[vpp-dev] gtpu tunneling decap-next ip4 issue

Hi, all Let me ask about a GTPu issue. Although I tried to overlay IPv4 packets with GTP-u, it didn't work by 17.10. Actually vpp was rebooted silently when I sent ping. Could someone help or provide GTPu IPv4 sample configuration ? My situation: By following diagram, when I sent ping from 10.9

Re: [vpp-dev] C++ compilation uses a lot of memory

Thanks, Ole, misery loves company:-) The only question I'd ask about your build change is whether build-root/Makefile should be modified to allow command line args, or whether it should stay as is and you write the configuration options into a locally modified copy of build-data/platforms/$(PLATFO

Re: [vpp-dev] Question about IP reassembly in 17.10 release

Hi Marion, Thanks for this info, it’s clear for me now. Best Regards Michael From: Damjan Marion [mailto:dmarion.li...@gmail.com] Sent: 2017年11月1日 17:41 To: Yu, Michael A. (NSB - CN/Qingdao) Cc: vpp-dev Subject: Re: [vpp-dev] Question about IP reassembly in 17.10 release On 30 Oct 2017, at

Re: [vpp-dev] C++ compilation uses a lot of memory

Burt et al, > I hadn't built new code in a few weeks. I have a machine with 8 hyperthreads, > and I have only 8GB memory minus what is used for hugepages. I ran out of > virtual memory when doing a build, in the new C++ code. Just a heads up for > anyone without huge memory per processor thread

[vpp-dev] VPP @ IETF 100

Guys, VPP is in a couple of projects at the IETF 100 Hackathon in Singapore. https://www.ietf.org/registration/MeetingWiki/wiki/100hackathon If you aren't able to join there, but are in Singapore later in the week. Would you be interested in a small VPP developer gathering? Lunch, dinner or hid

[vpp-dev] C++ compilation uses a lot of memory

I hadn't built new code in a few weeks. I have a machine with 8 hyperthreads, and I have only 8GB memory minus what is used for hugepages. I ran out of virtual memory when doing a build, in the new C++ code. Just a heads up for anyone without huge memory per processor thread. My work around is goin

Re: [vpp-dev] vpp building problem

What you have done is build, you need to do the installation, and configuration (if needed), before you can start the service. Take a look at this link: https://wiki.fd.io/view/VPP/Installing_VPP_binaries_from_packages I am on CentOS, the build packages are under extras/rpm. If your build is succ

Re: [vpp-dev] vpp building problem

Hello, If you want to run it as a service then I suggest you use the pre-built packages, or build the packages yourself and install those. Only the act of installing a package prepares the underlying system to run VPP as a system service. The build instruction you gave leaves the built binarie

Re: [vpp-dev] VPP low throughput

Avi, You can tune a number of things to get higher throughput - Using VPP worker threads - Up the queue size to 1024 from the default 256 (done via qemu launch) - Enable multi-queues (done via qemu launch) - Up the number of streams when invoking iperf3 using –P Nevertheless, 10s Mbps is unusual

[vpp-dev] vpp building problem

Hi all i'm trying to install vpp from source on Ubuntu 16.04 server (without virtualization). I obeyed the following this link: https://wiki.fd.io/view/VPP/Pulling,_Building,_Running,_Hacking_and_Pushing_VPP_Code#Pulling_anonymously_.28https.29 So i did as such: sudo -s cd / git clone https://ge

Re: [vpp-dev] 50GE interface support on VPP

Thanks for the info Daniel. -Nitin From: Bernier, Daniel Sent: Wednesday, November 1, 2017 6:58 PM To: Saxena, Nitin; Damjan Marion (damarion) Cc: vpp-dev@lists.fd.io Subject: Re: [vpp-dev] 50GE interface support on VPP Hi, I have had the same issue and sti

[vpp-dev] VPP low throughput

It is my 1st time running a VPP. My setup is 2 VM's with a vpp in the middle (connected with vhost-user) as per - https://wiki.fd.io/view/VPP/Use_VPP_to_connect_VMs_Using_Vhost-User_Interface But get a very low throughput with iperf3 - 10s Mbps . any idea ? VPP is always running with DPDK -

Re: [vpp-dev] [SFC] Query regarding SFC classifier configuration for ip4 traffic

Hi Hongjun, Thank your very much for your support. I have tried this configuration for UDP packets and below is my configuration classify table mask l3 ip4 proto classify session l2-input-hit-next input-node nsh-classifier table-index 0 match l3 ip4 proto 17 opaque-index 47615 set int l2 bridge

Re: [vpp-dev] 50GE interface support on VPP

Hi, I have had the same issue and still working on the fix with Mellanox on this one. ​Damjan is right it is just cosmetic (although annoying). On the Linux kernel side, it implies moving to kernel 4.8 and above and a newer version of ethtools. On the VPP side, it just requires Mellanox to adve

Re: [vpp-dev] 50GE interface support on VPP

I havent ran testpmd but with VPP I am able to switch traffic between two ports. Both ports in VPP bridge. Seems fine right -Nitin From: Damjan Marion (damarion) Sent: Wednesday, November 1, 2017 5:50:57 PM To: Saxena, Nitin Cc: Bernier, Daniel; vpp-dev@lists.f

Re: [vpp-dev] 50GE interface support on VPP

Currently it is just cosmetic….. Does it work with testpmd? — Damjan On 1 Nov 2017, at 13:14, Saxena, Nitin mailto:nitin.sax...@cavium.com>> wrote: Ok Thanks I will debug where the problem lies. However is this just a display issue or problem lies with data path as well because I am able to

Re: [vpp-dev] 50GE interface support on VPP

Ok Thanks I will debug where the problem lies. However is this just a display issue or problem lies with data path as well because I am able to receive packets via this NIC to VPP from outside world? Any concern here? -Nitin From: Damjan Marion (damarion) Se

Re: [vpp-dev] 50GE interface support on VPP

mlx5 dpdk driver is telling us that speed_capa = 0, so no much love here. You should get at least ETH_LINK_SPEED_50G bit set by dpdk driver. — Damjan On 1 Nov 2017, at 12:55, Saxena, Nitin mailto:nitin.sax...@cavium.com>> wrote: Here is the detail (gdb) p *dev_info $1 = {pci_dev = 0x51c4e0,

Re: [vpp-dev] 50GE interface support on VPP

Here is the detail (gdb) p *dev_info $1 = {pci_dev = 0x51c4e0, driver_name = 0x76eb38a8 "net_mlx5", if_index = 8, min_rx_bufsize = 32, max_rx_pktlen = 65536, max_rx_queues = 65535, max_tx_queues = 65535, max_mac_addrs = 128, max_hash_mac_addrs = 0, max_vfs = 0, max_vmdq_pools = 0, rx_offl

Re: [vpp-dev] 50GE interface support on VPP

Can you put breakpoint to port_type_from_speed_capa and catpture dev_info. I.e: $ make build debug (gdb) b port_type_from_speed_capa (gdb) r (gdb) p * dev_info — Damjan On 1 Nov 2017, at 12:34, Saxena, Nitin mailto:nitin.sax...@cavium.com>> wrote: Please find show pci output DBGvpp# show

Re: [vpp-dev] 50GE interface support on VPP

Please find show pci output DBGvpp# show pci Address Sock VID:PID Link Speed Driver Product Name Vital Product Data :0b:00.0 0 14e4:16a1 8.0 GT/s x8 bnx2x OCP 10GbE Dual Port SFP+ Adapter :32:00.1 0 15b3:1013 8.0 GT/s x16 mlx5_

[vpp-dev] MACCHIATObin and VPP

If people are interested, there is ongoing work[1] to bring VPP up on Marvell MACCHIATObin[2] board. Interesting ARM64 community board with SFP+ ports. [1] https://github.com/MarvellEmbeddedProcessors/vpp-marvell [2] http://macchiatobin.net — Damjan

Re: [vpp-dev] Question about IP reassembly in 17.10 release

> On 30 Oct 2017, at 08:21, Yu, Michael A. (NSB - CN/Qingdao) > wrote: > > Hi, > > From https://docs.fd.io/vpp/17.10/release_notes_1710.html > , I found that “IP > reassembly” is claimed to support in 17.10 release. > But after checked

Re: [vpp-dev] 50GE interface support on VPP

Can you share “show pci” output from VPP? — Damjan On 30 Oct 2017, at 14:22, Saxena, Nitin mailto:nitin.sax...@cavium.com>> wrote: Hi Damjan, I am still seeing UnkownEthernet32/0/0/0 interface with Mellanox Connect X-4 NIC. I am using vpp v17.10 tag. I think the specified gerrit patch in fol

Re: [vpp-dev] dpdk output function

Hi Yuliang, It will call: vnet_interface_output_node() from src/vnet/interface_output.c There is also a TenGigabitEthernet5/0/1-tx node. Since you Are using DPDK on this interface, that will call: dpdk_interface_tx() from src/plugins/dpdk/device/device.c hth neale From: on behalf of Yu