Hi Joydev,
Sounds like an IXIA may be required.
I spend most of my time in the wifi space and don't have current awareness
of 10G testing at your scale. Testing 960G of IP traffic is quite a lot.
The packet sizes will impact performance, e.g. smaller packets require more
processing per byte transferred. CPU based systems will struggle with
small packet sizes. Years ago, when I did test multiple 10G for a router
company we purchased expensive IXIAs and expensive IXIA line cards.
Basically, specialized test hardware was required to test the switching
ASICs.
In theory, one could throw many hosts with 10G NICs at the problem. But to
coordinate the 96 10G iperf flows would require something like src/flows
found in iperf2
<https://sourceforge.net/p/iperf2/code/ci/master/tree/flows/> (though this
is in the very early proto stages so I wouldn't recommend it unless
developing in python and 3.5+ asyncio was part of the project.) If I took
this approach I'd focus on server class systems for those hosts and NICs.
You'd probably need 48 of them and they likely wouldn't fill a 10G pipe
when using small packet sizes.
Bob
On Tue, Jun 20, 2017 at 7:49 AM, Jana, Joydev <joydev.j...@coriant.com>
wrote:
> Hello Bob,
>
> Thanks for the pros and cons data which is very useful for us to analyse
> and make the best decision of our test equipment choice.
>
> I had a chat with one of the switch vendor and his comment was also as
> unsure about the CPU capability even if forwarding are done with latest
> Broadcom Tridant or Tomahawk chip and they are planning to investigate this
> for us.
>
>
>
> As described in my previous mail about the application, we want to
> simultaneously send 10G IP traffic and monitor the flow/any errored
> frames/drops/packet sequences/Jitter or Latency/throughput/QoS behaviour
> etc and with bi-directional flow for 48 pairs.
>
> Are you aware of any user application similar to our plan and does it
> really work when using such tools, compared to dedicated test equipment?
>
>
>
> Regards,
>
> Joydev
>
>
>
> *From:* Bob McMahon [mailto:bob.mcma...@broadcom.com]
> *Sent:* 19 June 2017 20:34
> *To:* Jana, Joydev
> *Cc:* iperf-users@lists.sourceforge.net
> *Subject:* Re: [Iperf-users] Iperf usage in Telecom network testing
>
>
>
> Hi Joydev,
>
>
>
> Yes, the number of physical ports is limited by the hardware/host. The
> number of iperf flows is not, i.e. a single host can have multiple iperf
> streams.
>
>
>
> As standard testing there are no switches which iperf is compiled for.
> One would have to talk with the switch vendor to have them supply iperf for
> their switch. I also would be a bit wary about this because switches tend
> to have less substantial CPUs than let's say an Intel server. So, while
> it's ability to switch traffic is high performance, the ability to source
> and sink traffic may not be.
>
>
>
> Another option is to use vlans on the host with the NIC connected to a
> vlan capable switch as the mux and demux. Connect a quality server to the
> switch over a trunk port. Maybe use a 10G NIC port to a switch with ten 1G
> ports.
>
>
>
> For many of my setups I use the Intel quad port server adapter
> <https://www.intel.com/content/www/us/en/ethernet-products/gigabit-server-adapters/overview.html>
> to get multiple NICs. Also, you may need to install some policy routes
> when sourcing traffic out of the same host with multiple NIC ports. This
> allows one to have an iperf flow use a particular NIC port.
>
>
>
> Bob
>
>
>
> On Sun, Jun 18, 2017 at 10:46 PM, Jana, Joydev <joydev.j...@coriant.com>
> wrote:
>
> Hello,
>
> We are investigating the possibility to use any open source IP traffic
> generator/Analyser app that can be installed in any device to be able to
> verify error free flow.
>
> Reading through the documentation of Iperf, this tool is possible to do so
> when installed into Client/server setup which can be Windows or Linux based
> Machine.
>
> This means the physical ports that can be used from the Linus Machine are
> connected to the DUT/Network under test.
>
> This also means that number of ports available from the Generator/Analyser
> are guided by number of NIC ports available.
>
> Is this correct understanding?
>
>
>
> This leads to another question:
>
> Can Iperf be installed into Switches which has say 50 to 100 ports
> available and each can be used as a Generator/Analyser port?
>
> If so, do you have any recommendation to type of switch in use by already
> users who are using Iperf in such application requiring say 50 ports used
> simultaneously to generate and analyse IP traffic packets.
>
>
>
> Thanks,
>
> Joydev Jana
>
> Coriant Oy
>
> Sinimaentie 6
>
> 02630 Finland
>
> +358 40 757 4862 <+358%2040%207574862>
>
> joydev.j...@coriant.com
>
>
>
>
>
>
> ------------------------------------------------------------
> ------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> Iperf-users mailing list
> Iperf-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/iperf-users
>
>
>
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Iperf-users mailing list
Iperf-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/iperf-users
