One of my customers handed me a copy of this paper and asked why are we using xen if it is so slow... http://morse.colorado.edu/~tlen5710/12s/VMware.pdf which is a fairly damning report of performance under Xen in comparison to VMWare, citing worse than 50% overhead of Xen in comparison to physical. Has anyone seen this paper before? James
On Wed, May 23, 2012 at 8:03 AM, James Harper <james.harper@bendigoit.com.au> wrote:> One of my customers handed me a copy of this paper and asked why are we using xen if it is so slow... > > http://morse.colorado.edu/~tlen5710/12s/VMware.pdf > > which is a fairly damning report of performance under Xen in comparison to VMWare, citing worse than 50% overhead of Xen in comparison to physical. > > Has anyone seen this paper before?IIRC vmware doesn''t allow people to publish benchmark results like that. Anyway, taking a quick glance, the benchmark was using SQL 2005 (which implies windows), but searching for the words "pv", "para", or "driver" returns no result. So it''s only normal that xen performs terrible on that benchmark. -- Fajar
Hadn''t seen it, but the fact that they used 32 bit Windows Server 2003 means that I think it would be - to say the least - disingenuous to extrapolate those findings to Xen in general. Running windows means that it was running HVM and that''s a killer for IO based testing. In other words, if you - like the vast majority of Xen users - are running PV guests, this is 100% irrelevant. On Tue, May 22, 2012 at 9:03 PM, James Harper <james.harper@bendigoit.com.au> wrote:> One of my customers handed me a copy of this paper and asked why are we > using xen if it is so slow... > > http://morse.colorado.edu/~tlen5710/12s/VMware.pdf > > which is a fairly damning report of performance under Xen in comparison to > VMWare, citing worse than 50% overhead of Xen in comparison to physical. > > Has anyone seen this paper before? > > James > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xen.org > http://lists.xen.org/xen-users >_______________________________________________ Xen-users mailing list Xen-users@lists.xen.org http://lists.xen.org/xen-users
> -----Original Message----- > From: xen-users-bounces@lists.xen.org [mailto:xen-users-bounces@lists.xen.org] On > Behalf Of Fajar A. Nugraha > Sent: Tuesday, May 22, 2012 9:14 PM > > > http://morse.colorado.edu/~tlen5710/12s/VMware.pdf > > Anyway, taking a quick glance, the benchmark was using SQL 2005 (which implies > windows), but searching for the words "pv", "para", or "driver" returns no result. So > it''s only normal that xen performs terrible on that benchmark.100% agreed. You can''t expect that paper to be objective given the source, can you? Anyway, we have some fairly large Xen deployments with multiple, clustered dom0 hosts and hundreds of Linux PV guests. I can tell you confidently that the I/O bottleneck is our SAN implementation, not the network, not blkback/blkfront, or anything else in Xen. I''ve gotten throughput of 200MB/s sequential from PV guests, which is the limit of our network, but unfortunately random I/O is the norm for us so we never really achieve throughput anywhere near our theoretical limit. My advice: If you want performance and you''re willing to do it the "Xen way" with Linux PV guests (or maybe Windows HVM and suitable drivers), Xen will do fine. If on the other hand you want something that will work "just like VMWare" and support Windows natively with no special drivers, then yeah, the VMWare paper is probably really relevant to you. (And even then I thought VMWare needed virtual drivers to perform optimally. At least on Linux.) -Jeff
I''m guessing that the benchmarks were run without any optimized drivers for Windows under Xen. That would be the only way to explain the dismal performance. Windows is known to perform poorly under Xen if you simply install Windows and don''t install optimized drives. In all fairness, Windows performs poorly on native hardware if you don''t install optimized drivers, so testers should expect to have to install optimized drivers when testing Windows regardless of the environment. I have emailed them about this, and I will post anything I hear back from them. -----Original Message----- From: xen-users-bounces@lists.xen.org [mailto:xen-users-bounces@lists.xen.org] On Behalf Of James Harper Sent: Tuesday, May 22, 2012 8:03 PM To: xen-users@lists.xen.org Subject: [Xen-users] Xen vs VMWare comparison paper One of my customers handed me a copy of this paper and asked why are we using xen if it is so slow... http://morse.colorado.edu/~tlen5710/12s/VMware.pdf which is a fairly damning report of performance under Xen in comparison to VMWare, citing worse than 50% overhead of Xen in comparison to physical. Has anyone seen this paper before? James _______________________________________________ Xen-users mailing list Xen-users@lists.xen.org http://lists.xen.org/xen-users
On 23 May 2012 19:11, <admin@xenhive.com> wrote:> I''m guessing that the benchmarks were run without any optimized drivers for > Windows under Xen. That would be the only way to explain the dismal > performance. Windows is known to perform poorly under Xen if you simply > install Windows and don''t install optimized drives. In all fairness, > Windows performs poorly on native hardware if you don''t install optimized > drivers, so testers should expect to have to install optimized drivers when > testing Windows regardless of the environment. I have emailed them about > this, and I will post anything I hear back from them. >100% agree. And also, that most of Xen users run PV guests, so this is usually irrelevant. But, what about this site that it''s considered an important reference for students here: http://www.phoronix.com/scan.php?page=article&item=ubuntu_1204_virt&num=1 I don''t see the point of running an HVM Linux.. am I missing something here?
> > But, what about this site that it''s considered an important reference > for students here: > > http://www.phoronix.com/scan.php?page=article&item=ubuntu_1204_virt&num=1 > > I don''t see the point of running an HVM Linux.. am I missing something here?Yes. HVM Linux for most workloads is actualy faster than PV one. I''ve done benchmarks a while ago on a decent hardware (Xeon 56xx, EPT, lots of optimizaton for VMENTRY / VMEXIT, etc). Look for what VMware has been doing from a long time - it has switched every x86_64/AMD64 guest to hardware virtualization (there''s a paper about that that shows what they use for which kind of guest). PV still has its place in some specific workloads, but the tendency is now reversed.
eva wrote:>I don''t see the point of running an HVM Linux.. am I missing something here?It depends on what level of control/customisation you have over the guest. For most cases, there''s no point I can see. However, there will be cases where "appliances" are being employed where it may be required. The sort of situation I can think of might be where you buy am appliance which comes as a complete system - "insert [ CD | USB Stick ], switch on, wait while it installs". In that situation, the vendor may simply only support unmodified bare-metal installs. Running it as a PV guest will be right out, running it as an HV guest you will probably get away with - but not necessarily. Slightly different these days now that standard kernels tend to have the Xen stuff - so you could probably manage to run the appliance as a PV guest without having to use a different kernel. As an aside, at work we have some customers using a particular package (on Windows). This isn''t supplied as a disk, you have to call up the vendors, give them remote access to your server, and they will install and configure it. They will not work with a virtual machine - not even using Microsoft''s Hyper-V. I think for one customers, we let them install it, then imaged the machine into a VM. -- Simon Hobson Visit http://www.magpiesnestpublishing.co.uk/ for books by acclaimed author Gladys Hobson. Novels - poetry - short stories - ideal as Christmas stocking fillers. Some available as e-books.
Hello Admin, On Wed, May 23, 2012 at 1:11 PM, <admin@xenhive.com> wrote:> I''m guessing that the benchmarks were run without any optimized drivers for > Windows under Xen. That would be the only way to explain the dismal > performance. Windows is known to perform poorly under Xen if you simply > install Windows and don''t install optimized drives. In all fairness, > Windows performs poorly on native hardware if you don''t install optimized > drivers, so testers should expect to have to install optimized drivers when > testing Windows regardless of the environment. I have emailed them about > this, and I will post anything I hear back from them.I''m rather curious here because you said "optimized drivers" rather than saying "GPLPV." Is there a choice of [better than stock] drivers for a Windows HVM DomU that I''m not aware of? I ask because GPLPV doesn''t quite work for me yet in all situations. Regards, Andrew Bobulsky> -----Original Message----- > From: xen-users-bounces@lists.xen.org > [mailto:xen-users-bounces@lists.xen.org] On Behalf Of James Harper > Sent: Tuesday, May 22, 2012 8:03 PM > To: xen-users@lists.xen.org > Subject: [Xen-users] Xen vs VMWare comparison paper > > One of my customers handed me a copy of this paper and asked why are we > using xen if it is so slow... > > http://morse.colorado.edu/~tlen5710/12s/VMware.pdf > > which is a fairly damning report of performance under Xen in comparison to > VMWare, citing worse than 50% overhead of Xen in comparison to physical. > > Has anyone seen this paper before? > > James > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xen.org > http://lists.xen.org/xen-users > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xen.org > http://lists.xen.org/xen-users
GPLPV is an optimized driver. -----Original Message----- From: xen-users-bounces@lists.xen.org [mailto:xen-users-bounces@lists.xen.org] On Behalf Of Andrew Bobulsky Sent: Thursday, May 24, 2012 11:53 AM To: admin@xenhive.com Cc: xen-users@lists.xen.org Subject: Re: [Xen-users] Xen vs VMWare comparison paper Hello Admin, On Wed, May 23, 2012 at 1:11 PM, <admin@xenhive.com> wrote:> I''m guessing that the benchmarks were run without any optimized driversfor> Windows under Xen. That would be the only way to explain the dismal > performance. Windows is known to perform poorly under Xen if you simply > install Windows and don''t install optimized drives. In all fairness, > Windows performs poorly on native hardware if you don''t install optimized > drivers, so testers should expect to have to install optimized driverswhen> testing Windows regardless of the environment. I have emailed them about > this, and I will post anything I hear back from them.I''m rather curious here because you said "optimized drivers" rather than saying "GPLPV." Is there a choice of [better than stock] drivers for a Windows HVM DomU that I''m not aware of? I ask because GPLPV doesn''t quite work for me yet in all situations. Regards, Andrew Bobulsky> -----Original Message----- > From: xen-users-bounces@lists.xen.org > [mailto:xen-users-bounces@lists.xen.org] On Behalf Of James Harper > Sent: Tuesday, May 22, 2012 8:03 PM > To: xen-users@lists.xen.org > Subject: [Xen-users] Xen vs VMWare comparison paper > > One of my customers handed me a copy of this paper and asked why are we > using xen if it is so slow... > > http://morse.colorado.edu/~tlen5710/12s/VMware.pdf > > which is a fairly damning report of performance under Xen in comparison to > VMWare, citing worse than 50% overhead of Xen in comparison to physical. > > Has anyone seen this paper before? > > James > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xen.org > http://lists.xen.org/xen-users > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xen.org > http://lists.xen.org/xen-users_______________________________________________ Xen-users mailing list Xen-users@lists.xen.org http://lists.xen.org/xen-users
> Hello Admin, > > On Wed, May 23, 2012 at 1:11 PM, <admin@xenhive.com> wrote: > > I''m guessing that the benchmarks were run without any optimized > > drivers for Windows under Xen. That would be the only way to explain > > the dismal performance. Windows is known to perform poorly under Xen > > if you simply install Windows and don''t install optimized drives. In > > all fairness, Windows performs poorly on native hardware if you don''t > > install optimized drivers, so testers should expect to have to install > > optimized drivers when testing Windows regardless of the environment. > > I have emailed them about this, and I will post anything I hear back from > them. > > I''m rather curious here because you said "optimized drivers" rather than > saying "GPLPV." Is there a choice of [better than stock] drivers for a > Windows HVM DomU that I''m not aware of? I ask because GPLPV doesn''t > quite work for me yet in all situations. >If they were running the Citrix branded Xen product then Citrix''s own drivers would be the obvious choice. In what way is GPLPV letting you down? You asked about USB a while back... I''ve made some progress there but it''s still very much a work in progress. Thanks James
Hello James, On Thu, May 24, 2012 at 6:25 PM, James Harper <james.harper@bendigoit.com.au> wrote:>> Hello Admin, >> >> On Wed, May 23, 2012 at 1:11 PM, <admin@xenhive.com> wrote: >> > I''m guessing that the benchmarks were run without any optimized >> > drivers for Windows under Xen. That would be the only way to explain >> > the dismal performance. Windows is known to perform poorly under Xen >> > if you simply install Windows and don''t install optimized drives. In >> > all fairness, Windows performs poorly on native hardware if you don''t >> > install optimized drivers, so testers should expect to have to install >> > optimized drivers when testing Windows regardless of the environment. >> > I have emailed them about this, and I will post anything I hear back from >> them. >> >> I''m rather curious here because you said "optimized drivers" rather than >> saying "GPLPV." Is there a choice of [better than stock] drivers for a >> Windows HVM DomU that I''m not aware of? I ask because GPLPV doesn''t >> quite work for me yet in all situations. >> > > If they were running the Citrix branded Xen product then Citrix''s own drivers would be the obvious choice. > > In what way is GPLPV letting you down? You asked about USB a while back... I''ve made some progress there but it''s still very much a work in progress.Heh, I remember I tried the Citrix drivers on Xen a while back... that didn''t go well :D While GPLPV seemed to work pretty well with my Xen guests, it invariably caused a BSOD on atimkpag.sys (I think that''s the spelling) when I passed my Radeon 5850 through with IOMMU. I really wish I knew why, especially because it seems to work very well for Casey, albeit on an Intel setup with a Radeon 6XXX model card. If you''d like, I can offer my time and/or access to my hardware to help debug the issue, but would definitely need some direction on exactly what to do :P Cheers, Andrew Bobulsky