[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: kvm and bonding




----- Original Message -----
From: "Stan Hoeppner" <stan@hardwarefreak.com>
To: debian-user@lists.debian.org
Sent: Monday, October 24, 2011 7:23:49 AM
Subject: Re: kvm and bonding

On 10/23/2011 3:05 PM, Jesus arteche wrote:
> hey guys,
> 
> do you know if it is possible to increase the bandwidth in a virtual machine
> running on kvm server with etehrnet bonding (LACP), or the bandwidth is
> limited by kvm per virtual machine and not for network adaptor???

If you haven't created any host kernel control groups to limit network
bandwidth to your guests, then each guest should have full access to all
available NIC bandwidth provided by the host kernel.  Thus, for example,
if you create a bonded ethernet adapter of 4 physical GbE NICs in the
host kernel and configure the GbE switch ports accordingly, any KVM
guest will be able to use the potential bandwidth of 4 Gb/s.  Actual
bonded throughput will depend on many factors.  Don't expect to get all
4 Gb/s from a single FTP Get operation.

Please read the following for more info.

Note:  You may not need additional link bandwidth.  You may simply need
to optimize your KVM network configuration to use bridge mode instead of
the default user-mode networking.

http://publib.boulder.ibm.com/infocenter/lnxinfo/v3r0m0/index.jsp?topic=%2Fliaat%2Fliaatseccgroups.htm

Read this for creating control groups on Debian:
http://linuxaria.com/article/introduction-to-cgroups-the-linux-conrol-group?lang=en

You may not need to bond another interface to increase network
performance.  You may simply need to optimize what you already have.
KVM by default enables user mode networking

Read the networking section of this:
http://www.linux-kvm.org/page/Tuning_KVM

Also read this:
http://www.linux-kvm.org/page/Networking_Performance

-- 
Stan
<snip>
We use bridging but there was a problem when we set up our systems a few years ago where bridging and bonding interfere with each other.  There were some improvements made in this area but I do not know the final results and we have not reinvestigated.

I believe the problem was with using bonding in alb mode and had to do with the arp method used to distribute MAC addresses for receive traffic.  tlb worked fine.  There should be a fairly extensive bugzilla report on it - John


Reply to: