[dpdk-users] TAP-PMD & tap-device PPS via Kernel

Giridharan, Ganesan ggiridharan at rbbn.com
Fri Jun 1 12:44:11 CEST 2018

Good morning..

Sharing my test results. Might be helpful, in case if this path is desired.

Hypervisor : Hyper-V
Guest : Linux 4.14
Processor : 2690v2
# of vCpus : 4
Packet flow:
               Host --> Kernel -->dpdk-->Kernel-->Host ( data plane traffic )
Packet types: UDP streams
I was able to comfortably reach abot ~28K-30K pps. Packet loss occurs at about or after that. Traffic was real-time latency sensitive.

Great day

From: Wiles, Keith <keith.wiles at intel.com>
Sent: Tuesday, May 29, 2018 11:53 AM
To: Giridharan, Ganesan <ggiridharan at rbbn.com>
Cc: users at dpdk.org
Subject: Re: [dpdk-users] TAP-PMD & tap-device PPS via Kernel

NOTICE: This email was received from an EXTERNAL sender

> On May 29, 2018, at 11:12 AM, Giridharan, Ganesan <ggiridharan at rbbn.com<mailto:ggiridharan at rbbn.com>> wrote:
> Good morning. Would like to know some numbers, especially PPS, using TAP device and TAP PMD combo to fetch packets to DPDK application. Wondering do those numbers change as Hypervisor platforms change? Assuming the resources (vCpu & memory) are unchanged across platforms. Mainly interested in PPS measured on ESXi/KVM/Hyper-V platforms.

I have never run TAP PMD in ESXi/KVM/Hyper-V so I do not know the numbers. The performance of the DPDK TAP PMD is not going to be much better then any other design using tun/tap as the kernel is still involved in the movement of data.

The TAP PMD was created to allow me to test applications with DPDK on a system without real NICs like my MACBook Pro I have a port of DPDK for a MAC. The TAP interface should only be used for testing or if you need connectivity to the host/another TAP device. With Pktgen I use it to connect to wireshark to decode packets.

> Appreciate the share if anyone might possess test results.
> Thanks
> Great day
> --GG


More information about the users mailing list