Dobrica Pavlinušić's random unstructured stuff
Virtualization workshop: Revision 38

Materijali za Virtualizacija na Linuxu -- jednostavan izbor zar ne?

Contents


Hardware

Disk performance
Contents: [virtualization_workshop]


Have many disks. More disk spindles brings more than capacity alone! (Same as in databases)

Speed

Disk platter transfer speed

If you think that disk has constant transfer speed, ZCAV has interesting graphs

Individial disks

Slow laptop 2.5" 5400 disk

dpavlin@llin:~$ sudo hdparm -i /dev/sda

/dev/sda:

 Model=FUJITSU MHV2080BH                       , FwRev=00840028, SerialNo=        NW05T6B29HM5
 Config={ HardSect NotMFM HdSw>15uSec Fixed DTR>10Mbs }
 RawCHS=16383/16/63, TrkSize=0, SectSize=0, ECCbytes=4
 BuffType=DualPortCache, BuffSize=8192kB, MaxMultSect=16, MultSect=?16?
 CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=156301488
 IORDY=on/off, tPIO={min:240,w/IORDY:120}, tDMA={min:120,rec:120}
 PIO modes:  pio0 pio1 pio2 pio3 pio4 
 DMA modes:  mdma0 mdma1 mdma2 
 UDMA modes: udma0 udma1 udma2 udma3 udma4 *udma5 
 AdvancedPM=yes: mode=0x80 (128) WriteCache=enabled
 Drive conforms to: unknown:  ATA/ATAPI-3,4,5,6,7

 * signifies the current active mode

dpavlin@llin:~$ sudo hdparm -tT /dev/sda
/dev/sda:
 Timing cached reads:   1566 MB in  2.00 seconds = 782.85 MB/sec
 Timing buffered disk reads:   66 MB in  3.03 seconds =  21.79 MB/sec

Interesting numbers are BuffSize (cache in disk) and MaxMultSect which we want to use for read-ahead param:

hdparm -m 16 -a 16 /dev/sda

This will decrease a bit speed of linerar buffer reads which hdparm uses, but we will pull from disk only blocks which are allready in cache, improving random read/write performance.

To find optimal readahead for your drive using hdparm access pattern you can use hdparm-readahead.pl which will try different combinations for you.

Faster (!) external 3.5 USB disk (no hdparm -i on USB), but just because it's another disk not loaded by system.

dpavlin@llin:~$ sudo hdparm -tT /dev/sdb

/dev/sdb:
 Timing cached reads:   1508 MB in  2.00 seconds = 753.72 MB/sec
 Timing buffered disk reads:   56 MB in  3.03 seconds =  18.48 MB/sec

Software RAID

Home-made RAID disk holder

Home-made software md RAID 5 array from SATA drives:

Content from RAID5 disk array


Note nice usage of construction metal stripes with holes which is usually used to hold fence. It has holes just the right size for screws to go through and hold disks nicely spaced (although a little bit more space would be ideal). It's soft enough to be bent at corners to produce nice and leveled space between it and case.

Blog post RAID5 for home describes setup in some details.



Drive info:

dpavlin@brr:~$ sudo hdparm -i /dev/sdd

/dev/sdd:

 Model=WDC WD5000AAKS-00YGA0                   , FwRev=12.01C02, SerialNo=     WD-WCAS80929678
 Config={ HardSect NotMFM HdSw>15uSec SpinMotCtl Fixed DTR>5Mbs FmtGapReq }
 RawCHS=16383/16/63, TrkSize=0, SectSize=0, ECCbytes=50
 BuffType=unknown, BuffSize=16384kB, MaxMultSect=16, MultSect=?16?
 CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=976773168
 IORDY=on/off, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120}
 PIO modes:  pio0 pio3 pio4 
 DMA modes:  mdma0 mdma1 mdma2 
 UDMA modes: udma0 udma1 udma2 udma3 udma4 udma5 *udma6 
 AdvancedPM=no WriteCache=enabled
 Drive conforms to: Unspecified:  ATA/ATAPI-1,2,3,4,5,6,7

 * signifies the current active mode

Speed of individual drives in array:

dpavlin@brr:~$ sudo hdparm -tT /dev/sda /dev/sdb /dev/sdd

/dev/sda:
 Timing cached reads:   1982 MB in  2.00 seconds = 991.18 MB/sec
 Timing buffered disk reads:  232 MB in  3.03 seconds =  76.67 MB/sec

/dev/sdb:
 Timing cached reads:   2010 MB in  2.00 seconds = 1004.95 MB/sec
 Timing buffered disk reads:  228 MB in  3.01 seconds =  75.85 MB/sec

/dev/sdd:
 Timing cached reads:   2006 MB in  2.00 seconds = 1003.01 MB/sec
 Timing buffered disk reads:  230 MB in  3.01 seconds =  76.47 MB/sec

How are hey assembled into /dev/md0 RAID 5 array:

dpavlin@brr:~$ cat /proc/mdstat 
Personalities : [raid6] [raid5] [raid4] 
md0 : active raid5 sdd1[0] sda1[2] sdb1[1]
      976767872 blocks level 5, 64k chunk, algorithm 2 [3/3] [UUU]

Speed of array

dpavlin@brr:~$ sudo hdparm -tT /dev/md0

/dev/md0:
 Timing cached reads:   1986 MB in  2.00 seconds = 993.20 MB/sec
 Timing buffered disk reads:  434 MB in  3.01 seconds = 144.41 MB/sec

As expected RAID 5 speed is 75 + 75 + 0 (parity disk) ~ 144 MB/sec

Temperature

Disks don't like it hot!

root@brr:~# hddtemp /dev/sda /dev/sdb /dev/sdd
/dev/sda: WDC WD5000AAKS-00YGA0: 33°C
/dev/sdb: WDC WD5000AAKS-00YGA0: 32°C
/dev/sdd: WDC WD5000AAKS-00YGA0: 32°C

On output above, middle disk is /dev/sda so it's 1° hotter than other two. I could mitigate this with additional fan on front of case, but it's making enough noise already, so I'll leave it as is.

Data security

Smart

root@brr:~# smartctl --all /dev/sda | head -20
smartctl version 5.38 [i686-pc-linux-gnu] Copyright (C) 2002-8 Bruce Allen
Home page is http://smartmontools.sourceforge.net/

=== START OF INFORMATION SECTION ===
Model Family:     Western Digital Caviar Second Generation Serial ATA family
Device Model:     WDC WD5000AAKS-00YGA0
Serial Number:    WD-WCAS80815866
Firmware Version: 12.01C02
User Capacity:    500,107,862,016 bytes
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   8
ATA Standard is:  Exact ATA specification draft version not indicated
Local Time is:    Sat Oct 11 00:27:01 2008 CEST
SMART support is: Available - device has SMART capability.
SMART support is: Enabled

=== START OF READ SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED

Before you start to beleve in SMART as solution to all disk health problems, read Failure Trends in a Large Disk Drive Population

http://media.arstechnica.com/staff.media/failurehd.png

See also Bad block HOWTO for smartmontools if you ever get smart errors and don't just want to throw out your disk.

RAID

Also interesting is Some RAID Issues

Read also Why RAID 5 stops working in 2009



CPU

Support for hardware virtualization:

egrep '^flags.*(vmx|svm)' /proc/cpuinfo

  vmx/svm no vmx/svn
USB kvm qemu+kqemu
no USB VirtualBox  

How much CPU do I use? :-)

dpavlin@brr:~$ cpufreq-info 
cpufrequtils 004: cpufreq-info (C) Dominik Brodowski 2004-2006
Report errors and bugs to cpufreq@lists.linux.org.uk, please.
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which need to switch frequency at the same time: 0
  hardware limits: 2.40 GHz - 3.20 GHz
  available frequency steps: 3.20 GHz, 2.80 GHz, 2.40 GHz
  available cpufreq governors: userspace, powersave, ondemand, conservative, performance
  current policy: frequency should be within 2.40 GHz and 3.20 GHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 2.40 GHz.
  cpufreq stats: 3.20 GHz:1.80%, 2.80 GHz:0.00%, 2.40 GHz:98.20%  (17)

KVM/QEMU

kqemu module compilation on Debian:

module-assistant a-i kqemu

Windows

Remove them:

cd c:\windows\system32\drivers
del agp440.sys
del intelppm.dll

Startup script:

# 3M RFID 810
usbdev=0403:6001

sudo chown -R $USER /proc/bus/usb/*

kvm -m 512 -hda win-xp.vmdk -no-acpi -std-vga -monitor stdio -usb -usbdevice host:$usbdev

USB sniffing

info usbhost

VirtualBox

Seems to be best supported right now (package in Debian, optional drivers for Windows, starting unmodified VMWare machines -- after you guess right settings that is!)

OSE version (no USB!) comes in Debian, compile vboxdrv with:

root@llin:~# module-assistant a-i virtualbox-ose

OpenVZ

disk speed

dpavlin@zut:~$ sudo hdparm -tT /dev/cciss/c1d0 /dev/sda

/dev/cciss/c1d0:
 Timing cached reads:   2184 MB in  2.00 seconds = 1092.39 MB/sec
 Timing buffered disk reads:  324 MB in  3.02 seconds = 107.40 MB/sec

/dev/sda:
 Timing cached reads:   2144 MB in  2.00 seconds = 1071.89 MB/sec
 Timing buffered disk reads:  136 MB in  3.02 seconds =  45.02 MB/sec

Insert joke about enterprise storage

Add disk space to VE

First, resize logical volume:

root@koha-hw:~# vgextend -L +80G /dev/vg/vz
vgextend: invalid option -- L
  Error during parsing of command line.

root@koha-hw:~# lvextend -L +80G /dev/vg/vz
  Extending logical volume vz to 100.00 GB
  Logical volume vz successfully resized

root@koha-hw:~# resize2fs /dev/vg/vz 
resize2fs 1.40-WIP (14-Nov-2006)
Filesystem at /dev/vg/vz is mounted on /vz; on-line resizing required
old desc_blocks = 2, new_desc_blocks = 7
Performing an on-line resize of /dev/vg/vz to 26214400 (4k) blocks.
The filesystem on /dev/vg/vz is now 26214400 blocks long.

root@koha-hw:~# df -h /vz/
Filesystem            Size  Used Avail Use% Mounted on
/dev/mapper/vg-vz      99G   20G   79G  21% /vz

Then, take a look how much space does VEs take:

root@koha-hw:~# vzlist -o veid,diskspace,diskspace.s,diskspace.h,diskinodes,diskinodes.s,diskspace.h
      VEID   DQBLOCKS DQBLOCKS.S DQBLOCKS.H   DQINODES DQINODES.S DQBLOCKS.H
    212052   11717220   15728640   20971520      61001     286527   20971520
    212226    6407804   10485760   12582912      69011     435472   12582912

alternativly, you can also execute df inside VEs:

root@koha-hw:~# vzlist -o veid -H | xargs -i sh -c "echo --{}-- ; vzctl exec {} df -h"
--212052--
Filesystem            Size  Used Avail Use% Mounted on
simfs                  15G   12G  3.9G  75% /
tmpfs                 2.0G     0  2.0G   0% /lib/init/rw
tmpfs                 2.0G     0  2.0G   0% /dev/shm
--212226--
Filesystem            Size  Used Avail Use% Mounted on
simfs                  10G  6.2G  3.9G  62% /
tmpfs                 2.0G     0  2.0G   0% /lib/init/rw
tmpfs                 2.0G     0  2.0G   0% /dev/shm

next, we will set diskpace on both VEs (becase we want them to share all available resources) to new logical volume size:

root@koha-hw:~# vzlist -o veid -H | xargs -i vzctl set {} --diskspace 100G:100G --save
Saved parameters for VE 212052
Saved parameters for VE 212226

This VEs are not in production, and one is development version of another. When we move to production, we want to enforce more strict limit on disk usage, to protect production machine from running out of disk space in case the development one goes wild.

VMWare

Convert image to monolithic growable disk

This format is supported by other emulators, so it's a best choice.

dpavlin@llin:/rest/vmware/winxp$ vmware-vdiskmanager -r Windows\ XP\ Professional.vmdk -t 0 /mnt/usb/vmware/win-xp.vmdk 
Using log file /tmp/vmware-dpavlin/vdiskmanager.log
Creating a monolithic growable disk '/mnt/usb/vmware/win-xp.vmdk'
  Convert: 57% done.

Resize disk image

dpavlin@llin:/mnt/usb/vmware$ qemu-img info win-xp.vmdk
(VMDK) image open: flags=0x2 filename=win-xp.vmdk
image: win-xp.vmdk
file format: vmdk
virtual size: 3.0G (3221225472 bytes)
disk size: 3.0G

There is a way to extend image using only qemu-img, but that involves converting image to raw and appending zeros at end to produce larger image. However, we will do that using VMWare's vmware-vdiskmanager

dpavlin@llin:/mnt/usb/vmware$ vmware-vdiskmanager -x 6Gb win-xp.vmdk
Using log file /tmp/vmware-dpavlin/vdiskmanager.log
  Grow: 100% done.
The old geometry C/H/S of the disk is: 6241/16/63
The new geometry C/H/S of the disk is: 12483/16/63
Disk expansion completed successfully.

WARNING: If the virtual disk is partitioned, you must use a third-party
         utility in the virtual machine to expand the size of the
         partitions. For more information, see:
         http://www.vmware.com/support/kb/enduser/std_adp.php?p_faqid=1647

This will make disk unbootable, so we will have to resize partition. Download GParted live CD and resize partition using it...

kvm -m 512 -hda win-xp.vmdk -no-acpi -std-vga -cdrom /rest/iso/gparted-live-0.3.9-4.iso -boot d

Convert vmdk to qcow

dpavlin@llin:/mnt/usb/vmware$ qemu-img convert -O qcow win-xp.vmdk win-xp.qcow
(VMDK) image open: flags=0x2 filename=win-xp.vmdk
dpavlin@llin:/mnt/usb/vmware$ ls -al win-xp.*
-rw-r--r-- 1 dpavlin dpavlin 3190906880 Oct  9 17:41 win-xp.qcow
-rw------- 1 dpavlin dpavlin 3208577024 Oct  9 17:35 win-xp.vmdk

Xen

disk speed

this is domU

root@vega:~# uname -a
Linux vega 2.6.18-6-xen-amd64 #1 SMP Mon Jun 16 23:42:47 UTC 2008 x86_64 GNU/Linux
root@vega:~# hdparm -tT /dev/hda1

/dev/hda1:
 Timing cached reads:   5488 MB in  2.00 seconds = 2750.74 MB/sec
 Timing buffered disk reads:  318 MB in  3.00 seconds = 105.98 MB/sec

Guest OS

Solaris

It will not boot pass "Loading Nexenta..." stage without kvm module loaded.

# to install from iso image
kvm -m 512 -hda solaris.vmdk -cdrom ../iso/nexenta-core-platform_1.0.1-b85-test4_x86.iso -boot d -net nic,model=rtl8139 -net user

# run after installation
kvm -m 512 -hda solaris.vmdk -net nic,model=rtl8139 -net user

Darwin