Re: [Xen-users] The bottleneck of using multinetwork card?

[ma qiang <maqiang1984@xxxxxxxxx>]

>On Mon, Sep 06, 2010 at 02:52:16PM +0800, ma qiang wrote:
>    I  did the same test one the same computer in dom0 using only one network
>    card,  the results show 1000M/sec.
>

>So what kind of guests are you using the benchmark from?

That is IXIA side based on centos 5.4 also.

On Mon, Sep 6, 2010 at 3:02 PM, Pasi Kärkkäinen <pasik@xxxxxx> wrote:

On Mon, Sep 06, 2010 at 02:58:32PM +0800, ma qiang wrote:
>    Sorry, I lost two words.
>     I reboot in bare linux not dom0, and have tested it on the same computer
>    using only one network card. the results show 1000M.
>

So what do you get from xen dom0?
In dom0 and bare linux, the results show both 1000M. But in domU, the results are bad.

 

Also if your guest is HVM, have you installed PV-on-HVM drivers?
without those drivers the performance will be bad from the guest.
PV-on-HVM drivers ? Sorry, I haven't installed that.  And I'm going to search that. 

 

-- Pasi

>    On Mon, Sep 6, 2010 at 2:57 PM, ma qiang <[1]maqiang1984@xxxxxxxxx> wrote:
>
>      20% is the vm's utilization of cpu, not dom0's.
>      My domU is hvm and centos linux.
>      "test server" is OK, and I have tested is on the same computer using
>      only one network card. the results show 1000M.
>

>      On Mon, Sep 6, 2010 at 2:50 PM, Pasi Kärkkäinen <[2]pasik@xxxxxx> wrote:
>
>        On Mon, Sep 06, 2010 at 02:44:59PM +0800, ma qiang wrote:
>        >    > Do you mean 250 Mbit/sec ?
>        >    Yes.
>        >    > You should get much more.. how are you measuring the
>        performance?
>        >    > What benchmark are you using?
>        >    I use an tool from ixia to test the network in the vm.
>
>        Try using some standard network performance testing tools
>        such as "iperf" or just regular ftp file transfers.
>        >    > Are both the eth0 and eth1 connected to the same network/vlan?
>        >    Yes.  I run my test from another computer to connect to vm0 and
>        vm1,  and
>        >    they all in the same switch.
>        >    To be noted my cards are both 1000m network card.
>        >
>
>        How is the 'test server' connected?
>
>        >    > When you run the network benchmark try running "xm top"
>        >    > to monitor the overall cpu usage.. also run "top" in dom0
>        >    > to see if there's a big cpu usage in dom0.
>        >    No, the cpu is less than 20%
>
>        20% on dom0?
>        How about the domUs? Do they have big cpu usage?
>
>        What kind of guests do you have? pv? hvm? linux? windows?
>
>        -- Pasi
>        >    On Mon, Sep 6, 2010 at 2:08 PM, Pasi Kärkkäinen

>        <[1][3]pasik@xxxxxx> wrote:
>        >    > On Mon, Sep 06, 2010 at 01:58:10PM +0800, ma qiang wrote:
>        >    >> Hi all,
>        >    >>     I have install xen 3.4.3 based on centos5.4.
>        >    >>     I changed a line in /etc/xen/xend-config.sxp like below:
>        >    >> (network-script network-bridge)   ---->    (network-script
>        >    my-network-bridge)
>        >    >>
>        >    >> and #cat /etc/xen/scripts/my-network-script prints as below:
>        >    >> #!/bin/sh
>        >    >> /etc/xen/scripts/network-bridge "$@" netdev=eth0 bridge=eth0
>        >    >> /etc/xen/scripts/network-bridge "$@" netdev=eth1 bridge=eth1
>        >    >>
>        >    >> Now I installed two vms, and set one vm using eth0, then set
>        the other
>        >    >> using eth1.
>        >    >> So, I can connect internet in both vms.
>        >    >>
>        >    >> But now my problem is that the throughput of two network card
>        is the
>        >    >> same as only one network card.
>        >    >> Another words, If I use the only one network card such as eth0
>        for the
>        >    >> two vms, the throughput of eth0 is about 250m.
>        >    >> But when I set one vm use eth0 and the other use eth1, the
>        total
>        >    >> throughput of eth0 and eth1 is about 250m too.
>        >    >>
>        >    >> That why?  Any bottleneck?
>        >    >> Thanks a lot.
>        >    >>
>        >    >
>        >    > Do you mean 250 Mbit/sec ?
>        >    >
>        >    > You should get much more.. how are you measuring the
>        performance?
>        >    > What benchmark are you using?
>        >    >
>        >    > Are both the eth0 and eth1 connected to the same network/vlan?
>        >    >
>        >    > When you run the network benchmark try running "xm top"
>        >    > to monitor the overall cpu usage.. also run "top" in dom0
>        >    > to see if there's a big cpu usage in dom0.
>        >    >
>        >    > -- Pasi
>        >    >
>        >    >
>        >
>        > References
>        >
>        >    Visible links

>        >    1. mailto:[4]pasik@xxxxxx
>
> References
>
>    Visible links

>    1. mailto:maqiang1984@xxxxxxxxx
>    2. mailto:pasik@xxxxxx

>    3. mailto:pasik@xxxxxx
>    4. mailto:pasik@xxxxxx

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