概要

こちらの投稿を参考にKVM上でMeltdown/Spectreパッチ適用後に性能試験を実施しました。
今回はとりあえずunixBenchのみです。適宜検証したら追記したいと思います。

Redhatの参考情報
https://access.redhat.com/ja/security/vulnerabilities/3311961

前提条件

検証するVMが動作するKVMホストは、パッチ適用済です。
検証するVMは、CentOS7.3、CPU:4core、Memory:16GB

バージョン情報

VM(パッチ未適用)

$ uname -a
Linux vm1 3.10.0-693.5.2.el7.x86_64 #1 SMP Fri Oct 13 10:46:25 EDT 2017 x86_64 x86_64 x86_64 GNU/Linux

VM(パッチ適用)

$ uname -a
Linux vm2 3.10.0-693.11.6.el7.x86_64 #1 SMP Thu Jan 4 01:06:37 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

脆弱性チェック

https://raw.githubusercontent.com/speed47/spectre-meltdown-checker/master/spectre-meltdown-checker.sh を実行

パッチ未適用VM

Spectre Variant 1,2,3に脆弱性を確認

$ sudo sh ./spectre-meltdown-checker.sh
Spectre and Meltdown mitigation detection tool v0.31

Checking for vulnerabilities against running kernel Linux 3.10.0-693.5.2.el7.x86_64 #1 SMP Fri Oct 13 10:46:25 EDT 2017 x86_64
CPU is QEMU Virtual CPU version 1.5.3

CVE-2017-5753 [bounds check bypass] aka 'Spectre Variant 1'
* Checking count of LFENCE opcodes in kernel:  NO
> STATUS:  VULNERABLE  (only 21 opcodes found, should be >= 70, heuristic to be improved when official patches become available)

CVE-2017-5715 [branch target injection] aka 'Spectre Variant 2'
* Mitigation 1
*   Hardware (CPU microcode) support for mitigation
*     The SPEC_CTRL MSR is available:  YES
*     The SPEC_CTRL CPUID feature bit is set:  NO
*   Kernel support for IBRS:  NO
*   IBRS enabled for Kernel space:  NO
*   IBRS enabled for User space:  NO
* Mitigation 2
*   Kernel compiled with retpoline option:  NO
*   Kernel compiled with a retpoline-aware compiler:  NO
> STATUS:  VULNERABLE  (IBRS hardware + kernel support OR kernel with retpoline are needed to mitigate the vulnerability)

CVE-2017-5754 [rogue data cache load] aka 'Meltdown' aka 'Variant 3'
* Kernel supports Page Table Isolation (PTI):  NO
* PTI enabled and active:  NO
* Checking if we're running under Xen PV (64 bits):  NO
> STATUS:  VULNERABLE  (PTI is needed to mitigate the vulnerability)

A false sense of security is worse than no security at all, see --disclaimer

パッチ適用VM

Spectre Variant 2に脆弱性を確認。それ以外はOK。

$ sudo sh ./spectre-meltdown-checker.sh
Spectre and Meltdown mitigation detection tool v0.31

Checking for vulnerabilities against running kernel Linux 3.10.0-693.11.6.el7.x86_64 #1 SMP Thu Jan 4 01:06:37 UTC 2018 x86_64
CPU is QEMU Virtual CPU version 1.5.3

CVE-2017-5753 [bounds check bypass] aka 'Spectre Variant 1'
* Checking count of LFENCE opcodes in kernel:  YES
> STATUS:  NOT VULNERABLE  (106 opcodes found, which is >= 70, heuristic to be improved when official patches become available)

CVE-2017-5715 [branch target injection] aka 'Spectre Variant 2'
* Mitigation 1
*   Hardware (CPU microcode) support for mitigation
*     The SPEC_CTRL MSR is available:  YES
*     The SPEC_CTRL CPUID feature bit is set:  NO
*   Kernel support for IBRS:  YES
*   IBRS enabled for Kernel space:  NO
*   IBRS enabled for User space:  NO
* Mitigation 2
*   Kernel compiled with retpoline option:  NO
*   Kernel compiled with a retpoline-aware compiler:  NO
> STATUS:  VULNERABLE  (IBRS hardware + kernel support OR kernel with retpoline are needed to mitigate the vulnerability)

CVE-2017-5754 [rogue data cache load] aka 'Meltdown' aka 'Variant 3'
* Kernel supports Page Table Isolation (PTI):  YES
* PTI enabled and active:  YES
* Checking if we're running under Xen PV (64 bits):  NO
> STATUS:  NOT VULNERABLE  (PTI mitigates the vulnerability)

A false sense of security is worse than no security at all, see --disclaimer

unixBenchの結果

4coreのテストのみを記載

パッチ未適用VM

------------------------------------------------------------------------
Benchmark Run: Wed Jan 17 2018 15:07:20 - 15:14:03
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables      131231880.5 lps   (10.0 s, 1 samples)
Double-Precision Whetstone                    16937.1 MWIPS (9.6 s, 1 samples)
Execl Throughput                              16987.2 lps   (29.3 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks       1389846.0 KBps  (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks          385034.0 KBps  (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks       4373700.0 KBps  (30.0 s, 1 samples)
Pipe Throughput                             6748830.2 lps   (10.0 s, 1 samples)
Pipe-based Context Switching                1417922.6 lps   (10.0 s, 1 samples)
Process Creation                              44711.5 lps   (30.0 s, 1 samples)
Shell Scripts (1 concurrent)                  21357.6 lpm   (60.0 s, 1 samples)
Shell Scripts (8 concurrent)                   3414.1 lpm   (60.0 s, 1 samples)
System Call Overhead                        6678039.5 lps   (10.0 s, 1 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0  131231880.5  11245.2
Double-Precision Whetstone                       55.0      16937.1   3079.5
Execl Throughput                                 43.0      16987.2   3950.5
File Copy 1024 bufsize 2000 maxblocks          3960.0    1389846.0   3509.7
File Copy 256 bufsize 500 maxblocks            1655.0     385034.0   2326.5
File Copy 4096 bufsize 8000 maxblocks          5800.0    4373700.0   7540.9
Pipe Throughput                               12440.0    6748830.2   5425.1
Pipe-based Context Switching                   4000.0    1417922.6   3544.8
Process Creation                                126.0      44711.5   3548.5
Shell Scripts (1 concurrent)                     42.4      21357.6   5037.2
Shell Scripts (8 concurrent)                      6.0       3414.1   5690.2
System Call Overhead                          15000.0    6678039.5   4452.0
                                                                   ========
System Benchmarks Index Score                                        4523.3

パッチ適用VM

------------------------------------------------------------------------
Benchmark Run: Wed Jan 17 2018 15:29:47 - 15:36:30
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables      128905413.5 lps   (10.0 s, 1 samples)
Double-Precision Whetstone                    16450.1 MWIPS (9.8 s, 1 samples)
Execl Throughput                              11097.3 lps   (29.0 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks       1203052.0 KBps  (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks          313511.0 KBps  (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks       3669716.0 KBps  (30.0 s, 1 samples)
Pipe Throughput                             2038905.8 lps   (10.0 s, 1 samples)
Pipe-based Context Switching                 638231.7 lps   (10.0 s, 1 samples)
Process Creation                              34631.1 lps   (30.0 s, 1 samples)
Shell Scripts (1 concurrent)                  16080.8 lpm   (60.0 s, 1 samples)
Shell Scripts (8 concurrent)                   2448.6 lpm   (60.0 s, 1 samples)
System Call Overhead                        1568599.1 lps   (10.0 s, 1 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0  128905413.5  11045.9
Double-Precision Whetstone                       55.0      16450.1   2990.9
Execl Throughput                                 43.0      11097.3   2580.8
File Copy 1024 bufsize 2000 maxblocks          3960.0    1203052.0   3038.0
File Copy 256 bufsize 500 maxblocks            1655.0     313511.0   1894.3
File Copy 4096 bufsize 8000 maxblocks          5800.0    3669716.0   6327.1
Pipe Throughput                               12440.0    2038905.8   1639.0
Pipe-based Context Switching                   4000.0     638231.7   1595.6
Process Creation                                126.0      34631.1   2748.5
Shell Scripts (1 concurrent)                     42.4      16080.8   3792.6
Shell Scripts (8 concurrent)                      6.0       2448.6   4081.0
System Call Overhead                          15000.0    1568599.1   1045.7
                                                                   ========
System Benchmarks Index Score                                        2905.0

unixBenchの性能比較結果

参考にしたqiitaの記事通りのテスト項目が著しく低下することを確認しました。

System Benchmarks Index Values	パッチ適用前	パッチ適用後	性能差(%)
Dhrystone 2 using register variables	11245.2	11045.9	98.23%
Double-Precision Whetstone	3079.5	2990.9	97.12%
Execl Throughput	3950.5	2580.8	65.33%
File Copy 1024 bufsize 2000 maxblocks	3509.7	3038	86.56%
File Copy 256 bufsize 500 maxblocks	2326.5	1894.3	81.42%
File Copy 4096 bufsize 8000 maxblocks	7540.9	6327.1	83.90%
Pipe Throughput	5425.1	1639	30.21%
Pipe-based Context Switching	3544.8	1595.6	45.01%
Process Creation	3548.5	2748.5	77.46%
Shell Scripts (1 concurrent)	5037.2	3792.6	75.29%
Shell Scripts (8 concurrent)	5690.2	4081	71.72%
System Call Overhead	4452	1045.7	23.49%
System Benchmarks Index Score	4523.3	2905.0	64.22%

テスト項目の意味は、こちらを参考にしました。
実際にちゃんと本番相当の負荷をかけて本格的なテストをしたかったですが、一旦ここまでとします。続報があれば追記します。

2018/01/23追記

mysqlslapの結果

こちらの記事を参考に、mysqlslapの性能を検証しました。
全体的に10%から20%前後の性能劣化が確認できました。特にinsert系の性能劣化が目立ちます。

テスト内容	テスト結果の分類	パッチ適用前	パッチ適用後	性能差(%)
50スレッド 1000行データ 1000クエリ read(テーブルスキャン)	Average	0.602	0.676	-12.29%
	Minimum	0.495	0.548	-10.71%
	Maximum	0.873	1.016	-16.38%
50スレッド 1000行データ 1000クエリ write(テーブルへの挿入)	Average	0.148	0.173	-16.89%
	Minimum	0.124	0.152	-22.58%
	Maximum	0.173	0.196	-13.29%
50スレッド 1000行データ 1000クエリ key(主キー読み取り)	Average	0.075	0.085	-13.33%
	Minimum	0.069	0.079	-14.49%
	Maximum	0.085	0.095	-11.76%
50スレッド 1000行データ 1000クエリ mixed(挿入とテーブルスキャンを半々)	Average	0.106	0.121	-14.15%
	Minimum	0.098	0.104	-6.12%
	Maximum	0.13	0.128	1.54%
50スレッド 1000行データ 1000クエリ update(更新)	Average	0.176	0.189	-7.39%
	Minimum	0.159	0.175	-10.06%
	Maximum	0.197	0.209	-6.09%

また続報があれば追記します。