I have been struggling with a problem for a couple of days and I hope someone has some ideas. I am using SLES 10 as a base for my Dom0 and I am trying to build a paravirtualized guest domain. I have tried lots of combinations of things but the disk during the install is increcdibly slow. I am fairly certain that this is something I have set wrong, as I see the same results on two very different hardware platforms. It takes about 2 hours for the first CD install; and that time is the same if I use the cd drive as a source of my install or if I use an NFS based source. Below is my domain config file --- I have tried lots of combinations of things but still the same result. -------------------------------------------------------------------------------------------------------------------------- disk = [ ''file:/var/lib/xen/images/basesles10/hda,hda,w'' ] memory = 348 vcpus = 1 builder = ''linux'' kernel = ''/tmp/YaST2-04473-dmQhaD/vmlinuz-2.6.16.21-0.8-xen'' ramdisk = ''/tmp/YaST2-04473-dmQhaD/inst-initrd'' name = ''basesles10'' ip = ''192.168.1.253'' netmask = ''255.255.255.0'' gateway = ''192.168.1.1'' interface = ''eth0'' vif = [ ''mac=00:16:3e:d1:dc:36,bridge=xenbr0'', ''mac=00:16:3e:e2:e6:a0,bridge=xenbr1'' ] localtime = 0 on_poweroff = ''destroy'' on_reboot = ''destroy'' on_crash = ''destroy'' import os, re arch = os.uname()[4] if re.search(''64'', arch): arch_libdir = ''lib64'' else: arch_libdir = ''lib'' device_model = ''/usr/'' + arch_libdir + ''/xen/bin/qemu-dm'' #---------------------------------------------------------------------------- # enable SDL library for graphics, default = 0 sdl=0 #---------------------------------------------------------------------------- # enable VNC library for graphics, default = 1 vnc=1 #---------------------------------------------------------------------------- # enable spawning vncviewer(only valid when vnc=1), default = 1 vncviewer=1 #---------------------------------------------------------------------------- # no graphics, use serial port #nographic=0 #---------------------------------------------------------------------------- # enable stdvga, default = 0 (use cirrus logic device model) stdvga=0 #----------------------------------------------------------------------------- # serial port re-direct to pty deivce, /dev/pts/n # then xm console or minicom can connect serial=''pty'' #---------------------------------------------------------------------------- # enable ne2000, default = 0(use pcnet) ne2000=0 #----------------------------------------------------------------------------- # enable audio support #audio=1 #----------------------------------------------------------------------------- # set the real time clock to local time [default=0 i.e. set to utc] #localtime=1 #----------------------------------------------------------------------------- # start in full screen #full-screen=1 extra = ''install=nfs://192.168.1.9//opt/software/SLES10/disk1 TERM=xterm textmode=0 vnc=1 hostip=192.168.1.253 netmask=255.255.255.0 ide=nodma gateway=192.168.1.1 netdevice=eth0'' ------------------------------------------------------------------------------------------------------------------ The ide=nodma is a farily recent addtion; it did not seem at have any effect on things. Oddly, I did not see this when I used full virtualization. That installation was very normal. I am positive that I need to set something slightly different - I just do not know what it it. Thanks for any suggestions. -- View this message in context: http://www.nabble.com/Slow-Disk-access-on-SLES-10-install-tf3338478.html#a9284951 Sent from the Xen - User mailing list archive at Nabble.com. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Matthias Lüneberg
2007-Mar-03 18:20 UTC
Re: [Xen-users] Slow Disk access on SLES 10 install
Hi, it seems to me that you use a file for the disk,>disk = [ ''file:/var/lib/xen/images/basesles10/hda,hda,w'' ]if you use a fresh SLES10 installation, you should read SLES10 Release Notes: http://www.novell.com/linux/releasenotes/i386/SUSE-SLES/10/RELEASE-NOTES.en.html Theres a paragraph called: "Sync Mode for Loopback Storage" which could help. Hope this helps! Regards, Matthias Am Samstag, den 03.03.2007, 05:20 -0800 schrieb D Sims:> I have been struggling with a problem for a couple of days and I hope someone > has some ideas. > > I am using SLES 10 as a base for my Dom0 and I am trying to build a > paravirtualized guest domain. > > I have tried lots of combinations of things but the disk during the install > is increcdibly slow. I am fairly certain that this is > something I have set wrong, as I see the same results on two very different > hardware platforms. It takes about 2 hours for the first CD install; and > that time is the same if I use the cd drive as a source of my install or if > I use an NFS based source. > > > Below is my domain config file --- I have tried lots of combinations of > things but still the same result. > > > -------------------------------------------------------------------------------------------------------------------------- > disk = [ ''file:/var/lib/xen/images/basesles10/hda,hda,w'' ] > memory = 348 > vcpus = 1 > builder = ''linux'' > kernel = ''/tmp/YaST2-04473-dmQhaD/vmlinuz-2.6.16.21-0.8-xen'' > ramdisk = ''/tmp/YaST2-04473-dmQhaD/inst-initrd'' > name = ''basesles10'' > ip = ''192.168.1.253'' > netmask = ''255.255.255.0'' > gateway = ''192.168.1.1'' > interface = ''eth0'' > vif = [ ''mac=00:16:3e:d1:dc:36,bridge=xenbr0'', > ''mac=00:16:3e:e2:e6:a0,bridge=xenbr1'' ] > localtime = 0 > on_poweroff = ''destroy'' > on_reboot = ''destroy'' > on_crash = ''destroy'' > > import os, re > arch = os.uname()[4] > if re.search(''64'', arch): > arch_libdir = ''lib64'' > else: > arch_libdir = ''lib'' > > device_model = ''/usr/'' + arch_libdir + ''/xen/bin/qemu-dm'' > > > #---------------------------------------------------------------------------- > # enable SDL library for graphics, default = 0 > sdl=0 > > #---------------------------------------------------------------------------- > # enable VNC library for graphics, default = 1 > vnc=1 > > #---------------------------------------------------------------------------- > # enable spawning vncviewer(only valid when vnc=1), default = 1 > vncviewer=1 > > #---------------------------------------------------------------------------- > # no graphics, use serial port > #nographic=0 > > #---------------------------------------------------------------------------- > # enable stdvga, default = 0 (use cirrus logic device model) > stdvga=0 > > #----------------------------------------------------------------------------- > # serial port re-direct to pty deivce, /dev/pts/n > # then xm console or minicom can connect > serial=''pty'' > > #---------------------------------------------------------------------------- > # enable ne2000, default = 0(use pcnet) > ne2000=0 > > > #----------------------------------------------------------------------------- > # enable audio support > #audio=1 > > > #----------------------------------------------------------------------------- > # set the real time clock to local time [default=0 i.e. set to utc] > #localtime=1 > > > #----------------------------------------------------------------------------- > # start in full screen > #full-screen=1 > > extra = ''install=nfs://192.168.1.9//opt/software/SLES10/disk1 TERM=xterm > textmode=0 vnc=1 hostip=192.168.1.253 netmask=255.255.255.0 ide=nodma > gateway=192.168.1.1 netdevice=eth0'' > > ------------------------------------------------------------------------------------------------------------------ > > The ide=nodma is a farily recent addtion; it did not seem at have any > effect on things. > > Oddly, I did not see this when I used full virtualization. That > installation was very normal. > > I am positive that I need to set something slightly different - I just do > not know what it it. > > Thanks for any suggestions._______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Matthias: You were 100% correct. That was the piece of information that I did not know that I needed. As soon as I changed the setting in the /etc/xen/scripts/block script from do_or_die losetup -y "$loopdev" "$file" do_or_die losetup "$loopdev" "$file" this made all the difference in the world. However, this does beg another question. What is the best practice for setting up a virtual disk for a guest domain? Is it better to live with the possibility of data corruption on a host crash and have normal disk access, or turn the flag on and be protected against corruption, but suffer terrible performance. Or is there a better way? Possibly to assign the guest domains their own disk partition that they can write to as they please? I have not tried this yet (I will tomorrow), but that sounds like a better bet. Or is there another way? Thanks a ton for your help. Dwain Sims Hi, it seems to me that you use a file for the disk,>disk = [ ''file:/var/lib/xen/images/basesles10/hda,hda,w'' ]if you use a fresh SLES10 installation, you should read SLES10 Release Notes: http://www.novell.com/linux/releasenotes/i386/SUSE-SLES/10/RELEASE-NOTES.en.html Theres a paragraph called: "Sync Mode for Loopback Storage" which could help. Hope this helps! Regards, Matthias Am Samstag, den 03.03.2007, 05:20 -0800 schrieb D Sims:> I have been struggling with a problem for a couple of days and I hope > someone > has some ideas. > > I am using SLES 10 as a base for my Dom0 and I am trying to build a > paravirtualized guest domain. > > I have tried lots of combinations of things but the disk during the > install > is increcdibly slow. I am fairly certain that this is > something I have set wrong, as I see the same results on two very > different > hardware platforms. It takes about 2 hours for the first CD install; and > that time is the same if I use the cd drive as a source of my install or > if > I use an NFS based source. >-- View this message in context: http://www.nabble.com/Slow-Disk-access-on-SLES-10-install-tf3338478.html#a9303459 Sent from the Xen - User mailing list archive at Nabble.com. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On 3 Mar 2007 at 5:20, D Sims wrote:> > I have been struggling with a problem for a couple of days and I hope someone > has some ideas. > > I am using SLES 10 as a base for my Dom0 and I am trying to build a > paravirtualized guest domain. > > I have tried lots of combinations of things but the disk during the install > is increcdibly slow. I am fairly certain that this is > something I have set wrong, as I see the same results on two very different > hardware platforms. It takes about 2 hours for the first CD install; and > that time is the same if I use the cd drive as a source of my install or if > I use an NFS based source.Pardon, but SLES10 is on DVD AFAIK. Ulrich _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Per Andreas Buer
2007-Mar-05 17:41 UTC
Re: [Xen-users] Slow Disk access on SLES 10 install
D Sims wrote:> Is it better to live with the possibility of data corruption on a host crash > and have normal disk access, or turn the flag on and be protected against > corruption, but suffer terrible performance. Or is there a better way? > Possibly to assign the guest domains their own disk partition that they can > write to as they please? I have not tried this yet (I will tomorrow), but > that sounds like a better bet. Or is there another way? >lvm gives you normal performace - and as a bonus - snapshots. I can''t say whether this is integrated with YaST in any way. Per. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Daniel P. Berrange
2007-Mar-05 17:50 UTC
Re: [Xen-users] Slow Disk access on SLES 10 install
On Sun, Mar 04, 2007 at 05:57:30PM -0800, D Sims wrote:> > Matthias: > > You were 100% correct. That was the piece of information that I did not > know that I needed. As soon as I changed the setting in the > /etc/xen/scripts/block script from > > do_or_die losetup -y "$loopdev" "$file" > > do_or_die losetup "$loopdev" "$file" > > this made all the difference in the world. > > However, this does beg another question. What is the best practice for > setting up a virtual disk for a guest domain? > > Is it better to live with the possibility of data corruption on a host crash > and have normal disk access, or turn the flag on and be protected against > corruption, but suffer terrible performance. Or is there a better way?Don''t use ''file:'' at all - loopback devices are just not a viable storage backend due to the horrific performance vs data integrity tradeoff you have to make. A physical disk device is the best option, but failing that use blktap for file based disks. eg use the tap:aio:/var/lib/xen/images/foo.img URL scheme for the disk. NB, if you want good performance with file based images, regardless of whether using blktap or loopback one should pre-allocate the entire disk image. Grow-on-demand images are always going to be slow because every time the host has to grow the image, it hits the journal on the host filesystem. Dan. -- |=- Red Hat, Engineering, Emerging Technologies, Boston. +1 978 392 2496 -=| |=- Perl modules: http://search.cpan.org/~danberr/ -=| |=- Projects: http://freshmeat.net/~danielpb/ -=| |=- GnuPG: 7D3B9505 F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 -=| _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users