Thank you for this information, Alfredo.

I am not worried about performance right now.  Our focus is to test some
PF_RING functionality for our application and see what we need to do to our
application to support PF_RING before we commit.  So, I am happy to do
everything in a guest VM (Centos7): install PF_RING, install DNA or ZC
drivers, and test our application under simulated conditions.  Right now,
we are happy to test either DNA or ZC (or both) just to prove things work,
mainly load balancing (packet hashing).  We will use tcpreplay to test our
application, not live traffic.  Once we are happy it works like we want, we
will move to a production-ready platform with real cards with ZC.

My challenge has been a disconnect between drivers on a guest VM and the
DNA/ZC drivers.  My first approach was to try Centos7 on VirtualBox with an
e1000 driver.  However, I can't seem to build the DNA e1000 driver from the
PF_RING source (make and compile errors) and there are no pre-built RPMs.
Also, reading through the source code, I notice that all of the scripts &
Makefile commands for building an e1000 driver are commented out.  So, I
assumed e1000 was unsupported.

My next approach is to try an Amazon EC2 VM that supports either igb ixgbe,
and hopefully use the prebuilt ZC driver.  If that fails, I guess I will
try VMware.

To summarize, I really want to use PF_RING and DNA or ZC in a
proof-of-concept for our application to show others that it will work;
performance is not important to me right now.  Is my use case reasonable
for test-driving PF_RING DNA or ZC?  Am I missing something obvious,
especially with building a DNA e1000 driver on Virtualbox?

Many thanks in advance,
John

On Wed, Feb 18, 2015 at 4:04 AM, Alfredo Cardigliano <[email protected]>
wrote:

> Hi John
> e1000 cards are supported by DNA only,
> e1000e cards are supported by DNA and ZC,
> please note that:
> - even if you use DNA/ZC drivers in the guest os, you still have all the
> overhead in the host os/hypervisor
>   (for 10G we use PCI-passthrough/Direct-IO with DNA/ZC for line-rate on
> VMs, or ZC with a master
>   process on the host sending packets to a sw queue for zero-copy
> host-guest communications)
> - you will not get a significative performance improvement at the rates
> supported by e1000 cards
> - in most case e1000 cards are not able to do line-rate anyway
>
> Alfredo
>
> > On 16 Feb 2015, at 20:15, John Zachary <[email protected]> wrote:
> >
> > Hello,
> >
> > I am looking for a link to the latest support of PF_RING with either ZC
> or DNA support on virtual guest machines.  We want to evaluate PF_RING for
> our application and wish to test it out.  I am trying to bring everything
> up on a Centos7 guest on VirtualBox 4.3.20r96996 with either virtio or
> intel 82545EM NIC support.  It seems that the latest version of PF_RING
> software doesn't support e1000 cards (and it looks like e1000e might be on
> its way out, too, yes?).  If anyone can point me towards a solution, either
> some way to force Vbox to work or suggest alternatives (VMware? something
> else?), I would really appreciate it.  I don't want to buy a license if the
> answer is "it doesn't work on VMware either" :)
> >
> > I've look all over the web to answer my questions before posting,
> assuming this has been asked before, but I don't find much information
> about PF_RING and virtualization in the user community.
> >
> > John
> > _______________________________________________
> > Ntop-misc mailing list
> > [email protected]
> > http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>
>
_______________________________________________
Ntop-misc mailing list
[email protected]
http://listgateway.unipi.it/mailman/listinfo/ntop-misc

Reply via email to