 40858230.7 lps (10.0 s, 7 samples)
Double-Precision Whetstone 4610.9 MWIPS (9.8 s, 7 samples)
Execl Throughput 5350.3 lps (29.9 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1172325.8 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 311188.0 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 2578243.0 KBps (30.0 s, 2 samples)
Pipe Throughput 1954746.6 lps (10.0 s, 7 samples)
Pipe-based Context Switching 418002.3 lps (10.0 s, 7 samples)
Process Creation 14349.6 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 6546.5 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 884.9 lpm (60.0 s, 2 samples)
System Call Overhead 3074144.1 lps (10.0 s, 7 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 40858230.7 3501.1
Double-Precision Whetstone 55.0 4610.9 838.3
Execl Throughput 43.0 5350.3 1244.2
File Copy 1024 bufsize 2000 maxblocks 3960.0 1172325.8 2960.4
File Copy 256 bufsize 500 maxblocks 1655.0 311188.0 1880.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 2578243.0 4445.2
Pipe Throughput 12440.0 1954746.6 1571.3
Pipe-based Context Switching 4000.0 418002.3 1045.0
Process Creation 126.0 14349.6 1138.9
Shell Scripts (1 concurrent) 42.4 6546.5 1544.0
Shell Scripts (8 concurrent) 6.0 884.9 1474.8
System Call Overhead 15000.0 3074144.1 2049.4
========
System Benchmarks Index Score 1743.4

 

 Spectre & Meltdown patched カーネル
2.6.32-696.18.7

BYTE UNIX Benchmarks (Version 5.1.3)

System: localhost.localdomain: GNU/Linux
OS: GNU/Linux -- 2.6.32-696.18.7.el6.x86_64 -- #1 SMP Thu Jan 4 17:31:22 UTC
2018
Machine: x86_64 (x86_64)
Language: en_US.utf8 (charmap="UTF-8", collate="UTF-8")
CPU 0: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz (5184.1 bogomips)
x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
23:57:53 up 1 min, 2 users, load average: 1.24, 0.53, 0.20; runlevel
2018-02-20

------------------------------------------------------------------------
Benchmark Run: 火 2月 20 2018 23:57:53 - 00:26:02
1 CPU in system; running 1 parallel copy of tests

Dhrystone 2 using register variables 37888385.8 lps (10.0 s, 7 samples)
Double-Precision Whetstone 4506.9 MWIPS (10.0 s, 7 samples)
Execl Throughput 3336.3 lps (29.9 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 341121.1 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 90343.2 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 1180761.0 KBps (30.0 s, 2 samples)
Pipe Throughput 492884.3 lps (10.0 s, 7 samples)
Pipe-based Context Switching 150710.6 lps (10.0 s, 7 samples)
Process Creation 11108.0 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 4923.0 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 682.2 lpm (60.0 s, 2 samples)
System Call Overhead 431462.7 lps (10.0 s, 7 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 37888385.8 3246.6
Double-Precision Whetstone 55.0 4506.9 819.4
Execl Throughput 43.0 3336.3 775.9
File Copy 1024 bufsize 2000 maxblocks 3960.0 341121.1 861.4
File Copy 256 bufsize 500 maxblocks 1655.0 90343.2 545.9
File Copy 4096 bufsize 8000 maxblocks 5800.0 1180761.0 2035.8
Pipe Throughput 12440.0 492884.3 396.2
Pipe-based Context Switching 4000.0 150710.6 376.8
Process Creation 126.0 11108.0 881.6
Shell Scripts (1 concurrent) 42.4 4923.0 1161.1
Shell Scripts (8 concurrent) 6.0 682.2 1136.9
System Call Overhead 15000.0 431462.7 287.6
========
System Benchmarks Index Score 824.5

 

 同カーネルで、カーネルオプションに、nopti を追加して実施した結果

BYTE UNIX Benchmarks (Version 5.1.3)

System: localhost.localdomain: GNU/Linux
OS: GNU/Linux -- 2.6.32-696.18.7.el6.x86_64 -- #1 SMP Thu Jan 4 17:31:22 UTC
2018
Machine: x86_64 (x86_64)
Language: en_US.utf8 (charmap="UTF-8", collate="UTF-8")
CPU 0: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz (5184.1 bogomips)
x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
03:50:07 up 1 min, 2 users, load average: 1.08, 0.44, 0.16; runlevel
2018-02-21

------------------------------------------------------------------------
Benchmark Run: 水 2月 21 2018 03:50:07 - 04:18:15
1 CPU in system; running 1 parallel copy of tests

Dhrystone 2 using register variables 38510363.9 lps (10.0 s, 7 samples)
Double-Precision Whetstone 4467.9 MWIPS (9.8 s, 7 samples)
Execl Throughput 4413.4 lps (29.9 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 870529.3 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 255228.5 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 2259120.5 KBps (30.0 s, 2 samples)
Pipe Throughput 1494173.8 lps (10.0 s, 7 samples)
Pipe-based Context Switching 216663.0 lps (10.0 s, 7 samples)
Process Creation 12555.5 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 6162.9 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 802.3 lpm (60.0 s, 2 samples)
System Call Overhead 1684728.3 lps (10.0 s, 7 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 38510363.9 3299.9
Double-Precision Whetstone 55.0 4467.9 812.4
Execl Throughput 43.0 4413.4 1026.4
File Copy 1024 bufsize 2000 maxblocks 3960.0 870529.3 2198.3
File Copy 256 bufsize 500 maxblocks 1655.0 255228.5 1542.2
File Copy 4096 bufsize 8000 maxblocks 5800.0 2259120.5 3895.0
Pipe Throughput 12440.0 1494173.8 1201.1
Pipe-based Context Switching 4000.0 216663.0 541.7
Process Creation 126.0 12555.5 996.5
Shell Scripts (1 concurrent) 42.4 6162.9 1453.5
Shell Scripts (8 concurrent) 6.0 802.3 1337.2
System Call Overhead 15000.0 1684728.3 1123.2
========
System Benchmarks Index Score 1388.7

 

 Page Table Isolation が効いたことにより、性能低下が認められました。
nopti により、かなり戻ったかも。
もっとも、それがいいかは別問題として。