QED : qcow2より早い、最速ディスクフォーマット #KVM

使った感想

イメージ容量が結構増える。断片化を極力しないように
まとまった領域を利用しているのかな

QED: A New, High Performance QEMU Disk Format - Phoronix

Linux-KVM mentions QED, the new QEMU Enhanced Disk format. This new disk format for QEMU/KVM is designed to be much faster than QCOW2 and other existing disk formats available to virtualization users.

2011年から存在してた様子。今2015年なのでもう安定していると見て良いのでは。
使ってみた感じ引っ掛かりがなく早い様子。

QCOW2のパフォーマンスは、qemu-0.14.0以降で大きく改善されている。

Fedora21 + QCOW2だと体感できる引っ掛かりがある。
- Fedora21 + QEDだと大丈夫っぽい

QEDはKVM専用のフォーマット

第7章イメージ形式の変換 - OpenStack 仮想マシンイメージガイド - カレント

qcow3というものも。

Features/Qcow3 - QEMU

ベンチマーク結果らしい

UnixBench (Fedora21上)

   BYTE UNIX Benchmarks (Version 5.1.3)

   System: wwwXX.local: GNU/Linux
   OS: GNU/Linux -- 2.6.18-402.el5PAE -- #1 SMP Tue Feb 10 18:29:18 EST 2015
   Machine: i686 (i386)
   Language: en_US.utf8 (charmap="UTF-8", collate="UTF-8")
   CPU 0: QEMU Virtual CPU version 2.1.3 (5600.0 bogomips)
          x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
   CPU 1: QEMU Virtual CPU version 2.1.3 (5599.7 bogomips)
          x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
   CPU 2: QEMU Virtual CPU version 2.1.3 (6157.8 bogomips)
          x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
   CPU 3: QEMU Virtual CPU version 2.1.3 (5944.2 bogomips)
          x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET
   16:05:08 up 8 min,  1 user,  load average: 0.00, 0.14, 0.13; runlevel

------------------------------------------------------------------------
Benchmark Run: 火  2月 17 2015 16:05:08 - 16:33:23
4 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables       19187540.4 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     2830.7 MWIPS (10.1 s, 7 samples)
Execl Throughput                               3150.0 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        486775.1 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          122197.1 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       1553066.6 KBps  (30.0 s, 2 samples)
Pipe Throughput                              648617.5 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 132098.6 lps   (10.0 s, 7 samples)
Process Creation                              13702.2 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   7559.2 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   2190.5 lpm   (60.0 s, 2 samples)
System Call Overhead                        1392085.2 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   19187540.4   1644.2
Double-Precision Whetstone                       55.0       2830.7    514.7
Execl Throughput                                 43.0       3150.0    732.6
File Copy 1024 bufsize 2000 maxblocks          3960.0     486775.1   1229.2
File Copy 256 bufsize 500 maxblocks            1655.0     122197.1    738.4
File Copy 4096 bufsize 8000 maxblocks          5800.0    1553066.6   2677.7
Pipe Throughput                               12440.0     648617.5    521.4
Pipe-based Context Switching                   4000.0     132098.6    330.2
Process Creation                                126.0      13702.2   1087.5
Shell Scripts (1 concurrent)                     42.4       7559.2   1782.8
Shell Scripts (8 concurrent)                      6.0       2190.5   3650.9
System Call Overhead                          15000.0    1392085.2    928.1
                                                                   ========
System Benchmarks Index Score                                        1045.4

------------------------------------------------------------------------
Benchmark Run: 火  2月 17 2015 16:33:23 - 17:01:41
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       39946752.2 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                    10119.7 MWIPS (9.9 s, 7 samples)
Execl Throughput                              13117.0 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        149799.0 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks           41334.2 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks        521744.6 KBps  (30.0 s, 2 samples)
Pipe Throughput                             1914502.7 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 748018.3 lps   (10.0 s, 7 samples)
Process Creation                              40092.2 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                  17342.0 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   2572.6 lpm   (60.0 s, 2 samples)
System Call Overhead                        3779586.3 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   39946752.2   3423.0
Double-Precision Whetstone                       55.0      10119.7   1839.9
Execl Throughput                                 43.0      13117.0   3050.5
File Copy 1024 bufsize 2000 maxblocks          3960.0     149799.0    378.3
File Copy 256 bufsize 500 maxblocks            1655.0      41334.2    249.8
File Copy 4096 bufsize 8000 maxblocks          5800.0     521744.6    899.6
Pipe Throughput                               12440.0    1914502.7   1539.0
Pipe-based Context Switching                   4000.0     748018.3   1870.0
Process Creation                                126.0      40092.2   3181.9
Shell Scripts (1 concurrent)                     42.4      17342.0   4090.1
Shell Scripts (8 concurrent)                      6.0       2572.6   4287.6
System Call Overhead                          15000.0    3779586.3   2519.7
                                                                   ========
System Benchmarks Index Score                                        1717.9