Re: [Xen-users] DomU - network performance and possibilities (HVM, PVM, pci-passthru, stubdom, PV on HVM...)

[Pasi Kärkkäinen <pasik@xxxxxx>]

On Tue, May 18, 2010 at 09:45:27PM +0800, lei yang wrote:
>    On Tue, May 18, 2010 at 6:19 PM, Pasi Kärkkäinen <[1]pasik@xxxxxx> wrote:
> 
>      On Tue, May 18, 2010 at 05:42:15PM +0800, lei yang wrote:
>      >    On Thu, May 13, 2010 at 9:12 PM, Pasi Kärkkäinen
>      <[1][2]pasik@xxxxxx> wrote:
>      >
>      >      Hello,
>      >
>      >      Matej: I didn't get your initial post to xen-users because I got
>      >      (automatically) unsubscribed
>      >      from the list, due to my email provider getting blacklisted
>      because of
>      >      some spam problems,
>      >      but now I subscribed again..
>      >
>      >      To answer your questions:
>      >
>      >      1) No, it's not possible to use VirtIO with Xen.
>      >      2) Yes, it's possible to use PV-on-HVM drivers in the guest with
>      pvops
>      >      dom0.
>      >
>      >      3) Take a look at:
>      >      [2][3]http://wiki.xensource.com/xenwiki/XenLinuxPVonHVMdrivers
>      >
>      >      Did you already make some benchmarks? If you need good latency
>      then I
>      >      think
>      >      PCI passthrough will help a lot there.. or Xen netchannel2.
>      >
>      >
>      >
>      >      Which one will be better for thoughput,  pci passthrogh or
>      netchanel2?
>      >
> 
>      pci passthrough.
> 
> 
>    Can you tell me why ?
>

Because with pci passthrough the guest has direct access to the NIC.
Netchannel and netchannel2 (bridged networking) goes through dom0.

-- Pasi

>    Thanks
>    Lei
> 
> 
> 
>      -- Pasi
>      >      Thanks
>      >      Lei
>      >
>      >
>      >
>      >      -- Pasi
>      >
>      >      Original email from Matej:
>      >
>      >      Hi,
>      >
>      >      I want to (need to) do a network performance analysis/benchmark
>      of
>      >      various Xen networking solutions (analyzing the pros and cons of
>      network
>      >      infrastructure virtualisation) on top of Xen 4.0.
>      >
>      >      So far I have Xen 4.0 running with various Dom0 kernels (git
>      stable
>      >      2.6.32.11, git stable 2.6.31.13, mercurial 2.6.18.8, Debian
>      distribution
>      >      2.6.26-2xen...want try some newer forward ported Xen kernel too).
>      >
>      >      AFAIK, there are these possibilities for DomU "architecture".
>      >
>      >      1. DomU as classic emulated HVM.
>      >      2. DomU as classic PVM (pv_ops kernel and Xen kernel)
>      >      3. DomU as classic emulated HVM with Xen PV drivers (don't know
>      whether
>      >      these work with pv_ops Dom0, but for that case I have the
>      2.6.18.8 Xen
>      >      linux)
>      >      4. DomU as classic PVM with pass-thru PCI NICs
>      >      5. DomU as classic HVM with pass-thru PCI NICs (don't have IOMMU
>      capable
>      >      HW :( )
>      >      6. DomU as HVM with stubdom (according to various papers, this
>      should be
>      >      a bit faster than classic HVM DomU)
>      >      7. DomU as HVM with virtio drivers (is this possible in xen4.0?)
>      >
>      >      I'm interested in the latency and bandwidth performance in
>      phys-DomU and
>      >      DomU-DomU scenarios (I have own tool for network latency
>      benchmarking
>      >      and for bandwidth there's the Iperf :) ). Also I will try to
>      experiment
>      >      a bit with the CPU pinning and schedulers in these scenarios.
>      >
>      >      Also would like to test and testbench OpenSwitch as linux-bridge
>      >      replacement (but I'm afraid I will not have time for this).
>      >
>      >      So my questions are:
>      >
>      >      1. Is possible to use VirtIO with XEN 4.0 (as another approach on
>      the
>      >      PV-on-HVM)?
>      >      2. Is possible to use the Xen HVM DomU PV drivers (unmodified
>      drivers?)
>      >      with pvops Dom0?
>      >      3. If you have some ideas or hints, I will be very thankful. :)
>      >
>      >      Regards
>      >
>      >      Matej
>      >
>      >      _______________________________________________
>      >      Xen-users mailing list
>      >      [3][4]Xen-users@xxxxxxxxxxxxxxxxxxx
>      >      [4][5]http://lists.xensource.com/xen-users
>      >
>      >    --
>      >    "We learn from failure, not from success!"
>      >
>      > References
>      >
>      >    Visible links
>      >    1. mailto:[6]pasik@xxxxxx
>      >    2. [7]http://wiki.xensource.com/xenwiki/XenLinuxPVonHVMdrivers
>      >    3. mailto:[8]Xen-users@xxxxxxxxxxxxxxxxxxx
>      >    4. [9]http://lists.xensource.com/xen-users
> 
>    --
>    "We learn from failure, not from success!"
> 
> References
> 
>    Visible links
>    1. mailto:pasik@xxxxxx
>    2. mailto:pasik@xxxxxx
>    3. http://wiki.xensource.com/xenwiki/XenLinuxPVonHVMdrivers
>    4. mailto:Xen-users@xxxxxxxxxxxxxxxxxxx
>    5. http://lists.xensource.com/xen-users
>    6. mailto:pasik@xxxxxx
>    7. http://wiki.xensource.com/xenwiki/XenLinuxPVonHVMdrivers
>    8. mailto:Xen-users@xxxxxxxxxxxxxxxxxxx
>    9. http://lists.xensource.com/xen-users



_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users