Question maximum throughput virtio interface / channel bounding inside KVM guest

Question maximum throughput virtio interface / channel bounding inside KVM guest

[Denny Fuchs]

[-- Attachment #1: Type: text/plain, Size: 662 bytes --]

hi,

I have Proxmox3 VE (based on Debian Wheezy) up and running with 2 LACP enabled 1Gb Intel interfaces, connected to a CISCO. I want to know, how I can get more than one gigabit inside a KVM guest (also Debian Wheezy). What are the limits for virtio ? For our backup and fileserver it would be nice, to get more throughput. I don't need LACP inside the guest too, Round-Robin (or something else) would be OK. I was able to create a bounding inside the backup guest with round-robin, but I think, it doesn't play well, because both virtual interfaces are in the same bridge, with the physical interfaces (connected via LACP).

any suggestions?

cu denny

[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 163 bytes --]

Re: Question maximum throughput virtio interface / channel bounding inside KVM guest

[Denny Fuchs]

hi,

Am 08.11.2013 um 18:01 schrieb Freddie Cash <fjwcash@gmail.com>:
> 
> Fire up the VM and start multiple connections to multiple servers and you should get an aggregate throughput over 1 Gbps.

[…]

that is, what I already have :-)


> AFAIK, there's no way to get more than 1 Gbps on a single connection between the VM and a single server using LACP.

the NFS fileserver has up to 112 diskless clients which connects at same time, so I didn’t mean single client <-> single server. 

So, what are the limits for virtio? I get next year 10Gb interfaces :-)

cu denny

Re: Question maximum throughput virtio interface / channel bounding inside KVM guest

[Denny Fuchs]

[-- Attachment #1: Type: text/plain, Size: 428 bytes --]

hi Freddie,

Am 12.11.2013 um 19:19 schrieb Freddie Cash <fjwcash@gmail.com>:

> 
> AFAIK, the limit for virtio network devices (using vhost-net kernel module) is "as fast as the CPU can pump data through the PCIe bus to the NIC".  Meaning, the NIC is the bottleneck, not the virtio device.


that was, what I wanted to hear. I wasn't sure, if there are any limits from the virtio side.

Thanks a lot :-)

cu denny

[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 163 bytes --]