Damjan,
In my case, we have a four-socket Skylake server filled with 2x25GbE NICs and
encryption accelerators---trying to see how much IPsec throughput we can manage.
Recently saw a ~50% performance increase by changing from 2MB hugepages (8192
pages)
to 1GB hugepages (16 pages) with VPP 18.04
2M works but 1G still fails.
I toned down the dpdk resource allocation to default:
dpdk
{
dev 0004:01:00.1
dev 0004:01:00.2
no-multi-seg
log-level debug
dev default
{
num-rx-queues 1
# num-tx-queues 4
num-rx-desc 2048
num-tx-desc 2048
}
# num-mbufs 128000
#
No, long back I had uncommented a few lines from the old config and never
bothered to change (I have the whole server for myself so had the liberty to be
lavish).
Let me retest with the patch.
Thank you
Sirshak Das
From: Damjan Marion
Sent: Tuesday, October 23, 2018 3:43 PM
To: Sirshak Das
Hi Daolin,
You can see the code below in gtpu4_encap node, if you set gtpu4_encap as next
node, the sw_if_index should be set correctly. In you case, it should be set to
the gtpu tunnel1.
sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
A workable solution is add a node as the next node
I submitted patch
Just for my curiosity what is the use case for allocating >=2 GB of wired
memory?
I thought nobody wants to do that so i did not pay attention.
—
Damjan
> On 23 Oct 2018, at 21:00, Lee Roberts wrote:
>
> In the following code, take a close look at line #61:
>
>
>
OMG, you are good in wasting memory. 1G pages, 2G per socket given to dpdk to
hang empty :)
128K buffers
For a start can you switch default page size to 2M. newer x86 kernels ignore it
but maybe it behaves
differently on aarch64...
In the meantime I will fix few coverity issues...
--
Hi Damjan,
I am getting the following error as well I don't know if its related to this
issue:
vlib_plugin_early_init:361: plugin path
/home/sirdas/code/commita/vpp/build-root/install-vpp-native/vpp/lib/vpp_plugins
load_one_plugin:117: Plugin disabled (default): abf_plugin.so
Folks,
Looks like my big physmem patch breaks non-root operation of VPP, working on it
and It will take a bit of time so as a workaround "make test" can be run with
sudo.
Let me know if any issues, and I will revert, but would like to avoid that due
to the size of patch.
--
Damjan
I believe you can cross connect two gtpu tunnel interfaces on the VPP in your
GW to achieve what you need:
DBGvpp# set int l2 xconnect ?
set interface l2 xconnectset interface l2 xconnect
You need to set L2 xconnect on both gtpu tunnel interfaces to each other.
Regards,
I don't know if VXLAN-GPE is implemented in VPP to handle IP payload without
ethernet header properly. You can try it to check what happens. I would not
be surprised if something need to be fixed in VXLAN-GPE encap/decap for it to
work properly. Hopefully the owner of VXLAN-GPE can comment
I look forward to it :)
Ed
On Tue, Oct 23, 2018 at 8:42 AM Ni, Hongjun wrote:
> Hi Ed,
>
>
>
> OK. I or some project proposer will join the Nov 8 8am PT TSC meeting and
> present the proposal.
>
>
>
> Thanks a lot,
>
> Hongjun
>
>
>
> *From:* t...@lists.fd.io [mailto:t...@lists.fd.io] *On
Hi Ed,
OK. I or some project proposer will join the Nov 8 8am PT TSC meeting and
present the proposal.
Thanks a lot,
Hongjun
From: t...@lists.fd.io [mailto:t...@lists.fd.io] On Behalf Of Edward Warnicke
Sent: Tuesday, October 23, 2018 9:15 PM
To: Ni, Hongjun
Cc: t...@lists.fd.io;
Hongjun,
Thank you for the proposal :) Per the FD.io technical charter, all
proposals must be out for public review for two weeks prior to approval by
the TSC. I believe this makes Nov 8 the earliest TSC meeting where we
could approve at the TSC. Would you like to schedule the project creation
Hello FD.io TSCs
Please accept this project proposal for Sweetcomb for consideration.
https://wiki.fd.io/view/Project_Proposals/Sweetcomb
This project has nine founding companies:
Intel, HuachenTel, China Mobile, China Unicom, China Telecom, NXP, Tieto,
Tencent, Alibaba.
If possible, I would
Read at face value, the vpp binary API memory segment (VA range 0x3000 and
above) is out of memory. The failing allocation is a request for 60 bytes,
which is not an unreasonable ask.
Before doing anything else, try increasing the size of the API segment and see
if the problem recurs.
We
Hi all,
I am facing an occasional VPP crash from dead_client_scan() when I restart
a client .
I am using VPP version v18.01.1-100~g3a6948c. Upgrading to a newer version
is not an option for me currently.
Here is the backtrace :
Program terminated with signal 6, Aborted.
*#0 0x2ad10a37f207
See https://wiki.fd.io/view/VPP/VPP_Home_Gateway for a NAT-based example.
From: vpp-dev@lists.fd.io On Behalf Of
wangchuan...@163.com
Sent: Tuesday, October 23, 2018 2:55 AM
To: vpp-dev
Subject: [vpp-dev] How translate some ctrl data to OS ?
Hi all,
PCA have running a VPP that have only 1
Hi,
I have a question about gtpu encap node in VPP. My network topology is:
[cid:image002.jpg@01D46AC2.C0DA7910]
VPP is running on GW in Linux VM with two IP interfaces, and eNB & CN in the
above graph are running on two different Linux VMs with only one IP interface.
After starting VPP and
Hi Alp Arslan,
To make tap_cli work, you need to revert the code as per this patch:
https://gerrit.fd.io/r/#/c/9467/
Thanks,
Hongjun
-Original Message-
From: vpp-dev@lists.fd.io [mailto:vpp-dev@lists.fd.io] On Behalf Of
alp.ars...@xflowresearch.com
Sent: Tuesday, October 23, 2018 4:01
Dell - Internal Use - Confidential
Dear Hongjun,
Thanks for the reply.
I am using the VPP version 18.07. If the tap_cli is still present, can you
please point me towards what could be the problem with the existing plugin.
I would like to see this one working a little bit, before starting
Hi Alp Arslan,
Since tap_cli code is going to be deprecated in favour of tapv2.
I suggest you leverage tapv2, and also need some rework for PPPoE plugin.
Thanks,
Hongjun
-Original Message-
From: vpp-dev@lists.fd.io [mailto:vpp-dev@lists.fd.io] On Behalf Of
alp.ars...@xflowresearch.com
Dear All,
I am trying to evaluate the support of pppoe in vpp. However, I am facing
some issues, here is my startup conf file.
unix {
nodaemon
log /var/log/vpp/vpp.log
full-coredump
cli-listen /run/vpp/cli.sock
gid vpp
}
api-trace {
on
}
api-segment {
gid vpp
}
socksvr {
Hi all,
PCA have running a VPP that have only 1 hw_if : GigabitEthernet2/0/0 , 1
ip:172.16.0.2/16, and already have vxlan [172.16.0.2 - 172.16.0.3].
How should I config the vpp to let me login the PCA'sOS from 172.16.0.4 by SSH ?
Is there some feature ? e.g. ip-port mapping
23 matches
Mail list logo