* 前置き
-- スペックなんかの違い
Instance vCPU ECU メモリ（GiB） インスタンスストレージ（GB） Linux/UNIX 料金
c3.2xlarge 8 28 15 2 x 80 SSD $0.511 /1 時間
c4.2xlarge 8 31 15 EBS のみ $0.588 /1 時間
このインスタンスを比較対象にしたのは、サービスでc3.2xlargeを使っているからです(安易)
ということで、テストインスタンスでベンチマークを取ってみます。
以下の3点がc3→c4に変わった際の売りになっているそうです
・CPUの世代がHaswellになってクロック数も上がりましたよ。
・ネットワークが最適化されて早くなりましたよ
・ディスク領域がEBSに最適化されて早くなりましたよ
-- お値段
気になるのは料金が結構上がっていることでして、たった$0.77と思うなかれ
安定運用していて1ヶ月単価になると結構な差がでてきます。
・c3.2xlarge：$367.92
・c4.2xlarge：$423.36
$55.44の差額です。たかい。
* ベンチ開始
-- ApacheBench
総アクセス100万
同時connection数がどこまでエラーなくレスポンスを返せるか。
5回同じ項目テストして、エラーが出なかったらOk扱いということで・・・
# ab -n 1000000 -c $cnt http://<hostname>/
→Apacheデフォルトの静的コンテンツ(5.4kb)
- c3.2xlarge
・同時接続370で挫ける
・さばいた秒間リクエスト数の最大は7284.85回/sec
・テスト所要時間は130秒前後
- c4.2xlarge
・同じく同時接続370で挫ける
・さばいた秒間リクエスト数の最大は19130.87回/sec
・テスト所要時間は50秒前後
--ApacheBench総評
・c4.2xlargeが圧勝しました。
-- UnixBench
- c4.2xlarge
# ./Run- c3.2xlarge
------------------------------------------------------------------------
Benchmark Run: Thu Jan 22 2015 09:53:19 - 10:21:26
8 CPUs in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 38182137.1 lps (10.0 s, 7 samples)
Double-Precision Whetstone 4571.3 MWIPS (10.2 s, 7 samples)
Execl Throughput 4164.7 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1131344.2 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 298746.9 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 3547221.1 KBps (30.0 s, 2 samples)
Pipe Throughput 1996255.2 lps (10.0 s, 7 samples)
Pipe-based Context Switching 52031.5 lps (10.0 s, 7 samples)
Process Creation 11620.3 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 9885.9 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 4096.4 lpm (60.0 s, 2 samples)
System Call Overhead 2905706.4 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 38182137.1 3271.8
Double-Precision Whetstone 55.0 4571.3 831.1
Execl Throughput 43.0 4164.7 968.5
File Copy 1024 bufsize 2000 maxblocks 3960.0 1131344.2 2856.9
File Copy 256 bufsize 500 maxblocks 1655.0 298746.9 1805.1
File Copy 4096 bufsize 8000 maxblocks 5800.0 3547221.1 6115.9
Pipe Throughput 12440.0 1996255.2 1604.7
Pipe-based Context Switching 4000.0 52031.5 130.1
Process Creation 126.0 11620.3 922.2
Shell Scripts (1 concurrent) 42.4 9885.9 2331.6
Shell Scripts (8 concurrent) 6.0 4096.4 6827.4
System Call Overhead 15000.0 2905706.4 1937.1
========
System Benchmarks Index Score 1676.5
------------------------------------------------------------------------
Benchmark Run: Thu Jan 22 2015 10:21:26 - 10:49:35
8 CPUs in system; running 8 parallel copies of tests
Dhrystone 2 using register variables 176334699.4 lps (10.0 s, 7 samples)
Double-Precision Whetstone 32053.7 MWIPS (10.0 s, 7 samples)
Execl Throughput 23787.5 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1011008.1 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 262963.6 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 3250921.8 KBps (30.0 s, 2 samples)
Pipe Throughput 9439625.3 lps (10.0 s, 7 samples)
Pipe-based Context Switching 1686610.1 lps (10.0 s, 7 samples)
Process Creation 66159.9 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 38267.9 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 5348.4 lpm (60.0 s, 2 samples)
System Call Overhead 9603210.3 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 176334699.4 15110.1
Double-Precision Whetstone 55.0 32053.7 5827.9
Execl Throughput 43.0 23787.5 5532.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 1011008.1 2553.1
File Copy 256 bufsize 500 maxblocks 1655.0 262963.6 1588.9
File Copy 4096 bufsize 8000 maxblocks 5800.0 3250921.8 5605.0
Pipe Throughput 12440.0 9439625.3 7588.1
Pipe-based Context Switching 4000.0 1686610.1 4216.5
Process Creation 126.0 66159.9 5250.8
Shell Scripts (1 concurrent) 42.4 38267.9 9025.4
Shell Scripts (8 concurrent) 6.0 5348.4 8914.0
System Call Overhead 15000.0 9603210.3 6402.1
========
System Benchmarks Index Score 5603.6
# ./Run
------------------------------------------------------------------------
Benchmark Run: Thu Jan 22 2015 06:29:18 - 06:57:23
8 CPUs in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 35890021.0 lps (10.0 s, 7 samples)
Double-Precision Whetstone 4399.9 MWIPS (9.9 s, 7 samples)
Execl Throughput 4173.9 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1030714.9 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 277165.0 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 3078109.2 KBps (30.0 s, 2 samples)
Pipe Throughput 1755376.0 lps (10.0 s, 7 samples)
Pipe-based Context Switching 45995.4 lps (10.0 s, 7 samples)
Process Creation 11708.5 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 8682.1 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 3581.8 lpm (60.0 s, 2 samples)
System Call Overhead 2423669.1 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 35890021.0 3075.4
Double-Precision Whetstone 55.0 4399.9 800.0
Execl Throughput 43.0 4173.9 970.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 1030714.9 2602.8
File Copy 256 bufsize 500 maxblocks 1655.0 277165.0 1674.7
File Copy 4096 bufsize 8000 maxblocks 5800.0 3078109.2 5307.1
Pipe Throughput 12440.0 1755376.0 1411.1
Pipe-based Context Switching 4000.0 45995.4 115.0
Process Creation 126.0 11708.5 929.2
Shell Scripts (1 concurrent) 42.4 8682.1 2047.7
Shell Scripts (8 concurrent) 6.0 3581.8 5969.7
System Call Overhead 15000.0 2423669.1 1615.8
========
System Benchmarks Index Score 1530.0
------------------------------------------------------------------------
Benchmark Run: Thu Jan 22 2015 06:57:23 - 07:25:33
8 CPUs in system; running 8 parallel copies of tests
Dhrystone 2 using register variables 160166373.8 lps (10.0 s, 7 samples)
Double-Precision Whetstone 29244.4 MWIPS (9.9 s, 7 samples)
Execl Throughput 23354.0 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 764912.1 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 198669.2 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 2652773.6 KBps (30.0 s, 2 samples)
Pipe Throughput 8402747.0 lps (10.0 s, 7 samples)
Pipe-based Context Switching 1447497.7 lps (10.0 s, 7 samples)
Process Creation 62231.2 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 33078.3 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 4566.4 lpm (60.0 s, 2 samples)
System Call Overhead 9414061.7 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 160166373.8 13724.6
Double-Precision Whetstone 55.0 29244.4 5317.2
Execl Throughput 43.0 23354.0 5431.2
File Copy 1024 bufsize 2000 maxblocks 3960.0 764912.1 1931.6
File Copy 256 bufsize 500 maxblocks 1655.0 198669.2 1200.4
File Copy 4096 bufsize 8000 maxblocks 5800.0 2652773.6 4573.7
Pipe Throughput 12440.0 8402747.0 6754.6
Pipe-based Context Switching 4000.0 1447497.7 3618.7
Process Creation 126.0 62231.2 4939.0
Shell Scripts (1 concurrent) 42.4 33078.3 7801.5
Shell Scripts (8 concurrent) 6.0 4566.4 7610.6
System Call Overhead 15000.0 9414061.7 6276.0
========
System Benchmarks Index Score 4894.7
-- UnixBench総評
・こちらも当然c4.2xlargeが勝利
単体Core性能：c4.2xlarge 1676.5 vs c3.2xlarge 1530.0
Core合算性能：c4.2xlarge 5603.6 vs c3.2xlarge 4894.7
----
おわり
0 件のコメント:
コメントを投稿