We have some fairly robust hardware we are trying to set up a number of para-virtual servers in a clustered environment and we are seeing huge cpu issues. Specs: Dell 1950: 8 1u servers cpu MHz : 2660.038 dual quad cores 32G memory each Our first virual server was configured to use 8G and 4 cores. I tried running a quick test by compiling our application, simply running ''time make.'' The results were terrible: real 0m56.79s user 0m26.53s sys 0m35.46s In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem): real 0m42.99s user 0m29.85s sys 0m6.47s We then reconfigured the virtual server to use a single core and the results were almost identical to the 4 core test. Any insights as to what our problem might be would be greatly appreciated. -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Grant McWilliams
2008-Sep-25 15:34 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 8:22 AM, Geoffrey <lists@serioustechnology.com>wrote:> We have some fairly robust hardware we are trying to set up a number of > para-virtual servers in a clustered environment and we are seeing huge cpu > issues. Specs: > > Dell 1950: 8 1u servers > cpu MHz : 2660.038 > dual quad cores > 32G memory each > > Our first virual server was configured to use 8G and 4 cores. > > I tried running a quick test by compiling our application, simply running > ''time make.'' The results were terrible: > > real 0m56.79s > user 0m26.53s > sys 0m35.46s > > In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem): > > real 0m42.99s > user 0m29.85s > sys 0m6.47s > > We then reconfigured the virtual server to use a single core and the > results were almost identical to the 4 core test. > > Any insights as to what our problem might be would be greatly appreciated. > > -- > Until later, Geoffrey > > Those who would give up essential Liberty, to purchase a little > temporary Safety, deserve neither Liberty nor Safety. > - Benjamin Franklin > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >Wouldn''t you have to do make -j 4 to use 4 cores? It looks like on the laptop you''re using one 2.5 Ghz core and on the Xen VM you''re using one 2.66 Ghz core. -- Grant McWilliams Some people, when confronted with a problem, think "I know, I''ll use Windows." Now they have two problems. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Todd Deshane
2008-Sep-25 15:50 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 11:22 AM, Geoffrey <lists@serioustechnology.com> wrote:> We have some fairly robust hardware we are trying to set up a number of > para-virtual servers in a clustered environment and we are seeing huge cpu > issues. Specs: > > Dell 1950: 8 1u servers > cpu MHz : 2660.038 > dual quad cores > 32G memory each > > Our first virual server was configured to use 8G and 4 cores. > > I tried running a quick test by compiling our application, simply running > ''time make.'' The results were terrible: > > real 0m56.79s > user 0m26.53s > sys 0m35.46s > > In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem): > > real 0m42.99s > user 0m29.85s > sys 0m6.47s > > We then reconfigured the virtual server to use a single core and the results > were almost identical to the 4 core test. > > Any insights as to what our problem might be would be greatly appreciated. >Try using kernbench for testing. Also, what version of xen, guest kernel etc.? Those can really make a big difference. Cheers, Todd -- Todd Deshane http://todddeshane.net check out our book: http://runningxen.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Grant McWilliams wrote:> Wouldn''t you have to do make -j 4 to use 4 cores? It looks like on the > laptop you''re using one 2.5 Ghz core and on the Xen VM you''re using one 2.66 > Ghz core.Thank you, that has made a huge difference. Still, I''m concerned with the results now: laptop (make -j 3) (2 cores) real 0m16.70s user 0m26.82s sys 0m5.50s virtual server with 4 cores (make -j 5) (4 cores) real 0m22.36s user 0m26.24s sys 0m54.97s Is this a reasonable expectation with virtualization? -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Todd Deshane
2008-Sep-25 16:04 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey <lists@serioustechnology.com> wrote:> Grant McWilliams wrote: > >> Wouldn''t you have to do make -j 4 to use 4 cores? It looks like on the >> laptop you''re using one 2.5 Ghz core and on the Xen VM you''re using one >> 2.66 >> Ghz core. > > Thank you, that has made a huge difference. Still, I''m concerned with the > results now: > > laptop (make -j 3) (2 cores) > > real 0m16.70s > user 0m26.82s > sys 0m5.50s > > virtual server with 4 cores (make -j 5) (4 cores) > > real 0m22.36s > user 0m26.24s > sys 0m54.97s > > > Is this a reasonable expectation with virtualization? >This doesn''t seem quite right to me, try kernbench and also make sure the versions of xen and guest kernels are the same on the server and laptop for a good comparision. The overhead of Xen PV should be pretty low vs native. Todd> -- > Until later, Geoffrey > > Those who would give up essential Liberty, to purchase a little > temporary Safety, deserve neither Liberty nor Safety. > - Benjamin Franklin > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >-- Todd Deshane http://todddeshane.net check out our book: http://runningxen.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Todd Deshane wrote:> On Thu, Sep 25, 2008 at 11:22 AM, Geoffrey <lists@serioustechnology.com> wrote: >> We have some fairly robust hardware we are trying to set up a number of >> para-virtual servers in a clustered environment and we are seeing huge cpu >> issues. Specs: >> >> Dell 1950: 8 1u servers >> cpu MHz : 2660.038 >> dual quad cores >> 32G memory each >> >> Our first virual server was configured to use 8G and 4 cores. >> >> I tried running a quick test by compiling our application, simply running >> ''time make.'' The results were terrible: >> >> real 0m56.79s >> user 0m26.53s >> sys 0m35.46s >> >> In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem): >> >> real 0m42.99s >> user 0m29.85s >> sys 0m6.47s >> >> We then reconfigured the virtual server to use a single core and the results >> were almost identical to the 4 core test. >> >> Any insights as to what our problem might be would be greatly appreciated. >> > > Try using kernbench for testing.I''ll do that and post my results. Pointers as to where I can get it?> Also, what version of xen, guest kernel etc.?Red Hat advance server 5.2 kernel-xen-devel-2.6.18-92.el5 xen-3.0.3-64.el5_2.1 -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Todd Deshane wrote:> On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey <lists@serioustechnology.com> wrote:>> Is this a reasonable expectation with virtualization? >> > > This doesn''t seem quite right to me, try kernbench and > also make sure the versions of xen and guest kernels > are the same on the server and laptop for a good > comparision.I''m not running xen on the laptop. Laptop is RHEL 5.2, kernel: 2.6.18-92.1.10.el5> The overhead of Xen PV should be pretty low vs native.I was wrong when I said we were para-virtualizing, this is full virtualization. -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ross S. W. Walker
2008-Sep-25 16:24 UTC
RE: [Xen-users] major slow down with xen implementation
Geoffrey wrote:> Todd Deshane wrote: > > On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey <lists@serioustechnology.com> wrote: > >> Is this a reasonable expectation with virtualization? > > > > This doesn''t seem quite right to me, try kernbench and > > also make sure the versions of xen and guest kernels > > are the same on the server and laptop for a good > > comparision. > > I''m not running xen on the laptop. Laptop is RHEL 5.2, > kernel: 2.6.18-92.1.10.el5 > > > The overhead of Xen PV should be pretty low vs native. > > I was wrong when I said we were para-virtualizing, this is full > virtualization.Well, why not put up the xen config for the domU and see if anybody can suggest some tweaks, but if you are using RH + Xen it would be silly NOT to para-virtualize it. -Ross ______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ross S. W. Walker wrote:> Geoffrey wrote: >> Todd Deshane wrote: >>> On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey <lists@serioustechnology.com> wrote: >>>> Is this a reasonable expectation with virtualization? >>> This doesn''t seem quite right to me, try kernbench and >>> also make sure the versions of xen and guest kernels >>> are the same on the server and laptop for a good >>> comparision. >> I''m not running xen on the laptop. Laptop is RHEL 5.2, >> kernel: 2.6.18-92.1.10.el5 >> >>> The overhead of Xen PV should be pretty low vs native. >> I was wrong when I said we were para-virtualizing, this is full >> virtualization. > > Well, why not put up the xen config for the domU and see if > anybody can suggest some tweaks, but if you are using RH + Xen > it would be silly NOT to para-virtualize it.Can''t para-virtualize. Running 64bit on the hardware so as to get access to the full 32GB memory. Running 32bit virtuals, because we have a third party app that won''t run on 64bit. I know... -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Javier Guerra
2008-Sep-25 17:16 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 11:55 AM, Geoffrey <lists@serioustechnology.com> wrote:> Can''t para-virtualize. Running 64bit on the hardware so as to get access to > the full 32GB memory. Running 32bit virtuals, because we have a third party > app that won''t run on 64bit. I know...it''s not easy to write a 32bit app that won''t run when the needed libraries are there; but i guess you''re already tired of analysing ldd output. if you''re stuck with HVM, then at least make sure that you''re using PV drivers. it really does a world of difference -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nathan Eisenberg
2008-Sep-25 17:23 UTC
RE: [Xen-users] major slow down with xen implementation
That seems like a pretty old version of XEN. Someone needs to give the RHAS repository guys a good poke. Not endorsing a specific distro here, but I know that the Ubuntu/Debian repos are quite up to date - if you''re up for trying a different Dom0 distro, I''d give that a shot on the same hardware and see what happens. Thank you, Nathan Eisenberg Sr. Systems Administrator Atlas Networks, LLC -----Original Message----- From: xen-users-bounces@lists.xensource.com [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of Geoffrey Sent: Thursday, September 25, 2008 9:08 AM To: xen-users@lists.xensource.com Subject: Re: [Xen-users] major slow down with xen implementation Todd Deshane wrote:> On Thu, Sep 25, 2008 at 11:22 AM, Geoffrey <lists@serioustechnology.com> wrote: >> We have some fairly robust hardware we are trying to set up a number of >> para-virtual servers in a clustered environment and we are seeing huge cpu >> issues. Specs: >> >> Dell 1950: 8 1u servers >> cpu MHz : 2660.038 >> dual quad cores >> 32G memory each >> >> Our first virual server was configured to use 8G and 4 cores. >> >> I tried running a quick test by compiling our application, simply running >> ''time make.'' The results were terrible: >> >> real 0m56.79s >> user 0m26.53s >> sys 0m35.46s >> >> In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem): >> >> real 0m42.99s >> user 0m29.85s >> sys 0m6.47s >> >> We then reconfigured the virtual server to use a single core and the results >> were almost identical to the 4 core test. >> >> Any insights as to what our problem might be would be greatly appreciated. >> > > Try using kernbench for testing.I''ll do that and post my results. Pointers as to where I can get it?> Also, what version of xen, guest kernel etc.?Red Hat advance server 5.2 kernel-xen-devel-2.6.18-92.el5 xen-3.0.3-64.el5_2.1 -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ 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
Grant McWilliams
2008-Sep-25 17:45 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 10:16 AM, Javier Guerra <javier@guerrag.com> wrote:> On Thu, Sep 25, 2008 at 11:55 AM, Geoffrey <lists@serioustechnology.com> > wrote: > > Can''t para-virtualize. Running 64bit on the hardware so as to get access > to > > the full 32GB memory. Running 32bit virtuals, because we have a third > party > > app that won''t run on 64bit. I know... > > it''s not easy to write a 32bit app that won''t run when the needed > libraries are there; but i guess you''re already tired of analysing ldd > output. > > -- > Javier > > >I''d assume that you''ve tried providing the relevant 32 bit libraries for the app but it''s dangerous to assume anything so I won''t. You can get most 32 bit apps to run on 64 bit if you have both 32 bit and 64 bit libs available. I think it''s worth the effort. But again you''ve probably already gone down this route. The overhead for HVM is quite a bit higher than PV. Also according to the Xensource wiki you can install a 32bit PAE PV on a 64bit DomU Hypervisor with Xen 3.1 and greater. *32bit Hypervisor* *64bit Hypervisor* *32bit PV DomU* Yes Yes (starting from Xen 3.1), kernel must be PAE *64bit PV DomU* No Yes *32bit HVM DomU* Yes Yes *64bit HVM DomU* No Yes http://wiki.xensource.com/xenwiki/XenFaq#head-5f7176b3909cb0382cece43a6a8fc25a3a114e93 It''s easy to install Xen 3.3 on CentOS 5 x86_64 if you use the rpms... Don''t know which distro you''re using. -- Grant McWilliams Some people, when confronted with a problem, think "I know, I''ll use Windows." Now they have two problems. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ross S. W. Walker
2008-Sep-25 18:14 UTC
RE: [Xen-users] major slow down with xen implementation
Grant McWilliams wrote:> > On Thu, Sep 25, 2008 at 10:16 AM, Javier Guerra <javier@guerrag.com> wrote: > > On Thu, Sep 25, 2008 at 11:55 AM, Geoffrey <lists@serioustechnology.com> wrote: > > Can''t para-virtualize. Running 64bit on the hardware so as to get access to > > the full 32GB memory. Running 32bit virtuals, because we have a third party > > app that won''t run on 64bit. I know... > > > it''s not easy to write a 32bit app that won''t run when the needed > libraries are there; but i guess you''re already tired of analysing ldd > output. > > I''d assume that you''ve tried providing the relevant 32 bit > libraries for the app but it''s dangerous to assume anything > so I won''t. You can get most > 32 bit apps to run on 64 bit if you have both 32 bit and 64 > bit libs available. I think it''s worth the effort. But again > you''ve probably already gone down > this route. The overhead for HVM is quite a bit higher than PV. > > Also according to the Xensource wiki you can install a 32bit > PAE PV on a 64bit DomU Hypervisor with Xen 3.1 and greater.Yes, and since RHEL Xen 3.0.3 is really the 3.1 hypervisor with the 3.0.3 userland utils (don''t ask) and since the RHEL kernels are now all PAE you should be able to run your 32bit PV domU on your 64bit dom0. Plus with pyGrub boot loading you don''t need to re-image to try it out, hell if your using LVM you don''t even need to take your existing domU down to test, just snapshot it. Install the Xen domU kernel in your HVM guest side-by-side with your native kernel, make an LVM snapshot, create a PV config for it and try to boot it (remember to console at boot so you can choose the Xen kernel). If it doesn''t work it will stop right after the Xen 3.1 banner, if it does then you can plan on switching to PV during the next maintenance period. NOTE: Remember to either not allocate or attach to a loop tap device the network interface of the test PV to prevent duplicate IP addresses. -Ross ______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Joseph L. Casale
2008-Sep-25 18:17 UTC
RE: [Xen-users] major slow down with xen implementation
>It's easy to install Xen 3.3 on CentOS 5 x86_64 if you use the rpms... Don't know which distro you're using.Was just about to post the same info you did… My experience with 3.3.0 has been rather seriously tainted. I ran extremely reliably for an extended period of time with 3.2.0/1 and was very happy. How have you found 3.3.0? Any Windows HVM's and passthrough for pv domu's? Thanks, jlc _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Grant McWilliams
2008-Sep-25 18:21 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 11:17 AM, Joseph L. Casale < JCasale@activenetwerx.com> wrote:> >It's easy to install Xen 3.3 on CentOS 5 x86_64 if you use the rpms... > Don't know which distro you're using. > > Was just about to post the same info you did… > My experience with 3.3.0 has been rather seriously tainted. I ran extremely > reliably for an extended period of time > with 3.2.0/1 and was very happy. How have you found 3.3.0? Any Windows > HVM's and passthrough for pv domu's? > > Thanks, > jlc > > _I've not had any problems as of yet but my needs are simpler. I'm only virtualizing Linux and the servers only need network access. -- Grant McWilliams Some people, when confronted with a problem, think "I know, I'll use Windows." Now they have two problems. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Javier Guerra wrote:> On Thu, Sep 25, 2008 at 11:55 AM, Geoffrey <lists@serioustechnology.com> wrote: >> Can''t para-virtualize. Running 64bit on the hardware so as to get access to >> the full 32GB memory. Running 32bit virtuals, because we have a third party >> app that won''t run on 64bit. I know... > > it''s not easy to write a 32bit app that won''t run when the needed > libraries are there; but i guess you''re already tired of analysing ldd > output.My thoughts exactly. In spite of our previous experiences, we are going to give it another shot.> if you''re stuck with HVM,then at least make sure that you''re using PV > drivers. it really does a world of differenceI''m not sure what you mean by PV drivers. -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ross S. W. Walker
2008-Sep-25 18:30 UTC
RE: [Xen-users] major slow down with xen implementation
Joseph L. Casale wrote:> > >It''s easy to install Xen 3.3 on CentOS 5 x86_64 if you use > the rpms... Don''t know which distro you''re using. > > Was just about to post the same info you did... > My experience with 3.3.0 has been rather seriously tainted. I > ran extremely reliably for an extended period of time > with 3.2.0/1 and was very happy. How have you found 3.3.0? > Any Windows HVM''s and passthrough for pv domu''s?I''m with you, I''ve bled against too many past Xen .0 releases to know it''s always better to go with the .1 or .2 releases. Try 3.2.2, use the 3.2.0 SRPM as the source to roll your own 3.2.2 RPM. -Ross ______________________________________________________________________ This e-mail, and any attachments thereto, is intended only for use by the addressee(s) named herein and may contain legally privileged and/or confidential information. If you are not the intended recipient of this e-mail, you are hereby notified that any dissemination, distribution or copying of this e-mail, and any attachments thereto, is strictly prohibited. If you have received this e-mail in error, please immediately notify the sender and permanently delete the original and any copy or printout thereof. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nathan Eisenberg wrote:> That seems like a pretty old version of XEN. Someone needs to give > the RHAS repository guys a good poke.Just our luck.> Not endorsing a specific distro here, but I know that the > Ubuntu/Debian repos are quite up to date - if you''re up for trying a > different Dom0 distro, I''d give that a shot on the same hardware and > see what happens.No can do, but thanks for the suggestion. -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Geoffrey wrote:> Ross S. W. Walker wrote: >> Geoffrey wrote: >>> Todd Deshane wrote: >>>> On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey >>>> <lists@serioustechnology.com> wrote: >>>>> Is this a reasonable expectation with virtualization? >>>> This doesn''t seem quite right to me, try kernbench and >>>> also make sure the versions of xen and guest kernels >>>> are the same on the server and laptop for a good >>>> comparision. >>> I''m not running xen on the laptop. Laptop is RHEL 5.2, >>> kernel: 2.6.18-92.1.10.el5 >>> >>>> The overhead of Xen PV should be pretty low vs native. >>> I was wrong when I said we were para-virtualizing, this is full >>> virtualization. >> >> Well, why not put up the xen config for the domU and see if >> anybody can suggest some tweaks, but if you are using RH + Xen >> it would be silly NOT to para-virtualize it. > > Can''t para-virtualize. Running 64bit on the hardware so as to get > access to the full 32GB memory. Running 32bit virtuals, because we > have a third party app that won''t run on 64bit. I know...So? Where is the problem? Check your Xen Caps [1] to confirm, that you can run 32bit para-virtualized domU. This is running without problems here: 64bit ubuntu dom0 and 32bit para domU ubuntu / debian. Greetz Age_M [1] xm info | grep xen_caps If the result shows xen-3.0-x86_32p is means you could run 32bit paravirtualized domUs _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Pasi Kärkkäinen
2008-Sep-26 07:40 UTC
Re: [Xen-users] major slow down with xen implementation
On Thu, Sep 25, 2008 at 12:55:20PM -0400, Geoffrey wrote:> Ross S. W. Walker wrote: > >Geoffrey wrote: > >>Todd Deshane wrote: > >>>On Thu, Sep 25, 2008 at 11:57 AM, Geoffrey <lists@serioustechnology.com> > >>>wrote: > >>>>Is this a reasonable expectation with virtualization? > >>>This doesn''t seem quite right to me, try kernbench and > >>>also make sure the versions of xen and guest kernels > >>>are the same on the server and laptop for a good > >>>comparision. > >>I''m not running xen on the laptop. Laptop is RHEL 5.2, > >>kernel: 2.6.18-92.1.10.el5 > >> > >>>The overhead of Xen PV should be pretty low vs native. > >>I was wrong when I said we were para-virtualizing, this is full > >>virtualization. > > > >Well, why not put up the xen config for the domU and see if > >anybody can suggest some tweaks, but if you are using RH + Xen > >it would be silly NOT to para-virtualize it. > > Can''t para-virtualize. Running 64bit on the hardware so as to get > access to the full 32GB memory. Running 32bit virtuals, because we have > a third party app that won''t run on 64bit. I know... >https://www.redhat.com/archives/rhelv5-announce/2008-May/msg00002.html "Red Hat Enterprise Linux 5.2 GA Announcement" Technology Previews ------------------- - 32-bit para-virtualized (PV) guests on 64-bit AMD64/Intel(r) 64 hosts So you could try it.. there were some problems with it in RHEL 5.1, and I assume those were fixed for 5.2. -- Pasi _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nemeth, Tamas
2008-Sep-26 08:32 UTC
Re: [Xen-users] major slow down with xen implementation
Hi all,
We have similar servers, and I use xen 3.2.1 (3.3.0 seems to be too
buggy). The hypervisor is 64bit, but all domains (including dom0) are
32bit PAE, because a 64bit dom0 - for example - can''t migrate a 32bit
domU, but a 32bit domU can migrate a 64bit domU, AFAIK. So, accordint to
my experiences, 32bit PAE domains (with the original xen.org kernels, if
you can afford it) running on a 64bit hypervisor seem to be the best
soulution to utilize a beast like your machine. Xen 3.2.1 seems to be
fast, and reliable.
I compiled xen the following way:
I downloaded the linux-2.6.18-xen-3.2.0 kernel and xen-3.2.1 separately,
and then first, i compiled the kernel by:
gzip -dc /proc/config.gz
make menuconfig
<SOME CONFIGURATION, especially don''t forget to set "Processor
type and
features" -> "High Memory Support" to 64GB in order to make it
a PAE
kernel>
make
<During the building process, i watch this video:
http://www.youtube.com/watch?v=PLyTrsk1Ey8>
make install
make modules_install
mkinitrd <IN SOME WAY>
After compiling and installing the kernel, I compiled and installed xen,
but I didn''t do a "make world" at all!! I did it the
following way:
On a 64 bit machine, or by cross-compiling (i''ve personally never
tried), I build the hypervisor:
make xen #this creates only the hypervisor (/boot/xen-3.3.0.gz)
make install-xen
After installing, I copy the hypervisor (/boot/xen-3.2.1.gz) to the
32bit target dom0.
And finally I compile and install the tools on the (32bit, of course)
dom0:
make tools #this creates thigs like the xm command, hvm environment,
#network scripts, etc.
make install-tools
Best regards,
Tamas
2008. 09. 25, csütörtök keltezéssel 11.22-kor Geoffrey ezt
írta:> We have some fairly robust hardware we are trying to set up a number of
> para-virtual servers in a clustered environment and we are seeing huge
> cpu issues. Specs:
>
> Dell 1950: 8 1u servers
> cpu MHz : 2660.038
> dual quad cores
> 32G memory each
>
> Our first virual server was configured to use 8G and 4 cores.
>
> I tried running a quick test by compiling our application, simply
> running ''time make.'' The results were terrible:
>
> real 0m56.79s
> user 0m26.53s
> sys 0m35.46s
>
> In comparison, my laptop (Intel Core2 Duo CPU @ 2.50GHz, 2G mem):
>
> real 0m42.99s
> user 0m29.85s
> sys 0m6.47s
>
> We then reconfigured the virtual server to use a single core and the
> results were almost identical to the 4 core test.
>
> Any insights as to what our problem might be would be greatly appreciated.
>
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
We have our solution. We have decided to go with 32bit para-virtual hosts on our 64bit hardware. Our speed issues are gone. I appreciate everyone''s input on this issue. -- Until later, Geoffrey Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety. - Benjamin Franklin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users