qemu-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-discuss] 4k Disk I/O issue


From: by morphin
Subject: [Qemu-discuss] 4k Disk I/O issue
Date: Tue, 22 Jan 2019 19:13:07 +0300

Hello.

I'm getting too much bad return from my customers at 4k random speeds. They
claim that SQL databases works much slower than usual, and I can't find a
solution.
I've done a lot of disk speed testing and I tried too many different
options for increase 4k random speed but still I can only reach half of the
physical disk speed.
Because of that I want to ask your advice. Please warn me if there is any
wrong in my tests and my xml.
Recently I'm using writeback - threads and "<driver queues='4'
iothread='4'/>" and I'm using CEPH with RBD but in this test I used the
same SSD RAW disk to be more consistent and fair.

Let's see our test environment:

*HOST OS:*
Kernel: 4.14.71-1-lts Arch-linux
qemu-headless 3.0.0-1
libvirt 4.7.0-1

*HOST HW:*
Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz
68GB++ Free ram
LSI-9300_8e HBA - FW=15.15.06.00

*DİSK:*
Source path: /dev/disk/by-id/ata-INTEL_SSDSC2KG480G7_BTYM72940CU2480BGN
Options: RAW, Writeback, threads

*GUEST:*
Win-2016
Driver: virtio-win-0.1.141
Guest xml: https://paste.ubuntu.com/p/w3W9BddzqY/
Also I tried q35-3.0

*TEST RESULTS*

*1. *First of all this is Pysical disk result without virtualization.

[image: maarif-ssdtest-fiziksel.jpeg]

*2. *i440FX - cache=none - IOmode=threads

[image: i440fx-none-io1-threads-cpu241.png]

*3. *i440FX - cache=directsync - IOmode=native

[image: ssdtest-directsync-native-io1.png]

*4**. *i440FX - cache=writeback - IOmode=threads - <driver queues='4'
iothread='4'/>

[image: i440fx-writeback-Q4-io4-threads+hugepage4096-cpu181.png]

*5. *i440FX - cache=writeback - IOmode=threads - <driver queues='8'
iothread='4'/>

[image: i440fx-writeback-Q8-io4-threads-cpu141.png]

*5. *i440FX - cache=writethrough - IOmode=threads - <driver queues='4'
iothread='4'/>

[image: i440fx-writethrough-io4-threads-cpu241.png]

*6. *i440FX - cache=unsafe - IOmode=threads - <driver queues='4'
iothread='4'/>

[image: i440fx-unsafe-io4-threads-cpu241.png]

*NOW LETS TRY Q35-3.0 CHIPSET*

*1. *Q35-3.0 - cache=none - IOmode=native - <driver queues='4'
iothread='1'/>

[image: q35-3.0-cachenone-Q4-io1-native-cpu142-Mem4.png]

*2. *Q35-3.0 - cache=none - IOmode=threads - <driver queues='8'
iothread='8'/>

[image: q35-3.0-cachenone-Q8-io8-threads-cpu142-Mem4.png]


*3. *Q35-3.0 - cache=writeback - IOmode=threads - <driver queues='4'
iothread='1'/>

[image: q35-3.0-writeback-Q4-io1-threads-cpu811-Mem4.png]

*4. *Q35-3.0 - cache=writeback - IOmode=threads - <driver queues='8'
iothread='8'/>

[image: q35-3.0-writeback-Q8-io8-threads-cpu142-Mem4.png]


*AND FINALLY I CHECKED OUR ENEMY TEST RESULTS "vmware 6.5 - local ssd
datastore & vmdk " *
They are closer to the physical disk as random and sequential.

[image: maarif-ssdtest-vmware-test1.png]

JPEG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image

PNG image


reply via email to

[Prev in Thread] Current Thread [Next in Thread]