Hello, we have a fresh Ubuntu 2.6.32-22-server install on Xen 3.4.0 and we observed really slow disk-write performance. Disk writes are so slow that the system is almost unusable. We ran a the classic dd benchmark a few times and the results are really discouraging: rd@ubuntu:~$ time sh -c "dd if=/dev/zero of=bigfile bs=8k count=1000000 && sync" 1000000+0 records in 1000000+0 records out 8192000000 bytes (8.2 GB) copied, 2082.15 s, 3.9 MB/s root@ubuntu:/home/rd# time sh -c "dd if=/dev/zero of=bigfile bs=8k count=1000000 && sync" 1000000+0 records in 1000000+0 records out 8192000000 bytes (8.2 GB) copied, 2352.29 s, 3.5 MB/s root@ubuntu:/home/rd# time sh -c "dd if=/dev/zero of=bigfile bs=8k count=1000000 && sync" 1000000+0 records in 1000000+0 records out 8192000000 bytes (8.2 GB) copied, 2544.19 s, 3.2 MB/s Then, we ran a postgresql benchmark - pgbench - that submits a few transactions against a small database. Simulating 15 clients running 200 transactions each we obtained this result, in transaction per second: tps: 1462.109434 (excluding connections establishing) where a non-virtualized workstation result is something like: tps: 5.833331 (excluding connections establishing) Running the dd benchmark against the raw disk we obtain good results (120MB/s) and a couple of Windows 2003 server guests hosted by the same Xen host seem to run fine (but I have no benchmark fo this). The disk is a HP ProLiant ML350 G6 Server series, Ubuntu is fully virtualized, 4 cpu, 4 GB RAM. Do you have any hint? Thanks _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
rudi wrote:> Then, we ran a postgresql benchmark - pgbench - that submits a few > transactions against a small database. Simulating 15 clients running 200 > transactions each we obtained this result, in transaction per second: > > tps: 1462.109434 (excluding connections establishing) > > where a non-virtualized workstation result is something like: > > tps: 5.833331 (excluding connections establishing)Of course what i meant was: 5.8 tps virtualized, 1462 tps non-virtualized. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Why not use paravirtualise? On 24 May 2010 17:20, rudi <rudolone@gmail.com> wrote:> Hello, > > we have a fresh Ubuntu 2.6.32-22-server install on Xen 3.4.0 and we > observed really slow disk-write performance. Disk writes are so slow that > the system is almost unusable. > We ran a the classic dd benchmark a few times and the results are really > discouraging: > > rd@ubuntu:~$ time sh -c "dd if=/dev/zero of=bigfile bs=8k count=1000000 && > sync" > 1000000+0 records in > 1000000+0 records out > 8192000000 bytes (8.2 GB) copied, 2082.15 s, 3.9 MB/s > > root@ubuntu:/home/rd# time sh -c "dd if=/dev/zero of=bigfile bs=8k > count=1000000 && sync" > 1000000+0 records in > 1000000+0 records out > 8192000000 bytes (8.2 GB) copied, 2352.29 s, 3.5 MB/s > > root@ubuntu:/home/rd# time sh -c "dd if=/dev/zero of=bigfile bs=8k > count=1000000 && sync" > 1000000+0 records in > 1000000+0 records out > 8192000000 bytes (8.2 GB) copied, 2544.19 s, 3.2 MB/s > > Then, we ran a postgresql benchmark - pgbench - that submits a few > transactions against a small database. Simulating 15 clients running 200 > transactions each we obtained this result, in transaction per second: > > tps: 1462.109434 (excluding connections establishing) > > where a non-virtualized workstation result is something like: > > tps: 5.833331 (excluding connections establishing) > > Running the dd benchmark against the raw disk we obtain good results > (120MB/s) and a couple of Windows 2003 server guests hosted by the same Xen > host seem to run fine (but I have no benchmark fo this). > > The disk is a HP ProLiant ML350 G6 Server series, > Ubuntu is fully virtualized, 4 cpu, 4 GB RAM. > > Do you have any hint? > > Thanks > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Boris Derzhavets
2010-May-24 17:01 UTC
[Xen-users] Attempt to set up Libvirt 0.7.7 & Xen 4.0 Dom0 (2.6.32.13) on top of F13 RC2
The most recent yum update has been run.
Installed Xen 4.0 and Libvirt like on F12. Via my experience xend will hang
for 5-7 min at startup and xen managed bridging will fail as well. This behavior
stays the same under either 2.6.32.13 or 2.6.33.3 pvops kernels. At the same
time PV guests might be unable to obtain IP via interface virbr0. Ubuntu Lucid
official xenified installer fails with DHCP request been sent to virbr0 under
2.6.32.13 kernel.
I''ve checked NAT bridge detection at Libvirt 0.7.1 & Xen 4.0
(2.6.32.13) on top of F12 and at Libvirt 0.7.0 & Xen 4.0 (2.6.32.13) on top
of Ubuntu 10.04 Server. Looks like only RH''s PV guests happily detect
virbr0 during install and runtime ( coming out of the LAN to the Net). Ubuntu
10.04 xenified PV installer and any OSOL (<= 134) requires old style network
bridge to Dom0.
In particular on F13 , i have to disable all xen managed network bridging except
(network-script /bin/true)
# chkconfig NetworkManager off
# chkconfig network on
Manually create a bridge
# service network start
Boris.
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
David Markey wrote:> Why not use paravirtualise?AFAIK paravirtualising Ubuntu is troublesome and requires a little work a the moment (a kernel patch and so on), so we did a quick test para-virtualising the standard "Oracle Enterprise Linux 5 64-bit" provided by Oracle, on top of Oracle-vm and running the same benchmarks. Results were sligthly better but still far from our expectations. Disk writes around 10-15 MB/s. My doubt is: should we expect much better results (and by consequence, there is a big misconfiguration/mistake somewhere in our stack) or these results are typical and unexceptional for this kind of systems? _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> -----Original Message----- > From: xen-users-bounces@lists.xensource.com [mailto:xen-users- > bounces@lists.xensource.com] On Behalf Of rudi > Sent: Monday, May 24, 2010 1:27 PM > To: xen-users@lists.xensource.com > Subject: Re: [Xen-users] Really slow disk write on Ubuntu > > Results were sligthly better but still far from our expectations. Disk > writes around 10-15 MB/s.As a data point, I logged into one of our virtual hosts and tried: $ dd if=/dev/zero of=/var/tmp/junk oflag=direct bs=64k count=1000 1000+0 records in 1000+0 records out 65536000 bytes (66 MB) copied, 0.971217 seconds, 67.5 MB/s $ dd if=/var/tmp/junk iflag=direct bs=64k of=/dev/null 1000+0 records in 1000+0 records out 65536000 bytes (66 MB) copied, 0.862861 seconds, 76.0 MB/s The disk array is currently shared by 15 domU''s running across 4 dom0''s. Not as fast as dedicated 15k RPM drives would be, but not bad either. Some of my physical hosts are faster, some are slower.> My doubt is: should we expect much better results (and by consequence, > there is a big misconfiguration/mistake somewhere in our stack) orthese> results are typical and unexceptional for this kind of systems?Are you using files or direct partitions for the guest image? What backend driver? -Jeff _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
I have a lot of Ubuntu 10.04 as paravirtual machines on top of the Xen without any modifications, no patches needed. On 24 May 2010 14:26, rudi <rudolone@gmail.com> wrote:> David Markey wrote: > >> Why not use paravirtualise? >> > > AFAIK paravirtualising Ubuntu is troublesome and requires a little work a > the moment (a kernel patch and so on), so we did a quick test > para-virtualising the standard "Oracle Enterprise Linux 5 64-bit" provided > by Oracle, on top of Oracle-vm and running the same benchmarks. > > Results were sligthly better but still far from our expectations. Disk > writes around 10-15 MB/s. > > My doubt is: should we expect much better results (and by consequence, > there is a big misconfiguration/mistake somewhere in our stack) or these > results are typical and unexceptional for this kind of systems? > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, May 24, 2010 at 07:26:30PM +0200, rudi wrote:> David Markey wrote: >> Why not use paravirtualise? > > AFAIK paravirtualising Ubuntu is troublesome and requires a little work > a the moment (a kernel patch and so on) >No it doesn''t require patches. Ubuntu 10.04 ships with Xen PV domU capable kernel, just like earlier versions of Ubuntu. I installed Ubuntu 10.04 as a Xen PV guest last week, using the standard kernel provided by the distro.. you can also boot the ubuntu installer as a Xen PV guest so it''s pretty easy to do the install aswell. -- Pasi , so we did a quick test> para-virtualising the standard "Oracle Enterprise Linux 5 64-bit" > provided by Oracle, on top of Oracle-vm and running the same benchmarks. > > Results were sligthly better but still far from our expectations. Disk > writes around 10-15 MB/s. > > My doubt is: should we expect much better results (and by consequence, > there is a big misconfiguration/mistake somewhere in our stack) or these > results are typical and unexceptional for this kind of systems? > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Fajar A. Nugraha
2010-May-25 04:33 UTC
Re: [Xen-users] Attempt to set up Libvirt 0.7.7 & Xen 4.0 Dom0 (2.6.32.13) on top of F13 RC2
On Tue, May 25, 2010 at 12:01 AM, Boris Derzhavets <bderzhavets@yahoo.com> wrote:> > The most recent yum update has been run. > Installed Xen 4.0 and Libvirt like on F12. Via my experience xend will hang for 5-7 min at startup and xen managed bridging will fail as well. This behavior stays the same under either 2.6.32.13 or 2.6.33.3 pvops kernels. At the same time PV guests might be unable to obtain IP via interface virbr0. Ubuntu Lucid official xenified installer fails with DHCP request been sent to virbr0 under 2.6.32.13 kernel. > I''ve checked NAT bridge detection at Libvirt 0.7.1 & Xen 4.0 (2.6.32.13) on top of F12 and at Libvirt 0.7.0 & Xen 4.0 (2.6.32.13) on top of Ubuntu 10.04 Server. Looks like only RH''s PV guests happily detect virbr0 during install and runtime ( coming out of the LAN to the Net). Ubuntu 10.04 xenified PV installer and any OSOL (<= 134) requires old style network bridge to Dom0. > > In particular on F13 , i have to disable all xen managed network bridging except > (network-script /bin/true) > > # chkconfig NetworkManager off > # chkconfig network on > Manually create a bridge > # service network start... so, in short, you wouldn''t recommend this combination for anything other than testing? -- Fajar _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Boris Derzhavets
2010-May-25 04:56 UTC
Re: [Xen-users] Attempt to set up Libvirt 0.7.7 & Xen 4.0 Dom0 (2.6.32.13) on top of F13 RC2
Not at all, i do recommend :- http://bderzhavets.wordpress.com/2010/05/18/set-up-libvirt-0-7-7xen-4-0-on-top-fedora-13-rc2/ Boris. P.S. I remember very well Solaris Nevada ( b70/b80) testing as PV DomU at native Xen 3.1 F8 Dom0 with xenified kernel ( not sure 2.6.21 or 2.6.24 ) . NAT worked for SNV guest just fine. I might be doing something not quite properly ( at Dom0) in meantime. --- On Tue, 5/25/10, Fajar A. Nugraha <fajar@fajar.net> wrote: From: Fajar A. Nugraha <fajar@fajar.net> Subject: Re: [Xen-users] Attempt to set up Libvirt 0.7.7 & Xen 4.0 Dom0 (2.6.32.13) on top of F13 RC2 To: "Boris Derzhavets" <bderzhavets@yahoo.com> Cc: xen-users@lists.xensource.com Date: Tuesday, May 25, 2010, 12:33 AM On Tue, May 25, 2010 at 12:01 AM, Boris Derzhavets <bderzhavets@yahoo.com> wrote:> > The most recent yum update has been run. > Installed Xen 4.0 and Libvirt like on F12. Via my experience xend will hang for 5-7 min at startup and xen managed bridging will fail as well. This behavior stays the same under either 2.6.32.13 or 2.6.33.3 pvops kernels. At the same time PV guests might be unable to obtain IP via interface virbr0. Ubuntu Lucid official xenified installer fails with DHCP request been sent to virbr0 under 2.6.32.13 kernel. > I''ve checked NAT bridge detection at Libvirt 0.7.1 & Xen 4.0 (2.6.32.13) on top of F12 and at Libvirt 0.7.0 & Xen 4.0 (2.6.32.13) on top of Ubuntu 10.04 Server. Looks like only RH''s PV guests happily detect virbr0 during install and runtime ( coming out of the LAN to the Net). Ubuntu 10.04 xenified PV installer and any OSOL (<= 134) requires old style network bridge to Dom0. > > In particular on F13 , i have to disable all xen managed network bridging except > (network-script /bin/true) > > # chkconfig NetworkManager off > # chkconfig network on > Manually create a bridge > # service network start... so, in short, you wouldn''t recommend this combination for anything other than testing? -- Fajar _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Jeff Sturm wrote:>> My doubt is: should we expect much better results >> (and by consequence, there is a big misconfiguration/mistake >> somewhere in our stack) or these results are typical and >> unexceptional for this kind of systems? > Are you using files or direct partitions for the guest image? What > backend driver?We are using files. What do you mean for backend driver? BTW, we repeated those benchmarks with different installations, and it seems that disk writes are really slow on OracleVM only. SUSE Entreprise+Xen, as well as VmWare+Ubuntu have reasonable performances. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> -----Original Message----- > From: xen-users-bounces@lists.xensource.com [mailto:xen-users- > bounces@lists.xensource.com] On Behalf Of rudi > Sent: Friday, May 28, 2010 8:39 AM > To: xen-users@lists.xensource.com > Subject: Re: [Xen-users] Really slow disk write on Ubuntu > > We are using files. What do you mean for backend driver?Whether you specify the virtual disk with phy:, file:, tap:aio:, etc. Couldn''t remember what those are called offhand. For our Xen instances we generally use phy: and stay away from file:. -Jeff _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users