Re: [vpp-dev] vpp_stats.py 03/18 changes cause issue

[Tianyu Li]

We also spotted the same error on local ubuntu 18.04 server, however ubuntu 
20.04 is working fine.

A patch is submitted and under review.
https://gerrit.fd.io/r/c/vpp/+/31972

Thanks,
Tianyu


From: vpp-dev@lists.fd.io  on behalf of hemant via 
lists.fd.io 
Sent: Thursday, April 15, 2021 6:04 AM
To: vpp-dev@lists.fd.io 
Subject: [vpp-dev] vpp_stats.py 03/18 changes cause issue


Ole,



My plugin’s python test (uses Python3) ran fine until I used a latest gerrit 
vpp repo today. I get the following error:



   super(TestUPF6, self).tearDown()

  File "/home/hemant/vpp/test/framework.py", line 798, in tearDown

self.logger.info(self.statistics.set_errors_str())

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line 230, 
in set_errors_str

error_counters = self.set_errors()

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line 213, 
in set_errors

self.connect()

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line 135, 
in connect

socket.close(mfd)

AttributeError: module 'socket' has no attribute 'close'



“git blame” shows this checkin that has caused the issue.



e66443c9b4 (Ole Troan   2021-03-18 11:12:01 +0100 135) 
socket.close(mfd)



Should we file a JIRA issue to fix the problem?



Thanks,



Hemant



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19206): https://lists.fd.io/g/vpp-dev/message/19206
Mute This Topic: https://lists.fd.io/mt/82104033/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] TX Queue Placement

[Feng Gao]

Hi Macros,

You could use "show interface" to check if there are some drop packets.
If so, it may mean the rx buffer isn't enough.

And you could use "show node counters" to check if there some drop
packets by other reasons.

On Wed, Apr 14, 2021 at 8:41 PM Marcos - Mgiga  wrote:
>
> Understood...
>
>
>
>
>
> Usually I use show run to monitor packet drops. Do you have another 
> suggestion for me to get a more precise information about it?
>
>
>
> Best Regards
>
>
>
> De: vpp-dev@lists.fd.io  Em nome de Damjan Marion via 
> lists.fd.io
> Enviada em: quarta-feira, 14 de abril de 2021 09:06
> Para: Marcos - Mgiga 
> Cc: vpp-dev@lists.fd.io
> Assunto: Re: [vpp-dev] TX Queue Placement
>
>
>
>
>
> General rule is that increasing those values are reducing number of rx drops 
> due to
>
> VPP being de-scheduled or busy or due to traffic bursts, but also that 
> degrades performance.
>
>
>
> I think what we have set as default is reasonable and you should leave it 
> unless you are experiencing some issues…
>
>
>
> —
>
> Damjan
>
>
>
>
>
> On 14.04.2021., at 13:45, Marcos - Mgiga  wrote:
>
>
>
> Hello Damjan,
>
>
>
> Thank you for clarifying...
>
>
>
> I also have a question about num-rx-desc and num-rt-desc parameters, hope you 
> dont mind to dicuss it in this e-mail
>
>
>
> I would like to understand what value fits better to my environment, do have 
> any thoughts about it?
>
>
>
> Best Regards
>
> De: vpp-dev@lists.fd.io  Em nome de Damjan Marion via 
> lists.fd.io
> Enviada em: quarta-feira, 14 de abril de 2021 08:24
> Para: Marcos - Mgiga 
> Cc: vpp-dev@lists.fd.io
> Assunto: Re: [vpp-dev] TX Queue Placement
>
>
>
>
>
>
>
>
> On 14.04.2021., at 13:21, Marcos - Mgiga  wrote:
>
>
>
> Hello,
>
>
>
> I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since 
> VPP is running on a NUMA system with two threads with 8 cores each, I would 
> like to pin TX / RX queue to proper NUMA nodes.
>
>
>
> Using set interface rx-placement I was able to associate rx queue to desired 
> cores, so I would like to know if is there any possibility to pin tx queue to 
> a certain workers as well.
>
>
>
>
>
> Not at the moment. tx queues are statically mapped (0 to main thread, 1 to 
> worker 0, 2 to worker 1, etc.).
>
> There are some plans to implement such capability…..
>
>
>
>
>
>
>
>
>
>
> 
>

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19205): https://lists.fd.io/g/vpp-dev/message/19205
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

[vpp-dev] Changes to .api files

[Paul Vinciguerra]

Hi Everyone.

I'd like to propose three more changes/cleanup to the api files/vppapigen.

1. I'd like to remove the logic in vpppigen that imperatively declares the
service definitions.  Instead of calculating the values, we would output
the necessary values to be added to the api file.

2. I'd like to change the existing message definitions to use the
corresponding enums in place of the primitives that they decay to.
u8 dscp -> vl_api_dscp_t dscp

3. I'd like to change the flag type enums to use the api enumflag type.

Objections?

Paul

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19204): https://lists.fd.io/g/vpp-dev/message/19204
Mute This Topic: https://lists.fd.io/mt/82107376/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

[vpp-dev] vpp_stats.py 03/18 changes cause issue

[hemant via lists.fd.io]

Ole,

 

My plugin's python test (uses Python3) ran fine until I used a latest gerrit
vpp repo today. I get the following error:

 

   super(TestUPF6, self).tearDown()

  File "/home/hemant/vpp/test/framework.py", line 798, in tearDown

self.logger.info(self.statistics.set_errors_str())

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line
230, in set_errors_str

error_counters = self.set_errors()

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line
213, in set_errors

self.connect()

  File "/home/hemant/vpp/src/vpp-api/python/vpp_papi/vpp_stats.py", line
135, in connect

socket.close(mfd)

AttributeError: module 'socket' has no attribute 'close' 

 

"git blame" shows this checkin that has caused the issue.

 

e66443c9b4 (Ole Troan   2021-03-18 11:12:01 +0100 135)
socket.close(mfd)

 

Should we file a JIRA issue to fix the problem?

 

Thanks,

 

Hemant

 



smime.p7s
Description: S/MIME cryptographic signature

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19203): https://lists.fd.io/g/vpp-dev/message/19203
Mute This Topic: https://lists.fd.io/mt/82104033/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

[vpp-dev] Memory requirements for CGN

[Marcos - Mgiga]

Hello VPP community,

 

According to Wiki page "Deterministic NAT prealocate vector with 1000
session slots (one session 15B) for each host from inside network range". 

 

In case a expect to have about 16 k users, with an address share ratio of 1
to 64 (1000 ports per user) is it correct to expect up to 2000 sessions per
user ( 1000 tcp and 1000 udp), so this prealocated slots can be an issue?

 

In the previous scenario is it correct to expect up to 480MB of memory usage
( 2000 sessions * 16000 users * 15)?

 

When using deterministic mode is "nat mode max translations per user" a
parameter that must be specified or is it calculated based on address
sharing ratio?

 

It is  correct to claim that since deterministic mode have an static mapping
from a range of internal IP/port to a range of external ip / port it doesnt
require a bi hash table so "user hash memory", "user hash buckets","
translation hash memory" and translation hash memory" are not available to
work in this mode. Does it make sense?

 

 

Thank you very Much

 

Best Regards

 

 


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19202): https://lists.fd.io/g/vpp-dev/message/19202
Mute This Topic: https://lists.fd.io/mt/82101219/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] vlib_buffer_clone behavior when trying to send to two interfaces

[Damjan Marion via lists.fd.io]

> On 14.04.2021., at 17:44, David Gohberg  wrote:
> 
> using DPDK with Mellanox MT27800 [ConnectX-5] (100G)

might be worth trying to use native rdma diver.

> show buffers indeed shows a leak:
> 
> Pool NameIndex NUMA  Size  Data Size  Total  Avail  Cached   Used 
>  
> default-numa-0 0 0   2496 2048   430185 430185 0   0  
>  
> default-numa-1 1 1   2496 2048   430185  71913101   
> 358171 
> 
> this keeps increasing until reaching 100% used


I would suspect your code first for buffer leaks…..

— 
Damjan
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19201): https://lists.fd.io/g/vpp-dev/message/19201
Mute This Topic: https://lists.fd.io/mt/81938105/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] vlib_buffer_clone behavior when trying to send to two interfaces

[David Gohberg]

using DPDK with Mellanox MT27800 [ConnectX-5] (100G)
show buffers indeed shows a leak:

Pool Name            Index NUMA  Size  Data Size  Total  Avail  Cached   Used
default-numa-0         0     0   2496     2048   430185 430185     0       0
default-numa-1         1     1   2496     2048   430185  71913    101   358171

this keeps increasing until reaching 100% used

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19200): https://lists.fd.io/g/vpp-dev/message/19200
Mute This Topic: https://lists.fd.io/mt/81938105/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] vlib_buffer_clone behavior when trying to send to two interfaces

[Damjan Marion via lists.fd.io]

> On 14.04.2021., at 17:15, David Gohberg  wrote:
> 
> When testing Damjan's version under traffic (using trex with a few thousand 
> pps)
> after about 5 minutes into the test vlib_buffer_clone fails to create 2 
> copies and vpp crashes due to double-free error.
> looks like the buffer pool leaks.

Are you using DPDK. What kind of NIC?

> 1. Is it the node function responsibility to vlib_buffer_free the cloned 
> buffer?

no, driver should do that….

> 2. How can I inspect the buffer pool memory usage?

show buffers

> I used the 'show memory' cli commands and didn't notice any thread memory 
> increase. 

buffer memory is pre-alllocated on startup and it not taken from the main heap 
so “show memory” will not give you any releveant data.

— 
Damjan



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19199): https://lists.fd.io/g/vpp-dev/message/19199
Mute This Topic: https://lists.fd.io/mt/81938105/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] vlib_buffer_clone behavior when trying to send to two interfaces

[David Gohberg]

When testing Damjan's version under traffic (using trex with a few thousand pps)
after about 5 minutes into the test vlib_buffer_clone fails to create 2 copies 
and vpp crashes due to double-free error.
looks like the buffer pool leaks.
1. Is it the node function responsibility to vlib_buffer_free the cloned buffer?
2. How can I inspect the buffer pool memory usage? I used the 'show memory' cli 
commands and didn't notice any thread memory increase.

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19198): https://lists.fd.io/g/vpp-dev/message/19198
Mute This Topic: https://lists.fd.io/mt/81938105/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01

[chetan bhasin]

Thanks a lot Mohammed!

Let me try with the rdma-core and patch mentioned by you and

Thanks,
Chetan

On Wed, Apr 14, 2021 at 6:20 PM Mohammed Hawari  wrote:

> Hi Chetan
>
> You are building DPDK with Mellanox support relying on dlopen and the glue
> library which is fine. The default approach implemented in the VPP build
> system consists of statically linking DPDK with the the libraries provided
> by rdma-core (as it’s also built by the VPP build system). The default
> approach should work fine, provided this patch
> https://gerrit.fd.io/r/c/vpp/+/31876 is applied (specific to CentOS). But
> what you should work just fine. Now the reason why  you need to comment
> dpdk_depends is kind of mysterious here…
>
> Regards,
>
> Mohammed
>
> On 14 Apr 2021, at 10:07, chetan bhasin 
> wrote:
>
> Hi,
>
> I have to do the following to enable Mellanox compilation under VPP21.01
> with dpdk 20.11.
>
> If I dont comment rdma-core dependencies , it will lead to undefined
> symbols .
>
> Can anybody please correct , is this the right way to do it?
>
> -DPDK_MLX4_PMD?= n
> -DPDK_MLX5_PMD?= n
> -DPDK_MLX5_COMMON_PMD ?= n
> *+DPDK_MLX4_PMD:= y*
> *+DPDK_MLX5_PMD:= y*
> *+DPDK_MLX5_COMMON_PMD = y*
>
> -DPDK_MLX_IBV_LINK?= static
> *+DPDK_MLX_IBV_LINK:= dlopen*
>
> -dpdk_depends:= rdma-core $(if $(ARCH_X86_64), ipsec-mb)
> *+#dpdk_depends   := rdma-core $(if $(ARCH_X86_64), ipsec-mb)*
>
> Thanks,
> Chetan
>
> On Tue, Apr 6, 2021 at 4:10 PM chetan bhasin via lists.fd.io
>  wrote:
>
>> Thanks Guys !
>>
>> We have a requirement to compile vppp with DPDK mellanox driver .
>>
>> Right now issue what i understood is , when we compile vpp_2101  , using
>> make build-release , it builds mellanox , but when it is trying to install
>> vpp-ext-dep rpm , it got failed because glue library created of mellanox
>> has some undefined symbols ,
>> For your reference , i have set following in dpdk.mk file
>>
>> -DPDK_MLX4_PMD?= n
>> -DPDK_MLX5_PMD?= n
>> +DPDK_MLX4_PMD:= y
>> +DPDK_MLX5_PMD:= y
>> -DPDK_MLX_IBV_LINK?= static
>> +DPDK_MLX_IBV_LINK:= shared
>>
>>  please check below logs for reference-
>>
>> FAILED: drivers/net/mlx4/librte_net_mlx4_glue.so.21.0
>> cc  -o drivers/net/mlx4/librte_net_mlx4_glue.so.21.0
>> 'drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o'
>> -Wl,--as-needed -Wl,--no-undefined -Wl,-O1 -shared -fPIC -Wl,--start-group
>> -Wl,-soname,librte_net_mlx4_glue.so.21.0 -Wl,--no-as-needed -pthread -lm
>> -ldl -lnuma -Wl,-export-dynamic -Wl,-h,librte_net_mlx4_glue.so.21.0
>> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libmlx4.so
>> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libibverbs.so
>> -Wl,--end-group
>> drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o: In
>> function `mlx4_glue_reg_mr':
>> mlx4_glue.c:(.text+0x277): undefined reference to `ibv_reg_mr_iova2'
>> collect2: error: ld returned 1 exit status
>> [8/547] Compiling C object 'drivers/a715181@@tmp_rte_net_mlx4@sta
>> /net_mlx4_mlx4_flow.c.o'
>>
>> ninja: build stopped: subcommand failed.
>>
>> On Thu, Apr 1, 2021 at 4:33 PM Юрий Иванов 
>> wrote:
>>
>>> Hi,
>>>
>>> Or you can use nativa rdma driver as written in this post and build vpp
>>> as usual. 
>>>
>>> Regards.
>>> --
>>> *От:* vpp-dev@lists.fd.io  от имени Mohammed
>>> Hawari 
>>> *Отправлено:* 1 апреля 2021 г. 13:16
>>> *Кому:* chetan bhasin 
>>> *Копия:* vpp-dev 
>>> *Тема:* Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01
>>>
>>>
>>> Hi Chetan,
>>>
>>> If you are using CentOS, I’d suggest to cherry-pick
>>> https://gerrit.fd.io/r/c/vpp/+/31876. Also please change the dpdk.mk to
>>> also set DPDK_MLX5_COMMON_PMD = y. I hope this solves your issue.
>>> Otherwise, please consider using the rdma native driver that does not rely
>>> on DPDK.
>>>
>>> Regards
>>>
>>> Mohammed
>>>
>>> On 30 Mar 2021, at 12:41, chetan bhasin 
>>> wrote:
>>>
>>> Hello Everyone,
>>>
>>> I am upgrading to vpp 2101 . I am facing a compilation issue after
>>> enabling Mellanox compilation in dpdk.mk .
>>>
>>> --- a/build/external/packages/dpdk.mk
>>> +++ b/build/external/packages/dpdk.mk
>>> @@ -14,8 +14,8 @@
>>>  DPDK_PKTMBUF_HEADROOM?= 128
>>>  DPDK_USE_LIBBSD  ?= n
>>>  DPDK_DEBUG   ?= n
>>> *-DPDK_MLX4_PMD?= n*
>>> *-DPDK_MLX5_PMD?= n*
>>> *+DPDK_MLX4_PMD?= y*
>>> *+DPDK_MLX5_PMD?= y*
>>>  DPDK_MLX5_COMMON_PMD ?= n
>>>
>>>
>>> Getting below errors . Anybody please help here ?
>>>
>>> [1344/1846] Compiling C object 'drivers/a715181@@tmp_rte_net_nfp@sta
>>> /net_nfp_nfpcore_nfp_nsp_cmds.c.o'
>>> [1345/1846] Linking target drivers/librte_net_mlx4.so.21.0
>>> FAILED: drivers/librte_net_mlx4.so.21.0
>>> cc  -o 

Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01

[Mohammed Hawari]

Hi Chetan

You are building DPDK with Mellanox support relying on dlopen and the glue 
library which is fine. The default approach implemented in the VPP build system 
consists of statically linking DPDK with the the libraries provided by 
rdma-core (as it’s also built by the VPP build system). The default approach 
should work fine, provided this patch https://gerrit.fd.io/r/c/vpp/+/31876 
 is applied (specific to CentOS). But 
what you should work just fine. Now the reason why  you need to comment 
dpdk_depends is kind of mysterious here…

Regards,

Mohammed
> On 14 Apr 2021, at 10:07, chetan bhasin  wrote:
> 
> Hi,
> 
> I have to do the following to enable Mellanox compilation under VPP21.01 with 
> dpdk 20.11.
> 
> If I dont comment rdma-core dependencies , it will lead to undefined symbols 
> . 
> 
> Can anybody please correct , is this the right way to do it?
> 
> -DPDK_MLX4_PMD?= n
> -DPDK_MLX5_PMD?= n
> -DPDK_MLX5_COMMON_PMD ?= n
> +DPDK_MLX4_PMD:= y
> +DPDK_MLX5_PMD:= y
> +DPDK_MLX5_COMMON_PMD = y
> 
> -DPDK_MLX_IBV_LINK?= static
> +DPDK_MLX_IBV_LINK:= dlopen
> 
> -dpdk_depends:= rdma-core $(if $(ARCH_X86_64), ipsec-mb)
> +#dpdk_depends   := rdma-core $(if $(ARCH_X86_64), ipsec-mb)
> 
> Thanks,
> Chetan
> 
> On Tue, Apr 6, 2021 at 4:10 PM chetan bhasin via lists.fd.io 
>   > wrote:
> Thanks Guys !
> 
> We have a requirement to compile vppp with DPDK mellanox driver . 
> 
> Right now issue what i understood is , when we compile vpp_2101  , using make 
> build-release , it builds mellanox , but when it is trying to install 
> vpp-ext-dep rpm , it got failed because glue library created of mellanox has 
> some undefined symbols ,
> For your reference , i have set following in dpdk.mk  file
> 
> -DPDK_MLX4_PMD?= n
> -DPDK_MLX5_PMD?= n
> +DPDK_MLX4_PMD:= y
> +DPDK_MLX5_PMD:= y
> -DPDK_MLX_IBV_LINK?= static
> +DPDK_MLX_IBV_LINK:= shared
> 
>  please check below logs for reference- 
> 
> FAILED: drivers/net/mlx4/librte_net_mlx4_glue.so.21.0
> cc  -o drivers/net/mlx4/librte_net_mlx4_glue.so.21.0 
> 'drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o' 
> -Wl,--as-needed -Wl,--no-undefined -Wl,-O1 -shared -fPIC -Wl,--start-group 
> -Wl,-soname,librte_net_mlx4_glue.so.21.0 -Wl,--no-as-needed -pthread -lm -ldl 
> -lnuma -Wl,-export-dynamic -Wl,-h,librte_net_mlx4_glue.so.21.0 
> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libmlx4.so 
> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libibverbs.so 
> -Wl,--end-group
> drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o: In function 
> `mlx4_glue_reg_mr':
> mlx4_glue.c:(.text+0x277): undefined reference to `ibv_reg_mr_iova2'
> collect2: error: ld returned 1 exit status
> [8/547] Compiling C object 
> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_flow.c.o'
> 
> ninja: build stopped: subcommand failed.
> 
> On Thu, Apr 1, 2021 at 4:33 PM Юрий Иванов  > wrote:
> Hi,
> 
> Or you can use nativa rdma driver as written in this post and build vpp as 
> usual. 
> 
> Regards.
> От: vpp-dev@lists.fd.io   > от имени Mohammed Hawari  >
> Отправлено: 1 апреля 2021 г. 13:16
> Кому: chetan bhasin  >
> Копия: vpp-dev mailto:vpp-dev@lists.fd.io>>
> Тема: Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01
>
> 
> Hi Chetan,
> 
> If you are using CentOS, I’d suggest to cherry-pick 
> https://gerrit.fd.io/r/c/vpp/+/31876 . 
> Also please change the dpdk.mk  to also set 
> DPDK_MLX5_COMMON_PMD = y. I hope this solves your issue. Otherwise, please 
> consider using the rdma native driver that does not rely on DPDK.
> 
> Regards
> 
> Mohammed
>> On 30 Mar 2021, at 12:41, chetan bhasin > > wrote:
>> 
>> Hello Everyone,
>> 
>> I am upgrading to vpp 2101 . I am facing a compilation issue after enabling 
>> Mellanox compilation in dpdk.mk  .
>> 
>> --- a/build/external/packages/dpdk.mk 
>> +++ b/build/external/packages/dpdk.mk 
>> @@ -14,8 +14,8 @@
>>  DPDK_PKTMBUF_HEADROOM?= 128
>>  DPDK_USE_LIBBSD  ?= n
>>  DPDK_DEBUG   ?= n
>> -DPDK_MLX4_PMD?= n
>> -DPDK_MLX5_PMD?= n
>> +DPDK_MLX4_PMD?= y
>> +DPDK_MLX5_PMD?= y
>>  DPDK_MLX5_COMMON_PMD ?= n
>> 
>> 
>> Getting below errors . Anybody please help here ?
>> 
>> [1344/1846] Compiling C object 
>> 'drivers/a715181@@tmp_rte_net_nfp@sta/net_nfp_nfpcore_nfp_nsp_cmds.c.o'
>> 

Re: [vpp-dev] TX Queue Placement

[Marcos - Mgiga]

Understood...

 

 

Usually I use show run to monitor packet drops. Do you have another suggestion 
for me to get a more precise information about it?

 

Best Regards

 

De: vpp-dev@lists.fd.io  Em nome de Damjan Marion via 
lists.fd.io
Enviada em: quarta-feira, 14 de abril de 2021 09:06
Para: Marcos - Mgiga 
Cc: vpp-dev@lists.fd.io
Assunto: Re: [vpp-dev] TX Queue Placement

 

 

General rule is that increasing those values are reducing number of rx drops 
due to 

VPP being de-scheduled or busy or due to traffic bursts, but also that degrades 
performance.

 

I think what we have set as default is reasonable and you should leave it 
unless you are experiencing some issues…

 

— 

Damjan

 





On 14.04.2021., at 13:45, Marcos - Mgiga mailto:mar...@mgiga.com.br> > wrote:

 

Hello Damjan,

 

Thank you for clarifying...

 

I also have a question about num-rx-desc and num-rt-desc parameters, hope you 
dont mind to dicuss it in this e-mail

 

I would like to understand what value fits better to my environment, do have 
any thoughts about it?

 

Best Regards

De: vpp-dev@lists.fd.io   mailto:vpp-dev@lists.fd.io> > Em nome de Damjan Marion via lists.fd.io 
 
Enviada em: quarta-feira, 14 de abril de 2021 08:24
Para: Marcos - Mgiga mailto:mar...@mgiga.com.br> >
Cc: vpp-dev@lists.fd.io  
Assunto: Re: [vpp-dev] TX Queue Placement

 

 






On 14.04.2021., at 13:21, Marcos - Mgiga mailto:mar...@mgiga.com.br> > wrote:

 

Hello,

 

I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since VPP 
is running on a NUMA system with two threads with 8 cores each, I would like to 
pin TX / RX queue to proper NUMA nodes.

 

Using set interface rx-placement I was able to associate rx queue to desired 
cores, so I would like to know if is there any possibility to pin tx queue to a 
certain workers as well.

 

 

Not at the moment. tx queues are statically mapped (0 to main thread, 1 to 
worker 0, 2 to worker 1, etc.).

There are some plans to implement such capability…..

 

 





 


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19195): https://lists.fd.io/g/vpp-dev/message/19195
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] TX Queue Placement

[Damjan Marion via lists.fd.io]

General rule is that increasing those values are reducing number of rx drops 
due to 
VPP being de-scheduled or busy or due to traffic bursts, but also that degrades 
performance.

I think what we have set as default is reasonable and you should leave it 
unless you are experiencing some issues…

— 
Damjan


> On 14.04.2021., at 13:45, Marcos - Mgiga  wrote:
> 
> Hello Damjan,
>
> Thank you for clarifying...
>
> I also have a question about num-rx-desc and num-rt-desc parameters, hope you 
> dont mind to dicuss it in this e-mail
>
> I would like to understand what value fits better to my environment, do have 
> any thoughts about it?
>
> Best Regards
> De: vpp-dev@lists.fd.io   > Em nome de Damjan Marion via lists.fd.io 
> 
> Enviada em: quarta-feira, 14 de abril de 2021 08:24
> Para: Marcos - Mgiga mailto:mar...@mgiga.com.br>>
> Cc: vpp-dev@lists.fd.io 
> Assunto: Re: [vpp-dev] TX Queue Placement
>
>
> 
> 
> On 14.04.2021., at 13:21, Marcos - Mgiga  > wrote:
>
> Hello,
>
> I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since 
> VPP is running on a NUMA system with two threads with 8 cores each, I would 
> like to pin TX / RX queue to proper NUMA nodes.
>
> Using set interface rx-placement I was able to associate rx queue to desired 
> cores, so I would like to know if is there any possibility to pin tx queue to 
> a certain workers as well.
>
>
> Not at the moment. tx queues are statically mapped (0 to main thread, 1 to 
> worker 0, 2 to worker 1, etc.).
> There are some plans to implement such capability…..
>
>
> 
> 


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19194): https://lists.fd.io/g/vpp-dev/message/19194
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] TX Queue Placement

[Marcos - Mgiga]

Hello Damjan,

 

Thank you for clarifying...

 

I also have a question about num-rx-desc and num-rt-desc parameters, hope you 
dont mind to dicuss it in this e-mail

 

I would like to understand what value fits better to my environment, do have 
any thoughts about it?

 

Best Regards

De: vpp-dev@lists.fd.io  Em nome de Damjan Marion via 
lists.fd.io
Enviada em: quarta-feira, 14 de abril de 2021 08:24
Para: Marcos - Mgiga 
Cc: vpp-dev@lists.fd.io
Assunto: Re: [vpp-dev] TX Queue Placement

 

 





On 14.04.2021., at 13:21, Marcos - Mgiga mailto:mar...@mgiga.com.br> > wrote:

 

Hello,

 

I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since VPP 
is running on a NUMA system with two threads with 8 cores each, I would like to 
pin TX / RX queue to proper NUMA nodes.

 

Using set interface rx-placement I was able to associate rx queue to desired 
cores, so I would like to know if is there any possibility to pin tx queue to a 
certain workers as well.

 

 

Not at the moment. tx queues are statically mapped (0 to main thread, 1 to 
worker 0, 2 to worker 1, etc.).

There are some plans to implement such capability…..

 

 


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19193): https://lists.fd.io/g/vpp-dev/message/19193
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] TX Queue Placement

[Damjan Marion via lists.fd.io]

> On 14.04.2021., at 13:21, Marcos - Mgiga  wrote:
> 
> Hello,
>  
> I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since 
> VPP is running on a NUMA system with two threads with 8 cores each, I would 
> like to pin TX / RX queue to proper NUMA nodes.
>  
> Using set interface rx-placement I was able to associate rx queue to desired 
> cores, so I would like to know if is there any possibility to pin tx queue to 
> a certain workers as well.
>  

Not at the moment. tx queues are statically mapped (0 to main thread, 1 to 
worker 0, 2 to worker 1, etc.).
There are some plans to implement such capability…..



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19192): https://lists.fd.io/g/vpp-dev/message/19192
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

[vpp-dev] TX Queue Placement

[Marcos - Mgiga]

Hello,

 

I increased VPP rx/tx queues in order to enable RSS on VPP instance. Since
VPP is running on a NUMA system with two threads with 8 cores each, I would
like to pin TX / RX queue to proper NUMA nodes.

 

Using set interface rx-placement I was able to associate rx queue to desired
cores, so I would like to know if is there any possibility to pin tx queue
to a certain workers as well.

 

 

Best regards

 

Marcos


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19191): https://lists.fd.io/g/vpp-dev/message/19191
Mute This Topic: https://lists.fd.io/mt/82088483/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

[vpp-dev] [vpp-api-dev] vppapigen: Backwards compatible API dictionary removal

[Ole Troan]

All,

As discussed during yesterdays community meeting we plan to merge this patch a 
week from now. April 21st.
Object now, or forever hold your peace. ;-)

Best regards,
Ole

API users,

This summer we found an error in the crc generation and to avoid forcing API 
users to update their manifests on short notice I put in 9f84e70c6d.
Which "cheats" and returns the old crc value instead of the correct crc value 
for a set of messages.

https://git.fd.io/vpp/tree/src/tools/vppapigen/vppapigen.py#n1143


In time for the next release I'd like to remove the "crc compatibility table" 
from the API compiler.

Paul has already a patch for that:
https://gerrit.fd.io/r/c/vpp/+/30345

The APIs themselves will not change, only the CRCs. This would affect anyone 
maintaining manifests.
I know CSIT is affected, but are there others too that will be affected by such 
a change?

I'll leave this issue open for 2 weeks or so, and unless there are objections 
I'd like to have 30345 merged before next release.

Best regards,
Ole








signature.asc
Description: Message signed with OpenPGP

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19190): https://lists.fd.io/g/vpp-dev/message/19190
Mute This Topic: https://lists.fd.io/mt/82088104/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01

[Damjan Marion via lists.fd.io]

> On 14.04.2021., at 10:07, chetan bhasin  wrote:
> 
> Hi,
> 
> I have to do the following to enable Mellanox compilation under VPP21.01 with 
> dpdk 20.11.
> 
> If I dont comment rdma-core dependencies , it will lead to undefined symbols 
> . 
> 
> Can anybody please correct , is this the right way to do it?

Right way to do it is to use rdma plugin instead of dpdk...

— 
Damjan



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19189): https://lists.fd.io/g/vpp-dev/message/19189
Mute This Topic: https://lists.fd.io/mt/81718770/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01

[chetan bhasin]

Hi,

I have to do the following to enable Mellanox compilation under VPP21.01
with dpdk 20.11.

If I dont comment rdma-core dependencies , it will lead to undefined
symbols .

Can anybody please correct , is this the right way to do it?

-DPDK_MLX4_PMD?= n
-DPDK_MLX5_PMD?= n
-DPDK_MLX5_COMMON_PMD ?= n
*+DPDK_MLX4_PMD:= y*
*+DPDK_MLX5_PMD:= y*
*+DPDK_MLX5_COMMON_PMD = y*

-DPDK_MLX_IBV_LINK?= static
*+DPDK_MLX_IBV_LINK:= dlopen*

-dpdk_depends:= rdma-core $(if $(ARCH_X86_64), ipsec-mb)
*+#dpdk_depends   := rdma-core $(if $(ARCH_X86_64), ipsec-mb)*

Thanks,
Chetan

On Tue, Apr 6, 2021 at 4:10 PM chetan bhasin via lists.fd.io
 wrote:

> Thanks Guys !
>
> We have a requirement to compile vppp with DPDK mellanox driver .
>
> Right now issue what i understood is , when we compile vpp_2101  , using
> make build-release , it builds mellanox , but when it is trying to install
> vpp-ext-dep rpm , it got failed because glue library created of mellanox
> has some undefined symbols ,
> For your reference , i have set following in dpdk.mk file
>
> -DPDK_MLX4_PMD?= n
> -DPDK_MLX5_PMD?= n
> +DPDK_MLX4_PMD:= y
> +DPDK_MLX5_PMD:= y
> -DPDK_MLX_IBV_LINK?= static
> +DPDK_MLX_IBV_LINK:= shared
>
>  please check below logs for reference-
>
> FAILED: drivers/net/mlx4/librte_net_mlx4_glue.so.21.0
> cc  -o drivers/net/mlx4/librte_net_mlx4_glue.so.21.0
> 'drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o'
> -Wl,--as-needed -Wl,--no-undefined -Wl,-O1 -shared -fPIC -Wl,--start-group
> -Wl,-soname,librte_net_mlx4_glue.so.21.0 -Wl,--no-as-needed -pthread -lm
> -ldl -lnuma -Wl,-export-dynamic -Wl,-h,librte_net_mlx4_glue.so.21.0
> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libmlx4.so
> /usr/lib/gcc/x86_64-redhat-linux/4.8.5/../../../../lib64/libibverbs.so
> -Wl,--end-group
> drivers/net/mlx4/8672f8e@@rte_net_mlx4_glue@sha/mlx4_glue.c.o: In
> function `mlx4_glue_reg_mr':
> mlx4_glue.c:(.text+0x277): undefined reference to `ibv_reg_mr_iova2'
> collect2: error: ld returned 1 exit status
> [8/547] Compiling C object 'drivers/a715181@@tmp_rte_net_mlx4@sta
> /net_mlx4_mlx4_flow.c.o'
>
> ninja: build stopped: subcommand failed.
>
> On Thu, Apr 1, 2021 at 4:33 PM Юрий Иванов  wrote:
>
>> Hi,
>>
>> Or you can use nativa rdma driver as written in this post and build vpp
>> as usual. 
>>
>> Regards.
>> --
>> *От:* vpp-dev@lists.fd.io  от имени Mohammed Hawari
>> 
>> *Отправлено:* 1 апреля 2021 г. 13:16
>> *Кому:* chetan bhasin 
>> *Копия:* vpp-dev 
>> *Тема:* Re: [vpp-dev] How to enable Mellanox compilation in VPP 21.01
>>
>>
>> Hi Chetan,
>>
>> If you are using CentOS, I’d suggest to cherry-pick
>> https://gerrit.fd.io/r/c/vpp/+/31876. Also please change the dpdk.mk to
>> also set DPDK_MLX5_COMMON_PMD = y. I hope this solves your issue.
>> Otherwise, please consider using the rdma native driver that does not rely
>> on DPDK.
>>
>> Regards
>>
>> Mohammed
>>
>> On 30 Mar 2021, at 12:41, chetan bhasin 
>> wrote:
>>
>> Hello Everyone,
>>
>> I am upgrading to vpp 2101 . I am facing a compilation issue after
>> enabling Mellanox compilation in dpdk.mk .
>>
>> --- a/build/external/packages/dpdk.mk
>> +++ b/build/external/packages/dpdk.mk
>> @@ -14,8 +14,8 @@
>>  DPDK_PKTMBUF_HEADROOM?= 128
>>  DPDK_USE_LIBBSD  ?= n
>>  DPDK_DEBUG   ?= n
>> *-DPDK_MLX4_PMD?= n*
>> *-DPDK_MLX5_PMD?= n*
>> *+DPDK_MLX4_PMD?= y*
>> *+DPDK_MLX5_PMD?= y*
>>  DPDK_MLX5_COMMON_PMD ?= n
>>
>>
>> Getting below errors . Anybody please help here ?
>>
>> [1344/1846] Compiling C object 'drivers/a715181@@tmp_rte_net_nfp@sta
>> /net_nfp_nfpcore_nfp_nsp_cmds.c.o'
>> [1345/1846] Linking target drivers/librte_net_mlx4.so.21.0
>> FAILED: drivers/librte_net_mlx4.so.21.0
>> cc  -o drivers/librte_net_mlx4.so.21.0 
>> 'drivers/a715181@@rte_net_mlx4@sha/meson-generated_.._rte_net_mlx4.pmd.c.o'
>> 'drivers/a715181@@tmp_r
>> te_net_mlx4@sta/net_mlx4_mlx4.c.o' 
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_ethdev.c.o'
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta
>> /net_mlx4_mlx4_flow.c.o' 
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_intr.c.o'
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_ml
>> x4_mp.c.o' 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_mr.c.o'
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_rxq.c.o' 'dri
>> vers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_rxtx.c.o'
>> 'drivers/a715181@@tmp_rte_net_mlx4@sta/net_mlx4_mlx4_txq.c.o'
>> 'drivers/a715181@
>> @tmp_rte_net_mlx4@sta/net_mlx4_mlx4_utils.c.o' 'drivers/a715181@
>> @tmp_rte_net_mlx4@sta/net_mlx4_mlx4_glue.c.o' -Wl,--as-needed
>> -Wl,--no-und
>> efined -Wl,-O1 -shared -fPIC -Wl,--start-group
>> 

Re: [vpp-dev] VPP crashes on vhost-user VM reboot

[Eyle Brinkhuis]

All,

Moving VPP to 21.01 seems to solve this issue for now.

Eyle

> On 9 Apr 2021, at 09:59, Eyle Brinkhuis  wrote:
> 
> Hi guys,
> 
> We are experiencing crashes left and right on VPP 20.05.1 that we installed 
> from packagecloud. VPP crashes when a VM with 3 interfaces (all three 
> connected to VPP on vhost-user) reboots or gets spawned. I was able to 
> capture a core dump, which is available here:
> https://surfdrive.surf.nl/files/index.php/s/qIR01agXQD4Jkha 
> 
> 
> Anyone want to take a look at this? I am planning on moving to 20.09 today, 
> and will see if that solves our problem but maybe it is something worth 
> looking at. I will report back with my experience on 20.09.
> 
> Regards,
> 
> Eyle
> 
> 
> 



signature.asc
Description: Message signed with OpenPGP

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19187): https://lists.fd.io/g/vpp-dev/message/19187
Mute This Topic: https://lists.fd.io/mt/81963208/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-