[dpdk-dev] Testing memnic for VM to VM transfer

[Hiroshi Shimamoto]

Hi,

> Subject: [dpdk-dev] Testing memnic for VM to VM transfer
> 
> Hi everyone:
>         We are interested in testing the performance of the memnic driver posted at     http://dpdk.org/browse/memnic/refs/.
> We want to compare its performance compared to other techniques to transfer packets between the guest and the kernel,
> predominately for VM to VM transfers.
> 
> We have downloaded the memnic components and have got it running in a guest VM.
> 
> The question we hope this group might be able to help with is what would be the best way to processes the packets in the
> kernel to get a VM to VM transfer.

I think there is no kernel code work with MEMNIC.
The recommend switching software on the host is Intel DPDK vSwitch hosted on 01.org and github.
https://github.com/01org/dpdk-ovs/tree/development

Intel DPDK vSwitch runs on userspace not kernel.

I introduced this mechanism to DPDK vSwitch and the guest drivers are maintained in dpdk.org.

thanks,
Hiroshi

> 
> A couple options might be possible
> 
> 
> 1.       Common shared buffer between two VMs.  With some utility/code to switch TX & RX rings between the two VMs.
> 
> VM1 application --- memnic  ---  common shared memory buffer on the host --- memnic  ---  VM2 application
> 
> 2.       Special purpose Kernel switching module
> 
> VM1 application --- memnic  ---  shared memory VM1  --- Kernel switching module  --- shared memory VM2  --- memnic  ---
> VM2 application
> 
> 3.       Existing Kernel switching module
> 
> VM1 application --- memnic  ---  shared memory VM1  --- existing Kernel switching module (e.g. OVS/linux Bridge/VETh pair)
> --- shared memory VM2  --- memnic  ---  VM2 application
> 
> Can anyone recommend which approach might be best or easiest?   We would like to avoid writing much (or any) kernel code
> so if there are already any open source code or test utilities that provide one of these options or would be a good starting
> point to start from,  a pointer would be much appreciated.
> 
> Thanks in advance
> 
> 
> John Joyce